AI Advancements and Challenges: From OpenAI's Crisis to Wall Street's Adoption

Silicon Valley's 'Audacity Crisis'

Two years ago, OpenAI released the public beta of DALL-E 2, an image-generation tool that immediately signified that we'd entered a new technological era. Trained off a huge body of data, DALL-E 2 produced unsettlingly good, delightful, and frequently unexpected outputs; my Twitter feed filled up with images derived from prompts such as close-up photo of brushing teeth with toothbrush covered with nacho cheese. Suddenly, it seemed as though machines could create just about anything in response to simple prompts. You likely know the story from there: A few months later, ChatGPT arrived, millions of people started using it, the student essay was pronounced dead, Web3 entrepreneurs nearly broke their ankles scrambling to pivot their companies to AI, and the technology industry was consumed by hype. The generative-AI revolution began in earnest. Where has it gotten us? Although enthusiasts eagerly use the technology to boost productivity and automate busywork, the drawbacks are also impossible to ignore. Social networks such as Facebook have been flooded with bizarre AI-generated slop images; search engines are floundering, trying to index an internet awash in hastily assembled, chatbot-written articles. Generative AI, we know for sure now, has been trained without permission on copyrighted media, which makes it all the more galling that the technology is competing against creative people for jobs and online attention; a backlash against AI companies scraping the internet for training data is in full swing. Yet these companies, emboldened by the success of their products and war chests of investor capital, have brushed these problems aside and unapologetically embraced a manifest-destiny attitude toward their technologies. Some of these firms are, in no uncertain terms, trying to rewrite the rules of society by doing whatever they can to create a godlike superintelligence (also known as artificial general intelligence, or AGI). Others seem more interested in using generative AI to build tools that repurpose others' creative work with little to no citation. In recent months, leaders within the AI industry are more brazenly expressing a paternalistic attitude about how the future will look -- including who will win (those who embrace their technology) and who will be left behind (those who do not). They're not asking us; they're telling us. As the journalist Joss Fong commented recently, "There's an audacity crisis happening in California." There are material concerns to contend with here. It is audacious to massively jeopardize your net-zero climate commitment in favor of advancing a technology that has told people to eat rocks, yet Google appears to have done just that, according to its latest environmental report. (In an emailed statement, a Google spokesperson, Corina Standiford, said that the company remains "dedicated to the sustainability goals we've set," including reaching net-zero emissions by 2030. According to the report, its emissions grew 13 percent in 2023, in large part because of the energy demands of generative AI.) And it is certainly audacious for companies such as Perplexity to use third-party tools to harvest information while ignoring long-standing online protocols that prevent websites from being scraped and having their content stolen. But I've found the rhetoric from AI leaders to be especially exasperating. This month, I spoke with OpenAI CEO Sam Altman and Thrive Global CEO Arianna Huffington after they announced their intention to build an AI health coach. The pair explicitly compared their nonexistent product to the New Deal. (They suggested that their product -- so theoretical, they could not tell me whether it would be an app or not -- could quickly become part of the health-care system's critical infrastructure.) But this audacity is about more than just grandiose press releases. In an interview at Dartmouth College last month, OpenAI's chief technology officer, Mira Murati, discussed AI's effects on labor, saying that, as a result of generative AI, "some creative jobs maybe will go away, but maybe they shouldn't have been there in the first place." She added later that "strictly repetitive" jobs are also likely on the chopping block. Her candor appears emblematic of OpenAI's very mission, which straightforwardly seeks to develop an intelligence capable of "turbocharging the global economy." Jobs that can be replaced, her words suggested, aren't just unworthy: They should never have existed. In the long arc of technological change, this may be true -- human operators of elevators, traffic signals, and telephones eventually gave way to automation -- but that doesn't mean that catastrophic job loss across several industries simultaneously is economically or morally acceptable. Read: AI has become a technology of faith Along these lines, Altman has said that generative AI will "create entirely new jobs." Other tech boosters have said the same. But if you listen closely, their language is cold and unsettling, offering insight into the kinds of labor that these people value -- and, by extension, the kinds that they don't. Altman has spoken of AGI possibly replacing the "the median human" worker's labor -- giving the impression that the least exceptional among us might be sacrificed in the name of progress. Even some inside the industry have expressed alarm at those in charge of this technology's future. Last month, Leopold Aschenbrenner, a former OpenAI employee, wrote a 165-page essay series warning readers about what's being built in San Francisco. "Few have the faintest glimmer of what is about to hit them," Aschenbrenner, who was reportedly fired this year for leaking company information, wrote. In Aschenbrenner's reckoning, he and "perhaps a few hundred people, most of them in San Francisco and the AI labs," have the "situational awareness" to anticipate the future, which will be marked by the arrival of AGI, geopolitical struggle, and radical cultural and economic change. Aschenbrenner's manifesto is a useful document in that it articulates how the architects of this technology see themselves: a small group of people bound together by their intellect, skill sets, and fate to help decide the shape of the future. Yet to read his treatise is to feel not FOMO, but alienation. The civilizational struggle he depicts bears little resemblance to the AI that the rest of us can see. "The fate of the world rests on these people," he writes of the Silicon Valley cohort building AI systems. This is not a call to action or a proposal for input; it's a statement of who is in charge. Unlike me, Aschenbrenner believes that a superintelligence is coming, and coming soon. His treatise contains quite a bit of grand speculation about the potential for AI models to drastically improve from here. (Skeptics have strongly pushed back on this assessment.) But his primary concern is that too few people wield too much power. "I don't think it can just be a small clique building this technology," he told me recently when I asked why he wrote the treatise. "I felt a sense of responsibility, by having ended up a part of this group, to tell people what they're thinking," he said, referring to the leaders at AI companies who believe they're on the cusp of achieving AGI. "And again, they might be right or they might be wrong, but people deserve to hear it." In our conversation, I found an unexpected overlap between us: Whether you believe that AI executives are delusional or genuinely on the verge of constructing a superintelligence, you should be concerned about how much power they've amassed. Having a class of builders with deep ambitions is part of a healthy, progressive society. Great technologists are, by nature, imbued with an audacious spirit to push the bounds of what is possible -- and that can be a very good thing for humanity indeed. None of this is to say that the technology is useless: AI undoubtedly has transformative potential (predicting how proteins fold is a genuine revelation, for example). But audacity can quickly turn into a liability when builders become untethered from reality, or when their hubris leads them to believe that it is their right to impose their values on the rest of us, in return for building God. Read: This is what it looks like when AI eats the world An industry is what it produces, and in 2024, these executive pronouncements and brazen actions, taken together, are the actual state of the artificial-intelligence industry two years into its latest revolution. The apocalyptic visions, the looming nature of superintelligence, and the struggle for the future of humanity -- all of these narratives are not facts but hypotheticals, however exciting, scary, or plausible. When you strip all of that away and focus on what's really there and what's really being said, the message is clear: These companies wish to be left alone to "scale in peace," a phrase that SSI, a new AI company co-founded by Ilya Sutskever, formerly OpenAI's chief scientist, used with no trace of self-awareness in announcing his company's mission. ("SSI" stands for "safe superintelligence," of course.) To do that, they'll need to commandeer all creative resources -- to eminent-domain the entire internet. The stakes demand it. We're to trust that they will build these tools safely, implement them responsibly, and share the wealth of their creations. We're to trust their values -- about the labor that's valuable and the creative pursuits that ought to exist -- as they remake the world in their image. We're to trust them because they are smart. We're to trust them as they achieve global scale with a technology that they say will be among the most disruptive in all of human history. Because they have seen the future, and because history has delivered them to this societal hinge point, marrying ambition and talent with just enough raw computing power to create God. To deny them this right is reckless, but also futile. It's possible, then, that generative AI's chief export is not image slop, voice clones, or lorem ipsum chatbot bullshit but instead unearned, entitled audacity. Yet another example of AI producing hallucinations -- not in the machines, but in the people who build them.

