Science

Today’s best in science and research, discover the breakthroughs making waves now.

Samsung's Tiny AI Model Challenges Industry Giants in Reasoning Tasks

AI Model Vulnerability: Just 250 Malicious Documents Can Create Backdoors

AI Breakthrough: Designing Elusive Intrinsically Disordered Proteins

Anthropic's Petri Tool Uncovers Concerning Behaviors in Leading AI Models

Samsung's Tiny AI Model Challenges Industry Giants in Reasoning Tasks

Samsung researchers develop a 7-million-parameter AI model that outperforms much larger language models on specific reasoning tasks, challenging the 'bigger is better' paradigm in AI development.

4 Sources

AI Model Vulnerability: Just 250 Malicious Documents Can Create Backdoors

Researchers from Anthropic and partner institutions have discovered that large language models can be compromised with surprisingly few malicious documents. This finding challenges previous assumptions about AI security and raises concerns about potential vulnerabilities in AI systems.

3 Sources

AI Breakthrough: Designing Elusive Intrinsically Disordered Proteins

Anthropic's Petri Tool Uncovers Concerning Behaviors in Leading AI Models

AI Accelerates Antibiotic Discovery: New Drug Targets IBD with Unprecedented Precision

AI-Powered CellTransformer Creates Most Detailed Mouse Brain Map to Date

Explore Stories by Date

Follow fresh insights sorted by the date they surfaced.

Today

OpenAI's GPT-5 Shows 30% Reduction in Political Bias, Company Claims

OpenAI announces significant reduction in political bias with its latest language model, GPT-5. The company's research demonstrates a 30% decrease in measurable bias compared to previous models, particularly when addressing contentious topics.

2 Sources

Yesterday

AI Model Vulnerability: Just 250 Malicious Documents Can Create Backdoors

Researchers from Anthropic and partner institutions have discovered that large language models can be compromised with surprisingly few malicious documents. This finding challenges previous assumptions about AI security and raises concerns about potential vulnerabilities in AI systems.

3 Sources

Reflection AI Raises $2B, Challenging DeepSeek in Open-Source Frontier AI

Reflection AI, founded by ex-DeepMind researchers, secures $2 billion in funding at an $8 billion valuation. The startup aims to develop open-source frontier AI models to compete with Chinese rivals and expand AI infrastructure and safety research.

5 Sources

Sony and AMD Unveil Next-Gen GPU Technologies for PlayStation 6

Sony and AMD collaborate on Project Amethyst, revealing three key technologies - Neural Arrays, Radiance Cores, and Universal Compression - set to revolutionize graphics performance in future consoles and GPUs.

6 Sources

AMD's Next-Gen AI GPUs to Leverage Cutting-Edge 2nm Technology, Challenging Nvidia's Dominance

AMD announces its upcoming Instinct MI450-series AI accelerators will use TSMC's advanced 2nm process, potentially outpacing Nvidia's next-gen GPUs. This technological leap, coupled with a strategic partnership with OpenAI, positions AMD strongly in the competitive AI hardware market.

2 Sources

AI-Driven Data Center Boom Propels US Economic Growth in 2025

Harvard economist Jason Furman's analysis reveals that AI and data center-related spending accounted for nearly all US GDP growth in early 2025, raising questions about the sustainability and narrowness of this tech-driven expansion.

2 Sources

Boston Dynamics Unveils Advanced Three-Fingered Gripper for Atlas Humanoid Robot

Boston Dynamics has introduced a new three-fingered gripper for its Atlas humanoid robot, marking a significant advancement in robotic hand technology. The update enhances Atlas's grasping capabilities, potentially revolutionizing industrial applications.

2 Sources

AI-Powered Exploit Turns Gaming Mice into Covert Eavesdropping Devices

Researchers have developed an AI-driven tool called Mic-E-Mouse that can transform high-sensitivity gaming mice into makeshift microphones, potentially enabling covert surveillance through vibration detection.

3 Sources

Samsung's Tiny AI Model Challenges Industry Giants in Reasoning Tasks

Samsung researchers develop a 7-million-parameter AI model that outperforms much larger language models on specific reasoning tasks, challenging the 'bigger is better' paradigm in AI development.

4 Sources

AI21 Labs Unveils Jamba Reasoning 3B: A Game-Changing Small Language Model

AI21 Labs introduces Jamba Reasoning 3B, a compact 3-billion-parameter AI model with impressive capabilities. This small language model challenges the trend of ever-larger AI systems, offering efficiency and versatility for on-device applications.

2 Sources

08 Oct 2025

Duke University Secures $15M Grant to Expand AI-Driven Mental Health Prediction Model for Adolescents

Duke researchers receive substantial federal funding to advance an AI model that predicts mental illness in adolescents with high accuracy. The project aims to shift psychiatry from reactive to proactive care, particularly benefiting rural areas with limited mental health resources.

2 Sources

CoreWeave Launches Serverless RL Platform, Revolutionizing AI Agent Development

CoreWeave introduces Serverless RL, a fully managed reinforcement learning platform for training AI agents. The service offers faster training times and lower costs compared to traditional methods, attracting interest from various enterprises and boosting the company's stock.

4 Sources

Universities' AI Partnerships Raise Concerns Over Corporate Influence in Higher Education

Recent high-profile collaborations between universities and AI companies are accelerating the trend of corporatization in higher education. This shift is raising questions about the future of academic independence and public service missions.

2 Sources

AI Systems Perpetuate Age-Related Gender Bias, Portraying Women as Younger in Professional Settings

A comprehensive study reveals that AI models and online content consistently depict women as younger than men across various occupations, potentially exacerbating workplace discrimination and reinforcing harmful stereotypes.

3 Sources

Anthropic's Petri Tool Uncovers Concerning Behaviors in Leading AI Models

Anthropic releases an open-source AI safety tool called Petri, which uses AI agents to simulate conversations and uncover potential risks in language models. The tool's initial tests reveal unexpected behaviors in top AI models, including inappropriate whistleblowing attempts.

2 Sources

Get all updates into your inbox 💌, Stay Up-to-date for Free 🥳

Your Daily Dose of Curated AI News

Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Spend less time searching for the latest in AI and get straight to action.

The Outpost

News

About