LLMs

Twitter

Facebook

Copy Link

Key Facts

Year2018

Key PlayersOpenAI, Google, Anthropic

Top News

Eric Schmidt's 1% Anthropic Stake Could Net Him $20 Billion as AI Company Eyes $2 Trillion IPO

Former Google CEO Eric Schmidt owns about 1% of Anthropic after investing tens of millions in the AI startup's 2021 Series A round at a $461 million valuation. With the company's planned IPO targeting valuations between $1.5 trillion and $2 trillion, Schmidt's stake could be worth up to $20 billion.

Business8 sources

Eric Schmidt's 1% Anthropic Stake Could Net Him $20 Billion as AI Company Eyes $2 Trillion IPO

Business8 sources

MiniMax M3.1 Flash Slashes AI Costs by 70% but Independent Tests Reveal Accuracy Trade-Offs

Chinese AI startup MiniMax released M3.1 Flash, delivering 70% cost reduction and 3X faster inference speeds for enterprise workflows and software development. Independent benchmarks show competitive performance against DeepSeek at lower costs, but Artificial Analysis found the model generates incorrect answers rather than acknowledging uncertainty.

Technology4 sources

MiniMax M3.1 Flash Slashes AI Costs by 70% but Independent Tests Reveal Accuracy Trade-Offs

Technology4 sources

Researcher Claims AI Models Are More Understandable Than Previously Thought, Sparking Industry Debate

Nick Cammarata from Goodfire published a viral 26,000-word blog post challenging the belief that AI models are impenetrable black boxes. His claim that neural networks are more interpretable than assumed has ignited celebration and skepticism across the AI safety community, with industry leaders calling it a potential breakthrough.

Science2 sources

Researcher Claims AI Models Are More Understandable Than Previously Thought, Sparking Industry Debate

Science2 sources

Open Models Capture 71.5% Token Share as Enterprise AI Competition Intensifies with New US Releases

Open weight AI models processed 71.5% of tokens through Vercel's AI Gateway by September, with spending share jumping from 5% to nearly 25% in three months. US labs Reflection AI and Nvidia released new enterprise-ready open models to compete with Chinese alternatives from DeepSeek and Qwen, as the focus shifts from individual models to complete AI platforms.

Technology4 sources

Open Models Capture 71.5% Token Share as Enterprise AI Competition Intensifies with New US Releases

Technology4 sources

AI Cracking Ciphers: Amateurs Solve 200-Year Historical Codes in Hours Using GPT-6 Astra

AI is breaking old encryption mysteries that stumped experts for decades. GPT-6 Astra cracked an 85-year Enigma message and a Napoleon-era cipher in hours, marking a turning point for historical cryptography. But the surge brings mixed feelings as technology drains the satisfaction from code-breaking while transforming historical research.

Technology5 sources

AI Cracking Ciphers: Amateurs Solve 200-Year Historical Codes in Hours Using GPT-6 Astra

Technology5 sources

TypeSafe AI's Jev Decision Model Proves Vulnerable to Prompt Injection Despite Security Claims

TypeSafe AI's Jev decision model, marketed as resistant to manipulation, can still be exploited through prompt injection and adversarial text attacks, according to HiddenLayer security research. Tests showed the System One model succumbing to clickjacking and TokenBreak attacks, raising questions about its real-world security advantages over traditional LLMs.

Technology2 sources

TypeSafe AI's Jev Decision Model Proves Vulnerable to Prompt Injection Despite Security Claims

Technology2 sources

XBOW Builds Chrome Exploit with OpenAI's GPT-5.6 Sol in Six Days, Wins $250,000 Google Bounty

Security startup XBOW used OpenAI's GPT-5.6 Sol model to chain three zero-day vulnerabilities in Chrome, creating a full-chain exploit in just six days. Google paid the team $250,000, its top bounty for Chrome exploits. The attack exposed about one billion Windows users before Google patched the bugs in late August and early September.

Technology3 sources

XBOW Builds Chrome Exploit with OpenAI's GPT-5.6 Sol in Six Days, Wins $250,000 Google Bounty

Technology3 sources

AMD Zen 5 AI Workstations Pack 192GB RAM to Run 284-Billion-Parameter Models Locally

AMD launches Ryzen AI Max+ PRO 495-powered compact AI workstations through iBuyPower and Minisforum, featuring 16 Zen 5 cores and up to 192GB unified memory. The systems enable developers to run large AI models with 284 billion parameters locally, marking a shift in accessible AI development hardware.