This Week in AI: How Kamala Harris might regulate AI | TechCrunch

Hiya, folks, welcome to TechCrunch's regular AI newsletter. Last Sunday, President Joe Biden announced that he no longer plans to seek reelection, instead offering his "full endorsement" of VP Kamala Harris to become the Democratic Party's nominee; in the days following, Harris secured support from the Democratic delegate majority. Harris has been outspoken on tech and AI policy; should she win the presidency, what would that mean for U.S. AI regulation? My colleague Anthony Ha penned a few words on this over the weekend. Harris and President Biden previously said they "reject the false choice that suggests we can either protect the public or advance innovation." At that time, Biden had issued an executive order calling for companies to set new standards around the development of AI. Harris said that the voluntary commitments were "an initial step toward a safer AI future with more to come" because "in the absence of regulation and strong government oversight, some technology companies choose to prioritize profit over the well-being of their customers, the safety of our communities, and the stability of our democracies." I also spoke with AI policy experts to get their views. For the most part, they said that they'd expect consistency with a Harris administration, as opposed to a dismantling of the current AI policy and general deregulation that Donald Trump's camp has championed. Lee Tiedrich, an AI consultant at the Global Partnership on Artificial Intelligence, told TechCrunch that Biden's endorsement of Harris could "increase the chances of maintaining continuity" in U.S. AI policy. "[This is] framed by the 2023 AI executive order and also marked by multilateralism through the United Nations, the G7, the OECD and other organizations," she said. "The executive order and related actions also call for more government oversight of AI, including through increased enforcement, greater agency AI rules and policies, a focus on safety and certain mandatory testing and disclosures for some large AI systems." Sarah Kreps, a professor of government at Cornell with a special interest in AI, noted that there's a perception within certain segments of the tech industry that the Biden administration leaned too aggressively into regulation and that the AI executive order was "micromanagement overkill." She doesn't anticipate that Harris would roll back any of the AI safety protocols instituted under Biden, but she does wonder whether a Harris administration might take a less top-down regulatory approach to placate critics. Krystal Kauffman, a research fellow at the Distributed AI Research Institute, agrees with Kreps and Tiedrich that Harris will most likely continue Biden's work to address the risks associated with AI use and seek to increase transparency around AI. However, she hopes that, should Harris clinch the presidential election, she'll cast a wider stakeholder net in formulating policy -- a net that captures the data workers whose plight (poor pay, poor working conditions and mental health challenges) often goes unacknowledged. "Harris must include the voices of data workers who help program AI in these important conversations going forward," Kauffman said. "We cannot continue to see closed-door meetings with tech CEOs as a means to work out policy. This will absolutely take us down the wrong path if it continues." Meta releases new models: Meta this week released Llama 3.1 405B, a text-generating and -analyzing model containing 405 billion parameters. Its largest "open" model yet, Llama 3.1 405B is making its way into various Meta platforms and apps, including the Meta AI experience across Facebook, Instagram and Messenger. Adobe refreshes Firefly: Adobe released new Firefly tools for Photoshop and Illustrator on Tuesday, offering graphic designers more ways to use the company's in-house AI models. Facial recognition at school: An English school has been formally reprimanded by the U.K.'s data protection regulator after it used facial-recognition technology without getting specific opt-in consent from students for processing their facial scans. Cohere raises half a billion: Cohere, a generative AI startup co-founded by ex-Google researchers, has raised $500 million in new cash from investors, including Cisco and AMD. Unlike many of its generative AI startup rivals, Cohere customizes AI models for big enterprises -- a key factor in its success. CIA AI director interview: As part of TechCrunch's ongoing Women in AI series, yours truly interviewed Lakshmi Raman, the director of AI at the CIA. We talked about her path to director as well as the CIA's use of AI, and the balance that needs to be struck between embracing new tech and deploying it responsibly. Ever heard of the transformer? It's the AI model architecture of choice for complex reasoning tasks, powering models like OpenAI's GPT-4o, Anthropic's Claude and many others. But, as powerful as transformers are, they have their flaws. And so researchers are investigating possible alternatives. One of the more promising candidates is state space models (SSM), which combine the qualities of several older types of AI models, such as recurrent neural networks and convolutional neural networks, to create a more computationally efficient architecture capable of ingesting long sequences of data (think novels and movies). And one of the strongest incarnations of SSMs yet, Mamba-2, was detailed in a paper this month by research scientists Tri Dao (a professor at Princeton) and Albert Gu (Carnegie Mellon). Like its predecessor Mamba, Mamba-2 can handle larger chunks of input data than transformer-based equivalents while remaining competitive, performance-wise, with transformer-based models on certain language-generation tasks. Dao and Gu imply that, should SSMs continue to improve, they'll someday run on commodity hardware -- and deliver more powerful generative AI applications than are possible with today's transformers. In another recent architecture-related development, a team of researchers developed a new type of generative AI model they claim can match -- or beat -- both the strongest transformers and Mamba in terms of efficiency. Called test-time training models (TTT models), the architecture can reason over millions of tokens, according to the researchers, potentially scaling up to billions of tokens in future, refined designs. (In generative AI, "tokens" are bits of raw text and other bite-sized data pieces.) Because TTT models can take in many more tokens than conventional models and do so without overly straining hardware resources, they're fit to power "next-gen" generative AI apps, the researchers believe. For a deeper dive into TTT models, check out our recent feature. Stability AI, the generative AI startup that investors, including Napster co-founder Sean Parker, recently swooped in to save from financial ruin, has caused quite a bit of controversy over its restrictive new product terms of use and licensing policies. Until recently, to use Stability AI's newest open AI image model, Stable Diffusion 3, commercially, organizations making less than $1 million a year in revenue had to sign up for a "creator" license that capped the total number of images they could generate to 6,000 per month. The bigger issue for many customers, though, was Stability's restrictive fine-tuning terms, which gave (or at least appeared to give) Stability AI the right to extract fees for and exert control over any model trained on images generated by Stable Diffusion 3. Stability AI's heavy-handed approach led CivitAI, one of the largest hosts of image-generating models, to impose a temporary ban on models based or trained on images from Stable Diffusion 3 while it sought legal counsel on the new license. "The concern is that from our current understanding, this license grants Stability AI too much power over the use of not only any models fine-tuned on Stable Diffusion 3, but on any other models that include Stable Diffusion 3 images in their data sets," CivitAI wrote in a post on its blog. In response to the blowback, Stability AI early this month said that it'll adjust the licensing terms for Stable Diffusion 3 to allow for more liberal commercial use. "As long as you don't use it for activities that are illegal, or clearly violate our license or acceptable use policy, Stability AI will never ask you to delete resulting images, fine-tunes or other derived products -- even if you never pay Stability AI," Stability clarified in a blog. The saga highlights the legal pitfalls that continue to plague generative AI -- and, relatedly, the extent to which "open" remains subject to interpretation. Call me a pessimist, but the growing number of controversially restrictive licenses suggests to me that the AI industry won't reach consensus -- or inch toward clarity -- anytime soon.

Is Meta's new Llama AI model a game changer?

People are buzzing about today's release of Meta's new Llama 3.1 model. What's notable is that this is Meta's largest Llama model to date, with 405 billion parameters. (Parameters are the adjustable variables in a neural network, and give a rough sense of how large an AI model is.) And according to benchmark performance figures that conveniently leaked onto Reddit the day ahead of the official release, Llama 3.1 exceeds the capabilities of OpenAI's latest and greatest model, GPT-4o, by a few percentage points across a number of measures, including some benchmarks designed to test reasoning. Not only that, but Llama 3.1 is, like the other Llama models Meta has released, an "open model," meaning anyone can potentially build their own applications on top of it without paying, and even modify the model in any way they desire. But the models Meta has released before have been smaller, and less capable than any of the proprietary models, such as OpenAI's GPT-4, Anthropic's Claude 3 Opus, or Gemini's Ultra or 1.5 Pro models. The fact that Meta's new Llama 3.1 may have now closed the gap to GPT-4o has a lot of people excited that this Llama 3.1 405B will be the model that finally enables many businesses to really unlock the return on investment from generative AI. Anton McGonnell, head of software products at SambaNova Systems, which builds AI hardware and software for big companies, said in a statement that Llama 3.1 405B might be a game changer because it will allow two things: one is that companies can use the 405B parameter model to create synthetic datasets that can be used to train or fine-tune small open models to hone them for specific applications. This "distillation" process has been possible before but there were often ethical concerns about how the data used for "distillation" had been sourced (with data being scraped from the web without consent, or derived from the use of poorly paid human contractors). McGonnell also applauded Meta's decision to release Llama 3.1 405B as part of a family of Llama models of different sizes (there are also upgraded 70 billion and 8 billion parameter sizes) and to release a suggested "Llama stack." This is a set of related software built on top of and around the AI models themselves. Meta's AI stack includes guardrails software, to prevent the AI models from generating harmful or dangerous content, and security software to try to prevent prompt injection attacks against the Llama models. The family of models and the AI stack, McGonnell said, create the possibility of chaining open models together in a way that would be especially cost-effective -- using a process in which parts of a user's query or an application are handled by small, fine-tuned models, and only those more difficult aspects that these models can't handle are handed off to the full-scale 405 billion parameter model. But McGonnell's enthusiasm aside, there's a catch -- actually a bunch of them. The model is so big that it can't easily be hosted on a single GPU or even a dozen of them. (Meta's 70 billion parameter version of Llama 3 can potentially be run on two high-end Nvidia GPUs.) That means companies might have to pay for a lot of their own very expensive GPUs in the cloud to run the model and they will need a lot of rare technical expertise in how to split an AI workload across those GPUs and then bring the results back together to produce an output. To overcome those two issues, Meta is partnering with a bunch of companies, such as the AI services and data analytics company Databricks and the cloud service providers AWS and Google Cloud, to host the model and offer tools and services around it. It has also partnered with Groq, a hardware company that builds an alternative computer chip to Nvidia's GPUs that is designed specifically for running AI workloads on trained models, to help try to lower the cost of running such a large model and also speed up the time it takes the model to generate an output. Such an arrangement starts to make access to Llama 3.1 405B look a lot more like accessing a proprietary model through an application programming interface (API), which is what OpenAI, Anthropic, and Google Gemini offer (Google also offers some open models, called Gemma). It's not clear yet how the costs of hosting and accessing your own Llama 3.1 model through one of Meta's partners will compare to simply building on top of OpenAI's GPT-4o or Claude Opus. Previously, some developers have reportedly complained that hosting their own version of Llama 3's 70 billion parameter model was sometimes more expensive than simply paying OpenAI on a per-token basis to access the more capable GPT-4 model. It also isn't clear yet how much developers will be able to tinker with the parameters of the Llama 3.1 model they are running on the servers of one of Meta's partners, which presumably may be using the same model to run inference for several customers in order to maximize the return on their own hardware investment to host such a big model. If these partners limit how much developers can adjust the model's weights, that may negate some of the advantages of using the open model. It also isn't clear yet exactly what commercial licensing restrictions Meta has placed on the use of Llama 3.1 405B. In the past, the restrictions Meta has placed around the licensing of its Llama models have led open-source software purists to complain that Meta has twisted the meaning of open-source beyond recognition and that these models should not be called "open-source software" at all. Hence the growing use of the term "open model" as opposed to "open-source model." As with all open models, there are also some real concerns about AI safety here. Llama has not revealed the results of any red-teaming or safety testing it has done of its own model. More capable models are generally more dangerous -- a bad actor could more easily use them to suggest recipes for bioweapons or chemical weapons, to develop malicious software code, or to run highly automated disinformation campaigns, phishing schemes, or frauds. And as with all open models, it is easy for a sophisticated AI developer to remove any guardrails Meta has engineered into the baseline model. Finally, as capable as Llama 3.1 405B may be, it will likely be superseded soon by even more capable proprietary models. Google is working on Project Astra, an AI model that will be more "agentic" -- able to take actions, not just generate text or images. At Fortune's Brainstorm Tech conference last week, Google's chief research scientist Jeff Dean told me that Google will likely begin rolling this model out to some test users as soon as the fall. OpenAI is known to be training GPT-5, which will certainly be more capable than GPT-4o and may also have agentic properties. Anthropic is no doubt training a model that goes beyond Claude 3 Opus, its most powerful model, and also working on an AI agent. All of this just underscores how competitive the AI "foundation model" -- models on which many different kinds of AI applications can be built -- has become and how difficult it will be for any AI startups working on such models to survive as independent entities. That may not bode well for investors in hot French AI startups Mistral and H, or other independent foundation model companies like Cohere, or even somewhat more specialized AI model companies such as Character AI and Essential AI. It may be that only the biggest tech players, or those closely associated with them, will be able to keep pushing the boundaries of what these models can do. The good news for the rest of us is that, despite the caveats I've listed above, this foundation model race is actually driving down the cost of implementing AI models. While overall AI spending is continuing to climb as companies begin to deploy AI models more widely across their organizations, on a per-output basis, "the cost of intelligence" is falling dramatically. This should mean more companies will begin to see a return on investment from generative AI, accelerating the dawn of this new AI era. Before we get to the news... If you want to learn more about AI and its likely impacts on our companies, our jobs, our society, and even our own personal lives, please consider picking up a copy of my new book, Mastering AI: A Survival Guide to Our Superpowered Future. It's out now in the U.S. from Simon & Schuster and you can order a copy today here. If you live in the U.K., the book will be published by Bedford Square Publishers next week and you can preorder a copy today here. Senate Democrats demand AI safety information from