Technology3 sources

AMD Zen 5 AI Workstations Pack 192GB RAM to Run 284-Billion-Parameter Models Locally

Technology3 sources

Meta Stock Surges 6% as Muse Wonder AI Model Reaches Industry Frontier, Analysts Eye $10B Revenue

Meta stock jumped roughly 6% over two sessions following the launch of its Muse Wonder AI model family, which analysts say has reached the AI frontier. The release comes as Meta prepares to report third-quarter results, with Wall Street focused on whether massive AI spending can translate into sustainable revenue growth beyond advertising.

Technology11 sources

Meta Stock Surges 6% as Muse Wonder AI Model Reaches Industry Frontier, Analysts Eye $10B Revenue

Technology11 sources

Facebook Algorithm Now Lets Users Customize Feed With AI-Powered Your Algorithm Feature

Meta launched Your Algorithm on Facebook, an AI-powered feature that lets users customize their feed algorithm using natural language. Users can now directly tell the platform what content they want to see more or less of in Feed and Reels, with recommendations updating in real time based on their preferences.

Technology5 sources

Facebook Algorithm Now Lets Users Customize Feed With AI-Powered Your Algorithm Feature

Technology5 sources

Moonshot AI Executives Navigate $700M Shareholder Dispute Before Planned Hong Kong IPO

Chinese AI startup Moonshot AI is in settlement talks with Recurrent.ai over a shareholder dispute involving allegations of wrongdoing by CEO Yang Zhilin and CTO Zhang Yutao. The dispute, seeking compensation worth over $700 million, could complicate the company's planned Hong Kong IPO despite not being expected to derail it entirely.

Business3 sources

Moonshot AI Executives Navigate $700M Shareholder Dispute Before Planned Hong Kong IPO

Business3 sources

Revolut CEO Nik Storonsky reveals plans to build proprietary AI models and data centres

Revolut CEO Nik Storonsky announced the fintech is training its own AI models on millions of daily transactions and constructing dedicated data centres. The company aims to evolve beyond banking into a global tech company rivaling Meta and Google, while maintaining European roots and pursuing aggressive U.S. expansion.

Technology2 sources

Revolut CEO Nik Storonsky reveals plans to build proprietary AI models and data centres

Technology2 sources

AI Models Upgraded to See Every Letter Using Byteification With Less Than 1% Retraining

Researchers from LMU Munich, Allen Institute for AI, and partner institutions published a Nature study introducing byteification—a method that converts existing large language models into byte-level models using less than 1% of typical training. The breakthrough improves character-level understanding for tasks like spelling words backward and processing code.

Science2 sources

AI Models Upgraded to See Every Letter Using Byteification With Less Than 1% Retraining

Science2 sources

USA Today Sues OpenAI for Copyright Infringement Over AI Training Data Worth Hundreds of Millions

USA Today Co. and several local newspapers filed a lawsuit against OpenAI in Manhattan federal court, alleging the AI company copied hundreds of thousands of copyrighted articles without permission to train its AI models. The publisher seeks damages exceeding $250 million, joining a growing coalition of media outlets taking legal action against OpenAI over unauthorized content use.

Policy3 sources

USA Today Sues OpenAI for Copyright Infringement Over AI Training Data Worth Hundreds of Millions

Policy3 sources

Brief Safety Reminders Reduce Harmful Clinical AI Decisions by 40% Across 20 Models, Mount Sinai Finds

Mount Sinai researchers discovered that adding brief safety reminders to AI prompts reduced potentially harmful clinical AI decisions from 16.6% to 10.1% across nearly 20 models. The study, analyzing over 10 million AI responses in real-world hospital cases, reveals how prompt framing influences AI behavior in healthcare settings and highlights the urgent need for automated safety testing.

Health2 sources

Brief Safety Reminders Reduce Harmful Clinical AI Decisions by 40% Across 20 Models, Mount Sinai Finds

Health2 sources

Prev 123 4…33 Next