OpenAI. Senate Democrats have written OpenAI to demand data on its AI safety efforts following employee warnings about rushed safety testing, according to a story in the Washington Post. Led by Sen. Brian Schatz of Hawaii, the lawmakers asked CEO Sam Altman to outline plans to prevent AI misuse, such as creating bioweapons or aiding cyberattacks, and to disclose information on employee agreements that could stifle whistleblowing. OpenAI has said it has removed non-disparagement terms from staff agreements that might make it difficult for employees to become whistleblowers. The Senate's letter also requests that OpenAI allow independent experts to assess its safety systems and provide AI models to the government for pre-deployment testing. The senators have asked OpenAI to respond to their letter by Aug. 13. Google appears to limit AI-generated overviews of search results. That's according to The Verge, which cited data collected by SEO company BrightEdge. The prevalence of Google's AI-generated search results dropped from 11% of queries on June 1 to 7% by June 30, BrightEdge found. This reduction follows adjustments made by Google to address bizarre results, such as suggesting users put glue on pizza or eat rocks. Google disputes the study's methodology, noting it only tracks users who opted into the experimental AI features. Google has said it remains committed to refining AI Overviews to enhance their usefulness and maintain user trust. Condé Nast asks Perplexity to stop scraping its content. According to a report in The Information, lawyers for the magazine publishing house have sent a cease and desist letter to the buzzy AI generative search company Perplexity asking it to stop scraping data from its magazines' web pages. Previously, Wired, a Condé Nast publication, had reported that Perplexity was continuing to cull data from web pages that had asked web crawling bots not to scrape their data using a protocol called "robots.txt." Perplexity had publicly stated it would abide by the robots.txt protocol and not scrape data from such pages, but in experiments, the magazine had caught a web crawler scraping newly set up pages immediately after reporters for the publication sent queries to Perplexity's search engine that included exact passages of text found on those new web pages. Perplexity has previously landed in hot water with Forbes for using information from its web pages without what Forbes considered adequate attribution. Cohere valued at $5.5 billion in new funding round. The Canadian foundation model company, which is targeting business customers, has been valued at $5.5 billion following a new $500 million investment round led by pension fund PSP Investments. You can read more in this Bloomberg story. Legal AI company Harvey valued at $1.5 billion in latest funding round. The startup published a blog post announcing a new $100 million Series C funding round led by GV (formerly Google Ventures), with participation from OpenAI, Kleiner Perkins, Sequoia, and other notable Silicon Valley investors. The company said the round valued it at $1.5 billion and that it would use the money to continue to scale. Harvey is an OpenAI partner and has built its legal copilot on top of OpenAI's GPT models. A new way to solve the reliability problems of today's LLMs? One of the biggest problems with today's LLM-based AI models is that they can be maddeningly unreliable. One minute they generate a wonderful and accurate answer to a complex physics problem. The next they can't answer a much simpler high school mathematics problem or even win a game of tick-tack-toe. They hallucinate, making up information. And it can be very difficult if not impossible to figure out exactly why the models have gone wrong when they do go wrong. Some AI experts, perhaps most notably Gary Marcus (who has long been a critic of pure deep learning approaches to AI), have suggested for a while now that hybrid systems, that include some elements of a neural network with some elements of deep learning systems or LLMs, could be the key to overcoming the big drawbacks of today's frontier AI systems. Now a group of researchers from the KU Leuven in Belgium, the University of Cambridge in England, and a number of Italian universities has proposed such a hybrid that they say achieves exactly this. They have developed an AI model they call a Concept-based Memory Reasoner (CMR). It works by using a neural network to break down a task into small conceptual chunks and then storing these chunks in memory. When confronted with a new task, it then must select from these chunks and combine them using symbolic rules to achieve an output. This allows human experts to inspect and verify the logic behind each decision, ensuring the AI is both accurate and transparent. In essence, it works by constraining what the neural network can do. It has to select from the set concepts it has seen from prior tasks and combine them according to a clear set of rules. This makes the output more reliable and easy for humans to interpret. The drawback may be that this means the CMR may not be able to deal with all the new situations we would want the model to deal with. But it is an interesting experiment and may point to an approach with which businesses may want to experiment. You can read the paper here on the non-peer-reviewed research repository arxiv.org. The rise of the AI gadget could free us from our smartphones. We just need to find the right device -- by David Meyer The U.S. reigns supreme in AI startups while China ensures chatbots have 'core socialist values' -- by Jason Ma Industry leaders say companies are adopting AI, but cost and reliability remain key challenges -- by Sharon Goldman July 21-27: International Conference on Machine Learning (ICML), Vienna, Austria Dec. 8-12: Neural Information Processing Systems (Neurips) 2024 in Vancouver, British Columbia Dec. 9-10: Fortune Brainstorm AI San Francisco (register here) The results are in from Sam Altman's big study of universal cash transfers. One of the really big questions among those who think we may be approaching artificial general intelligence (AGI), is what would happen to everyone who might be put out of work? Many, including OpenAI's Sam Altman, have postulated that some form of universal basic income (UBI) will be necessary -- and that you could fund it by taxing the profits of AI companies or the businesses that see their productivity soar and costs decrease because of AI. Altman and his OpenResearch foundation funded the largest randomized study to date of universal cash transfer (or UCT, not UBI exactly because the amounts weren't large enough to qualify as a full income). It involved some 3,000 people in Illinois and Texas. That's more people than any previous study. And it also involved much larger cash transfers than prior studies: $1,000 per month for those getting the transfers, and $50 a month for those in the control group. That study just reported its results yesterday. The findings: The cash transfers did have an impact. People spent more on necessities such as food, rent, and transportation. They were more likely to lend money to friends and relatives. Those getting the larger cash disbursement reported feeling more financially secure and their spending became less volatile month to month which might indicate better financial health. There was some small, but significant increase in people's willingness to dream of becoming entrepreneurs -- a 5% increase compared to the control group over the whole three years of the study -- but no impact on people actually following through on those dreams and starting companies. You can read more about the research in this Bloomberg story. The researchers admitted in their findings that the amounts of money being doled out were not enough to really make an impact on issues such as chronic health issues, lack of childcare, or lack of access to affordable housing. And I think the study also pointed to the real flaws of the idea of UBI. The amount of money you'd have to dole out to really give most Americans a basic income is so high there's no way you could ever raise it without the debt soaring or the tax burden on companies being suffocatingly high. I asked Altman about this directly last year when I interviewed him for my book, Mastering AI. Could UBI ever be affordable? "Today, no. But if global wealth 10xed, then sure," he said. I asked if AI could do that. "AI and energy together," he replied. (He also invested in fusion power to drive down the cost of energy.) Well, I guess we're still waiting for AGI and fusion then.

AI Chatbots Have a Donald Trump Problem

The main thing about chatbots is that they say things. You chat, and they chat back. Like most software interfaces, they're designed to do what you ask. Unlike most software interfaces, they do so by speaking, often in a human voice. This makes them compelling, funny, frustrating, and sometimes creepy. That they engage in conversation in the manner of an assistant, a friend, a fictional character, or a knowledgeable stranger is a big part of why they're valued at billions of dollars. But the fact that chatbots say things -- that they produce fresh claims, arguments, facts, or bullshit -- is also a huge liability for the companies that operate them. These aren't search engines pointing users to things other people have said or social media services stringing together posts by users with identities of their own. They're pieces of software producing outputs on behalf of their owners, making claims. This might sound like a small distinction, but on the internet, it's everything. Social-media companies, search engines, and countless other products that publish things online are able to do so profitably and without debilitating risk because of Section 230, originally enacted as part of the Communications Decency Act in 1996, which allows online service providers to host content posted by others without assuming liability (with some significant caveats). This isn't much use to companies that make chatbots. Chatbots perform roles associated with outside users -- someone to talk to, someone with an answer to your question, someone to help with your work -- but what they're doing is, in legal terms, much closer to automated, error-prone publishing. "I don't think you get Section 230 immunity on the fly if you generate a statement that seems to be defamatory," says Mark Lemley, director of the Stanford Program in Law, Science & Technology. Sam Altman has acknowledged the concern. "Certainly, companies like ours bear a lot of responsibility for the tools that we put out in the world," he said in a congressional hearing last year calling for new legal frameworks for AI. "But tool users do as well." Absent an immensely favorable regulatory change, which isn't the sort of thing that happens quickly, this is a problem for firms like Altman's OpenAI, whose chatbots are known to say things that turn out to be untrue. Chatbots, as Lemley and his colleagues have suggested, might be designed to minimize risk by avoiding certain subjects, linking out a lot, and citing outside material. Indeed, across the industry, chatbots and related products do seem to be getting cagier and more cautious as they become more theoretically capable, which doesn't exactly scream AGI. Some are doing more linking and quoting of outside sources, which is fine until your sources accuse you of plagiarism, theft, or destroying the business models that motivate them to publish in the first place. It also makes your AI product feel a little less novel and a lot more familiar -- it turns your chatbot into a search engine. This is about much more than legal concerns, however. The narrow question of legal liability gives us a clear way to think about a much more general problem for chatbots: not just that they might say something that could get their owners sued -- in the eyes of the law, large language model-powered chatbots are speaking for their owners -- but that they might say things that make their owners look bad. If ChatGPT says something wrong in response to a reasonable query, a user reasonably might feel it's OpenAI's fault. If Gemini generates answers that users think are politically biased, it's Google's fault. If a chatbot tends to give specific answers to contested questions, someone is always going to be mad, and they're going to be mad at the company that created the model. This, more than legal liability, is clearly front of mind for AI companies, which over the past two years have enjoyed their first experiences of politicized backlash around chatbot outputs. Attempts to contain these episodes into appeals to "AI safety," an imprecise term used to describe both the process of sussing out model bias in consumer software and efforts to prevent AI from killing every human on Earth, have resulted in messy backlash of their own. It helps explain stuff like this: It is the curse of the all-purpose AI: A personified chatbot for everyone is doomed to become a chatbot for no one. You might, in 2024, call it AI's Trump problem. This stubborn problem might shed light on another, more local mystery about chatbots: what AI companies want with the news media. I have a theory. In recent months, OpenAI has been partnering with news organizations, making payments to companies including Axel Springer, the Associated Press, and New York parent company Vox Media. These deals are covered by NDAs, but the payments are reportedly fairly substantial (other AI firms have insinuated that they're working on similar arrangements). According to OpenAI, and its partners, the value of these partnerships is fairly straightforward: News organizations get money, which they very much need; OpenAI gets to use their content for training but also include it in forthcoming OpenAI products, which will be more searchlike and provide users with up-to-date information. News organizations, and the people who work at them, are a data source with some value to OpenAI in a world where lots of people use ChatGPT (or related products), and those people expect it to be able to address the world around them. As OpenAI CEO Brad Lightcap said at the time of the Axel Springer partnership, such deals will give OpenAI users "new ways to access quality, real-time news content through our AI tools." But in the broader context of OpenAI's paid partners, news organizations stand out as, well, small. A partner like Stack Overflow, an online community for programmers, provides huge volumes of relevant training data and up-to-date third-party information that could make OpenAI's products more valuable to programmers. Reddit is likewise just massive (though presumably got paid a lot more) and serves as a bridge to all sorts of content, online and off. News organizations have years or decades of content and comments, sure, and offer training data in specific formats -- if OpenAI's goal is to automate news writing, such data is obviously helpful (although of limited monetary value; just ask the news industry). If news organizations have unique value as partners to companies like OpenAI, it probably comes down to three things. One, as OpenAI has suggested, is "quality, real-time news content" -- chatbots, if they're going to say things about the news, need new information gathered for them. Another is left unspoken: News organizations are probably seen as likely to sue AI firms, as the New York Times already has, and deals like this are a good way to get in front of that and to make claims about future models being trained on clean data -- not scraped or stolen -- more credible. (This will become more important as other high quality data sources dry up.) But the last reason, one that I think is both unacknowledged and quite important, is that licensing journalism -- not just straight news but analysis and especially opinion -- gives AI companies a way out of the liability dilemma. Questions chatbots can't answer can be thrown to outside sources. The much broader set of questions that chatbot companies don't want their chatbots to answer -- completely routine, normal, and likely popular lines of inquiry that will nonetheless upset or offend users -- can be handed off, too. A Google-killing chatbot that can't talk about Donald Trump isn't actually a Google-killing chatbot. An AI that can't talk about a much wider range of subjects about which its users are most fired up, excited, curious, or angry doesn't seem like much of a chatbot at all. It can no longer do the main thing that AI is supposed to do: say things. In the borrowed parlance of AI enthusiasts, it's nerfed. And so you bring in other people to do that. You can describe this role for the news media in different ways, compatible with an industry known for both collective self-aggrandizement and individual self-loathing. You might say that AI companies are outsourcing the difficult and costly task of making contentious and disputed claims to the industry that is qualified, or at least willing, to do it, hopefully paying enough to keep the enterprises afloat. Or you might say that the AI industry is paying the news media to eat shit as it attempts to automate the more lucrative parts of its business that produce less animosity -- that it's trying to buy its way through a near future of inevitable, perpetual user outrage and politically perilous backlash and contracting with one of the potential sources of the backlash to do so. Who can blame them? This isn't unprecedented: You might describe the less formal relationship between the news media and search engines or social media, which rewarded the news media with monetizable traffic in exchange for "quality, real-time news content," in broadly similar terms. Google and Facebook hastened the decline of print and digital advertising and disincentivized subscription models for publishers; at the same time, absent a better plan, the news media lent its most valuable content to the platforms (arguably at a cost, not a benefit, to their brands). But it's also distinct from what happened last time: AI firms have different needs. Social media feels alive and dynamic because it's full of other people whom the platforms are happy to let say what they want. Chatbots feel alive and dynamic because they're able to infinitely generate content of their own. The main question this would raise for media companies -- and, hey, maybe it's not about this at all! -- is whether they're being paid enough for the job. Functioning as newswires for what AI firms are hoping represents the future of the internet is arguably a pretty big task, and serving as a reputational sponge, or liability sink, for ruthless tech companies sounds like pretty thankless work.

From hype to hesitation, Wall Streeters open up about how they're really using AI

Speed and ease -- that's how generative AI is changing the game for finance professionals. Because AI singled out an emerging trend and identified a startup poised to benefit from it, Bain Capital Venture's Christina Melas-Kyriazi beat out other investors to lead a fundraising round mere days after discovering it. "We were able to do that because we could sort of identify it fast enough. Speed is really important here," she told Business Insider. And for T. Rowe Price's Sébastien Page to capture the "nuance" of how the firm's analysts felt about each sector in the past? "It would've been pretty much impossible," the head of global multi-asset and the firm's chief investment officer told BI. "It would have been a lot more anecdotal." Generative AI is threatening to upend many industries with its ability to crunch data and spit out new information in a humanlike way -- and Wall Street is no different. As use cases are being built, tested, and scaled among workers, Business Insider talked to 35 people in various roles across the finance industry to hear how AI adoption is happening on the ground. Many of these people were granted anonymity to speak freely about their experiences. From C-suiters to junior staff, most said they welcome generative AI's potential to boost efficiencies and cut grunt work. They believe that less administrative friction will allow more time for core work and raise the bar for critical thinking and analysis. "I hope it does replace a lot of the analyst work," one former junior banker at a New York-based investment bank told BI, referring to the all-nighters entry-level investment banking analysts spend assembling pitch decks or punching numbers into Excel spreadsheets. "A lot of the analyst work is bullshit." Financial advisors and analysts in JPMorgan's wealth and asset management business are saving a couple of hours a day, BI previously reported. At Man Group, a machine-learning tool can send brokers trades that could offer the best pricing on a specific stock or security based on historical execution data. "It's a real source of alpha," said Eric Burl, the head of discretionary at Man Group, the world's largest publicly listed hedge fund. While many express enthusiasm for AI tools that allow them to save time and potentially focus on bigger and better things, others are more cynical. Some raised doubts about the technology's reliability and usefulness, concerns about their firms' approach to using AI, and questions about how the technology will affect jobs or work-life balance. If AI tools can eliminate much of Wall Street's entry-level work, it could shake up typical career paths. But only time will tell how this will change Wall Street's work-till-you-drop culture. In a survey of 780 banking and capital-markets employees by Accenture Research, 62% of respondents expect generative AI to increase people's stress and burnout. "The hours aren't going to change, because they're a product of the culture more than actual workload," the former junior banker said. "People will always find things for you to do because the expectation is that you work 80 or 100 hours a week." Finance has always been and will no doubt remain a competitive business where success depends on the speed at which information is obtained and acted upon. So it should come as no surprise that in the age of AI, Wall Street firms have burst out of the gate to leverage the technology to get a leg up on the competition. In the race to unlock AI's potential, they're testing investing models, hiring top talent, and developing their own cutting-edge research. Private-equity firms such as Blackstone are building teams to leverage AI's cost-cutting and productivity-inducing benefits within the companies they own. The middle-market PE firm Thomas H. Lee, which launched a generative-AI coding tool for select portfolio companies, reported that its engineers were up to 30% more productive just four weeks after the rollout. Quantitative trading firms and hedge funds such as Two Sigma use powerful compute engines and AI chips to uncover new sources of investing alpha. Meanwhile, consumer banks like JPMorgan, with their sprawling technology footprints and troves of data about how consumers spend, save, and invest, have been focused on readying their data strategies to take full advantage of AI. But there's no one-size-fits-all approach to tapping AI's benefits, and finance doesn't have the best reputation for integrating tech. Because it is such a nascent field, firms are still trying to figure out the most effective, efficient, and safest way to develop and scale the technology before unleashing it on the masses. "There's definitely a first-mover advantage," Keri Smith, Accenture's global banking data and AI lead, told BI. Finance firms that have already been investing in technology modernization, like migrating to the cloud and making sure enterprise data is well organized and tagged, are poised to step out in front of the pack. Smith said that after 18 months of experimentation and development, Wall Street's understanding and use of AI has matured significantly. This has led to thornier questions about how companies can differentiate themselves from the crowd and how to best develop and train talent. These are the puzzles that chief information officers, chief technology officers, and data leaders who oversee their firms' AI strategies are expected to solve. Those roles now have "a very important seat at the table," Ken Griffin, the billionaire Citadel founder and CEO, said in May at the Milken Institute Global Conference. "They've got the attention of the CEO. How can we use technology again to really drive productivity?" The pressure to mine productivity gains has put some of Wall Street's tech leaders in a tricky place: how to gain a first-mover advantage while not exposing the enterprise to vulnerabilities from moving too fast. Goldman Sachs' tech chief, Marco Argenti, told BI he had to push back against engineers who wanted the firm to roll out a new generative-AI coding tool more quickly. "You go slow to go fast," Argenti said. "It's such a big revolution that you need to be able to go faster safely. So, at the beginning, you need to take careful steps so that you remove a lot of that toil" involved in building AI applications, he said, like protecting confidential data and retrieving information in a way that will minimize inaccuracies. "At the beginning of the process, yes, I had to curb people's enthusiasm a little bit," he said. At one midsize Wall Street investment bank, a senior banker uses ChatGPT daily -- a sign of how AI is seeping into the analog world of investment banking, where bankers' interpersonal connections and flair in the boardroom have long reigned supreme. The banker told BI he's "retraining" himself to use tools like ChatGPT and Copilot more frequently. It's transformed how he researches "everything," from data about industry sectors to ideating new proposals to bring to clients. "Rather than just brainstorming with colleagues, you might brainstorm with a robot now," he said, "which I've actually found to be pretty helpful because it spurs some new thoughts." The banker primarily uses AI to compress dense troves of information -- whether from research notes or dozens of meetings -- into more digestible takeaways. "Now I don't have to read six reports," he explained. "I can query those reports through AI and get a pretty snappy summary of what they are, and then I can take that and use my own brain at that point to put it all together." His junior team members are hopping on the bandwagon to accelerate tasks like making slide decks, even if they're sly about it. "I know for a fact that our analysts are using it quite a bit to help them write and create prose and bullet points," he said, adding: "They don't readily admit it -- until you sit down and have a couple beers with them." But with the elimination of some of the drudge work, there's the risk that some jobs could become obsolete. "Maybe we don't need as many analysts down the road as we otherwise would," the banker conceded. Since most roles in finance include a lot of collecting and processing data, there's no question that generative AI is set to shake up jobs on Wall Street. A report from Accenture Research found that capital-markets roles are ripe for AI-related job displacement. The consulting firm estimated that 72% of jobs within investment banks, asset managers, and wealth advisories have "higher potential" to be automated or augmented by AI. But just as Excel didn't replace accountants, tech leaders don't see AI displacing humans. "Employees with AI skills will replace people without AI skills," Andrew Chin, the chief AI officer at the $759 billion money manager AllianceBernstein, told BI. In many cases, AI is simply an enabler, Lisa Donahue, a partner at the global consulting firm AlixPartners, which is best known for its work cleaning up messy balance sheets and turning around troubled companies. "It enables executives to get information in a comprehensive way faster, which allows you to make your decisions faster and quickly move toward execution," Donahue said. AlixPartners, which advises private-equity firms, uses a proprietary, AI-powered diagnostics tool that draws on decades of the firm's consulting work to help buyout shops evaluate potential acquisition targets. When it comes to implementing and interpreting recommendations, "you're still going to need experienced people to execute," she said. Jobs across the industry vary, obviously, but if you're in the business of giving advice and influencing outcomes, it's time for those hard-for-a-machine-to-replicate skills to shine. One fundamental analyst at a large hedge fund, who's seen more accurate summaries of earning calls and research reports as the use of generative AI has become more widespread, said that the tech has allowed him "to be more thoughtful" thanks to the additional time he has. He uses the extra time to write notes for his portfolio manager or craft questions for the management teams he covers. A data scientist at a midsize hedge fund told BI that generative AI models are a "superpower for coders." One of his biggest use cases is solving coding problems in different coding languages. He's been freed up to go deeper or think more abstractly about his projects, knowing the coding chunk won't take as long as it used to. He compares it to when writers moved from typewriters to word processors, with spell-check, the ability to delete and move items around, and more. He said he'd even take on the expense of ChatGPT himself if his firm stopped covering it. Not everyone in finance is convinced that generative AI will bring radical changes. Some employees at firms that have long used models powered by AI to seek an edge, such as quantitative hedge funds or trading firms, say a lot of the benefits are overhyped. "There's a lot of talk, but I haven't seen anything yet that changed the world," a quantitative portfolio manager told BI, adding that "all those GPT models might present somewhat of an improvement, but we haven't seen anything dramatic like a breakthrough." Other cohorts, including fundamental investors who make their living by picking the best stocks, believe their investing style is too nuanced to rely on automation. It's a "lack of available use cases rather than a deliberate decision not to," a fundamental analyst at one of the world's biggest hedge funds told BI. Two other fundamental analysts BI spoke with agreed there was no explicit use of AI in their processes. Though some say OpenAI's debut of ChatGPT represented a step change in AI and machine-learning capabilities, these generative AI models can still spew out misinformation, which means humans need to check the work done by AI. Not doing so is a mistake that some have learned the hard way. At one large hedge fund, some analysts have had to redo entire reports after realizing the numbers pulled by ChatGPT were incorrect, according to a colleague at the firm. These reports, which typically take half a day or so to complete, were generated by the bot almost instantly, but the analysts realized it used the wrong revenues and profits to draw up the analysis. "We trust it as much as an intern," the hedge-fund employee said. "You have to check its work." An emerging challenge for Wall Street firms now is closing the gap between the staff and the technology, and some firms are finding a "bit of friction" with adoption, Accenture's Smith said. A January report by the firm found that 93% of finance workers and employees are very keen to leverage generative AI, but Smith said their organizations' rollout of AI tools and approach to upskilling are a source of frustration. "I wish I could use it more," one midsize private-equity firm vice president told BI. Their firm blocked employees from using publicly available generative-AI tools and built its own model using OpenAI. "For some reason, we decided to become software developers and build our own shit versus just buying off-the-shelf stuff from firms who do this for a living," they said. Only 7% of banking and capital-markets organizations are actively reskilling their workforces at scale, Smith added. "I think I'm using it?" said one wealth advisor. His employer, one of the world's largest brokerages, has developed an internal AI product that analyzes client data and generates reports. One of his interns showed him how to use ChatGPT; other team members have used it to summarize hundred-plus-page private-equity offerings. The advisor plans to spend more time this summer dabbling with it. An advisor at Northern Trust also cited age as a hurdle to adoption. "I can see them being very apprehensive," he said of his colleagues in their late 50s and early 60s. "Why would they want to learn something new when they are so close to retirement?" At Citibank, employees are encouraged to raise new AI use cases. One potential use case could be summarizing hundreds of pages of regulatory documents into a list of the relevant obligations, one Citi employee told BI. A human would still need to double-check and validate that the information was correct, "but that just saved you so much time that's wasted just reading," they said. However, the process for getting any type of AI approved is arduous and can take months, the person said. "A lot of checks and balances, a lot of validating, a lot of evidence-based artifacts need to be provided and committees needing to review and approve," they said of the AI approval process, adding that "it will just beat you up." Some firms are not only automating grunt work but also looking to replicate work that could be considered unique. JPMorgan's private bank launched an AI tool that acts as an assistant to its bankers. The firm's ultimate aim is to use generative AI to replicate the success of its best bankers for all advisors. The firm is working to train the AI copilot on data that breaks down how the bank's top advisors respond to emails, how they handle client interactions, and what their portfolios look like. One former JPMorgan executive director told BI he was unnerved by the prospect of an AI tool that helps advisors mimic how top performers respond to emails or communicate with clients. It could cause tension between advisors who think their best practices are being shared with competing advisors in the same market. "I think it would be weird to have someone's intelligence feeding a tool," he said.

Is generative AI project success in 2024 a realistic goal? A mid-year reassessment

In January 2024, I put an AI stake in the ground: Attention enterprises - your AI project success in 2024 is not a given. What will separate wins from failures? I revealed my top eight underrated keys for AI project success - and coined phrases like "AI overreach," for the pitfall of expecting AI to do more than it's currently capable of. So, as we regroup from a blitz of AI anticipation from enterprise keynotes round one, where are we now? Many questions remain unanswered, but we do have more clarity. 1. The AI dissonance factor kicks in - we're now running into a troubling issue I call AI dissonance, where research on generative AI adoption boomerangs from optimistic to cautious (the cautions trace back to concerns on security, data privacy, and trust). Therefore, you can cherry pick an AI study to bolster any AI narrative you want, from 'immature technology' to 'enterprise game changer,' so that doesn't help us much. 2. Gen AI adoption is mostly through vendor-driven pilots - don't go looking to project ROI for proof points either: not yet. To date, most enterprise gen AI adoption is within the shipped solutions of third party vendors, not home grown projects. Most of these projects are still in the pilot or early adopter phase, though as we head into the fall, we're starting to see a bit more scale, mostly in the context of basic productivity enhancements like AI meeting summaries, email composers, and job description generators (No, I don't consider that AI content "creation," but that's an axe to grind later on). 3. "Responsible" AI architectures separate enterprise AI - the biggest thing enterprise vendors got right this spring? Prioritizing their "responsible AI" platforms. I'm using "responsible" less about ethics (that's another potent discussion), and more about improving gen AI output accuracy. In my AI vendor deep dives, I've documented the use of small language models, RAG, knowledge graphs, reinforcement learning, individualized customer models, and industry-specific LLMs. These refined architectures don't eliminate gen AI's output issues, but they do result in more relevant/accurate output than out-of-the-box consumer LLMs can pull off. On the extreme/extremely interesting side, I just spoke with the CEO of Zapata AI, "home of the industrial generative AI revolution." Due to the unreliability of LLM output, they only use smaller, domain-specific generative models. Therefore, they sever the link between LLMs and what we call "generative AI." That forces a rethink, wouldn't you say? Not to mention the potential to reduce AI operating costs, a surging concern. Also see: Joe McKendrick's Your business is going to rely on hundreds of AI models. Here's why, and Andy Thurai's Beyond the Hype: The Future of Generative AI Isn't Just Massive GenerativeAI Models. 4. AI overreach is still a project nemesis - Companies are still indulging in AI overreach for a variety of ill-conceived reasons (poor AI tech selection, excessive/premature head count reductions, etc.) The constant face palm gen AI headlines of chatbots gone awry aren't doing enterprise AI vendors any favors. They reinforce the worst perceptions of generative AI at scale. Which brings us to: 5. The financial markets are wising up - the financial markets are getting savvier to the shortcomings of generative AI, including the landmark study by Goldman Sachs, summed up by 404 Media in this headline: Goldman Sachs: AI Is Overhyped, Wildly Expensive, and Unreliable. Does this mean generative AI is heading towards blockchain in the 'wildly overrated' category of enterprise tech? No. But it means there is a more sober dialogue now, where the ROI of gen AI outweighs the pressure to adopt for enterprise FOMO. Whether this report tells the whole story is a worthy debate, but a more exacting look at the pros and cons of gen AI is overdue. Training narrower models on industry-specific data is an alternate way forward. Consumer gen AI critics are not always aware that enterprise vendors are pushing into different approaches, informed by customer-specific data - and applying them to sensible use cases where 100 percent accuracy is not required. Contrast that with say, New York City refusing to take down a bot for small businesses that gives out bad (and sometimes illegal) advice. But here's the catch: enterprise perceptions of AI are influenced by stories like Google Gemini recommending glue on pizza. The necessity for human/adult supervision of the vast majority of gen AI use cases changes the ROI equation - something gen AI evangelists seem to struggle to grasp. Goldman Sachs can't afford that misstep. 6. The regulatory environment remains uncertain - though it's not yet in effect, the EU AI Act remains a formidable piece of legislation - causing "big AI" players to rattle economic sabers, e.g. Meta pulls plug on release of advanced AI model in EU. Meanwhile, the US regulatory mood is far from certain. In my view, the US has dragged its feet to the point of regulatory incompetence (those who caution against AI regulations obviously feel otherwise; I believe smart regulations are well worth the hair loss it takes to craft them). The November election will give a big ol' clue as to where we are headed. For now, regulatory clarity in the US eludes. 7. AI readiness is a thing - enterprises may or may not be eager to adopt (see: AI dissonance), but generative AI has provoked a vigorous data governance and quality conversation. Companies that test generative AI soon realize that "garbage in, garbage out" applies with gusto. Best cast: companies can now earn buy-in for long-postponed data quality projects - though some of the most relevant data may lie outside enterprise walls, or in API-hostile silos (check my Enterprise Month in Review podcast on data pitfalls with Maureen Blandford). Multi-year data overhauls won't get the green light. Stack up wins with faster analytics projects and AI apps, and maintain data cleansing buy-in. As I wrote about Accenture's gen AI earnings: AI readiness is a thing, and it's luring companies into tech (modernization) spending. That may end up boosting Accenture's numbers more than gen AI does. Stuart Lauchlan quotes Accenture CEO Julie Sweet: 'It is important to remember that while there is a near universal recognition now of the importance of AI, which is at the heart of re-invention, the ability to use gen AI at scale varies widely with clients on a continuum. With those which have strong digital cores genuinely seeking to move more quickly, while most clients are coming to the realization of the investments needed to truly implement AI across the enterprise, starting with a strong digital core from migrating applications and data to the cloud, building a new cognitive layer, implementing modern ERP and applications across the enterprise to a strong security layer.' 8. IP lawsuits are unresolved - many IP lawsuits against "big AI" are unresolved. Companies that want to build home grown generative AI apps will need lawyers on speed dial, scouring the terms and conditions of open source models. For now, most enterprises will be able to minimize IP lawsuit/copyright risk by consuming generative AI through their trusted vendor - but they will still need lawyers to hammer out contracts that address IP liability with those vendors. And they'll need to ask the right questions about how their own data will be used by vendors, including model training. Also see: Brian Sommer's Asking some tough questions for new software selections in a generative AI era. 9. AI pricing isn't sorted, therefore ROI isn't sorted - how can companies evaluate generative AI projects when AI pricing itself is a moving target? Even if a vendor embeds AI in its core licensing, we don't know how that will change as vendors reckon with the true costs of operational AI. Vendors won't love me for this, but I hold the view that charging for co-pilots isn't going to be viable long term: My napkin-scribble, non-crystal-ball prediction is: vendors won't be able to get away with charging for "co-pilots" and other forms of embedded AI, as those will simply be how software is expected to be consumed (though there could be overage charges for extreme overuse). Most AI services will likely be consumption-based. Value-based pricing could factor in, but bottom line: vendors will be able to charge an AI premium only for services/apps that let customers do something they've never done before - e.g. significant value creation or significant cost/labor savings that ends the debate on AI-driven productivity. When customers see transformative value, they'll pay a premium for it. Until then, we'll have adoption questions, and cautious AI evaluations. On diginomica, Brian Sommer sorted through AI pricing models in Clarity may be emerging in AI capabilities pricing. Here's how. Over at Constellation Research, Larry Dignan recently rounded up some AI pricing approaches. As I take my own AI certifications this summer, I get a heavy dose of the mathematical sophistication behind today's AI tools. And yet the concerns jump out too: How do we achieve trust in a deepfake world? Trust is now at a new premium - in AI and beyond. Will the need for trust around AI create openings for upstarts? The "Big AI" vendors may have the data and the scale, but they don't really excel at this trust and transparency thing. See Louis Columbus' How adversarial AI is creating shallow trust in deepfake world. AI as an accelerant - applying AI seems to bring out the best (or the worst) in companies. Have a KPI-obsessed, digital surveillance culture? Chances are, adding gen AI to that mix will make it worse - at least for the (remaining) employees. Have a well-governed data infrastructure and a reputation for employee well-being and skills management? I like your chances better. AI agents are an overhyped, kick-the-can-down-the-road distraction - AI agents should one day be an asset to compile workflows into a processes a user can summon - or that happen automatically. But for now, vendors that overhype agents are putting off today's AI with promises of a better tomorrow. My upcoming piece on agents will dig in, but for now: Yeah, stitching together unreliable workflows sounds awesome. Also see: New paper: AI agents that matter. Early use cases with traction - my year-end AI projects round-up included a couple interesting generative AI go-live stories, with documented value (Can enterprise LLMs achieve results without hallucinating? How LOOP Insurance is changing customer service with a gen AI bot) - but we haven't surfaced too many more since then. Beyond productivity assistants/co-pilots, the early gen AI use cases that keep coming up include service and support assistance (including first-level, front line support, though there are design/accuracy issues to bear down on there). Meeting summaries with action items seem really useful, if not earthshaking (generative AI certainly excels at summarizing and transforming one media into another, including longer to shorter videos, a solid B2B use case). Yes, some content generation - though I like the translation between different mediums (and languages) better than the so-called "creation" use cases. And, for vendors with big development teams, coding assistants are definitely in play. Don't call it "creation" call it content generation - I've put out an entire video and podcast series on my views on generative AI and "creativity," the most recent podcast with diginomica contributor Barb Mosher Zinck (also see Barb's Is content creation the best use for generative AI in marketing?). Will gen AI empower/amplify human creations? Or will we become admin assistants to robot creators? Chris Middleton has been killing it on this topic all year. In hits/misses, I went off: Why trivialize creation more than it actually is? Gen AI's content abilities are due to the sweat (non)-equity of all the artists who unwittingly provided the uncompensated training for these systems (not to mention there is still a human element involved in pulling out the right stuff with the right prompt). For now, moving beyond productivity obsessions would be nice. Where are the imaginative use cases? As I wrote in my Team Liquid use case: So many AI stories this spring were about productivity pilots. But what about creative app building? What about utilizing big data sources in imaginative new ways? It's also good to hear from vendors that aren't obsessed about prematurely squeezing the human out of the loop. We don't just judge you by your AI tech, we just you by the humanity in your AI vision. As in this from gen AI service startup Cresta, via a recent talk with Cresta CEO Ping Wu Accelerate onboarding and empower every agent to perform like your best with proactive co-pilots that suggest proven best practices, generate accurate answers, and reinforce compliance at exactly the right moments throughout each conversation. We could stand more talk like this from other vendors about empowering agents, rather than finding ways to prematurely eliminate them, and facing the wrath of customers - an unpleasant indicator of AI overreach. The energy costs of gen AI are heating up the ROI and ethics questions - e.g.AI's Energy Demands Are Out of Control. Welcome to the Internet's Hyper-Consumption Era. Avoiding the overuse of LLMs when smaller models will do will help with this issue, but this topic isn't going away. The science of AI rolls on - somebody send a memo to tech marketing teams - most vendors present gen AI as either a finished/amazing/"intelligent" product, or one that will inevitably evolve into something that will solve itself - even though the scaling of generative AI training data is reaching diminishing returns, which Sam Altman of OpenAI himself has conceded. Most enterprise vendors are putting RAG lipstick on a powerful but imperfect technology. Meanwhile, some of the key scientists who laid the groundwork for generative AI are back in the lab, aware that today's tech needs a breakthrough. That breakthrough into truly cognitive systems will come someday, and we need to talk about that day, and what it will mean. Those advances will come, but not by licensing Reddit content. The history of AI indicates these breakthroughs will likely come in surprising, and perhaps even non-commercial ways, perhaps in an obscure lab where no investors are breathing ROI fire. This powerful tech is good enough at some very useful things. It's also incredibly good at some not very nice things (e.g. deepfakes and disinformation at scale), so we have our hands full - without pretending these are cognitive systems. My gen AI transcript summaries are amazing, but in the next instance a rogue (unspoken) word like "Holocaust" pops up in my transcript, a word I would not publish without intention, a word any truly intelligent system would flag to my attention. It's a reminder that these tools, for all their enterprise spit and polish, are predicting the next word, not grasping the essence. Not yet. Given the planning we need for disrupted workforces, maybe those limitations are a good thing. I won't repeat the AI project tips I outlined in my January AI projects piece, except for this one: The customer should define their own "customer success" AI metrics for each project, agreed to by vendor and/or services partner. Ask potential vendors what their definition of AI success will be, and how it will be tracked - but in the end, it's your metrics that all parties should buy into. That's a good place for any AI project discussion to start. When the fall event season kicks in, I expect we'll hear more from customers in the midst of generative AI pilots - and we look forward to documenting their results and pitfalls. Note: I'll embed even more reference links tomorrow night.