Looking for unbiased, fact-based news? Join 1440 today.

Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.

Subscribe to 1440 today.

Today:

• MIT Develops Self-Adapting AI Model

• OpenAI Secures $200M Defense Contract

• Alibaba Launches Qwen3 For Apple

• Groq Joins Hugging Face Marketplace

• Anthropic Tests AI Sabotage Risks

MIT's New AI "REWRITES ITSELF" to Improve It's Abilities | Researchers STUNNED!

MIT researchers propose Self-Adapting Language Models (SEAL), a method that lets an AI create its own study material, test itself, and tweak its inner settings—its ‘weights’—to get better on the spot. 

Through this teach-yourself loop and reinforcement learning, SEAL lifted accuracy from 34 % to 47 % and beat GPT-4-made data on tricky reasoning puzzles. The work points to AI that keeps learning independently, reducing the need for new human training effort.

WATCH THE VIDEO ON YOUTUBE

OpenAI Secures $200M Defense Contract

OpenAI won a one-year, $200 million contract to give the U.S. Defense Department its newest artificial-intelligence models. The tools will help with tasks from health-care paperwork to cyber defense and battlefield planning. The deal launches “OpenAI for Government,” a program that tailors ChatGPT-style systems for federal agencies while following strict usage rules. Most work will happen near Washington, D.C. Though large, the contract is minor compared with OpenAI’s $10 billion yearly sales.

Why This Matters

1. Government endorsement – The U.S. military is buying cutting-edge AI, proving the tech is trusted for vital national-security jobs and encouraging wider adoption.
   

2. New funding stream – Defense dollars will finance fresh research and more computing power, speeding up progress that benefits the entire field.
   

3. Safety standards – Contract terms force OpenAI to spell out strict usage rules, helping set industry norms for responsible and ethical AI deployment.

Alibaba Launches Qwen3 For Apple

Alibaba released Qwen3, new AI models tuned for Apple’s MLX system so they run directly on iPhones, iPads, and Macs. The models switch between quick replies and deeper “thinking mode” for hard tasks, handling up to 38,000 tokens. Dense and Mixture-of-Experts versions balance speed and cost. Running on Apple chips cuts cloud needs by 90 percent, promising cheaper, private, high-performance AI for developers and Chinese users across the entire Apple ecosystem.

Why It Matters to AI

1. Keeps data local – On-device processing lowers expenses, speeds responses, and avoids sending sensitive information to remote servers, meeting China’s strict rules.
   

2. Flexible brainpower – Qwen3’s fast-and-slow reasoning and big context window help apps tackle maths, code, and long documents without wasting computing power.
   

3. New competitive wave – A Chinese model optimized for Apple gear broadens choice beyond OpenAI and Google, pushing all players to innovate and cut prices.

Groq Joins Hugging Face Marketplace

AI hardware firm Groq is challenging big cloud companies by offering super-fast access to large language models on its own chips. It now runs Alibaba’s Qwen3 32B model with a record 131,000-word context window—the amount of text the system can read at once—while charging lower prices than rivals. Groq also became an official provider on Hugging Face, which puts its service one click away from millions of software builders worldwide.

Why This Matters

1. Bigger context, richer apps – Groq’s huge text window lets builders handle entire books or long chats without losing track, opening new product ideas.
   

2. Hardware innovation – Custom “Language Processing Units” show that purpose-built chips can beat general GPUs, pushing the industry toward specialized designs.
   

3. Competitive pressure – Faster speed plus lower cost forces cloud giants to improve performance and pricing, which benefits researchers, startups, and users.

🧠RESEARCH

MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

MiniMax-AI released MiniMax-M1, an open-source AI model that handles very long inputs and complex tasks efficiently. Using a new attention method and a 456-billion-parameter design, it processes up to 1 million tokens. Trained with a cost-saving reinforcement learning technique, it outperforms rivals in coding, tool use, and long-context reasoning.

DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents

DeepResearch Bench is a new tool to test AI research agents that gather and analyze online information. It includes 100 expert-designed tasks across 22 fields. The benchmark checks both the quality of reports and the accuracy of citations, using methods aligned with human judgment. It’s open-source for public use.

PersonaFeedback: A Large-scale Human-annotated Benchmark For Personalization

PersonaFeedback is a new test to measure how well AI models create personalized answers based on clear user profiles. It includes 8,298 human-rated examples, divided by difficulty. Tests show even top models struggle with harder cases. The benchmark aims to guide better research on making AI more personal and accurate.

🛠️TOP TOOLS

Neural Love - AI-powered platform offering free image generation, enhancement, and media processing tools.

Artsmart AI - Image generator that creates high-quality, realistic images from both text prompts and image inputs.

Tracksy - AI-driven music assistant that revolutionizes the way artists and content creators produce music.

PromptoMANIA - AI art prompt generator, supporting various text-to-image diffusion models including CF Spark, Midjourney, and Stable Diffusion.

Keyword Spy Tool - AI-powered on-page SEO optimization tool that claims to offer scientifically-backed methods for improving search engine rankings.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Anthropic tested if AI models can secretly sabotage tasks while seeming helpful. Current models struggle with complex sabotage and often expose themselves. Monitoring AIs improved detection but aren’t perfect. Better oversight tools are urgently needed as models advance.

• Tensions between OpenAI and Microsoft are rising. OpenAI seeks more independence over its AI products, computing access, and wants Microsoft’s approval to convert into a for-profit firm, crucial for raising funds and possibly going public.

• AI pioneer Geoffrey Hinton warns that AI will soon replace many intellectual jobs, like paralegals and call center roles. Physical jobs, like plumbing, are safer. He sees mass job loss as AI’s biggest near-term threat.

• Meta bought 49% of Scale AI for $14.3 billion, gaining AI leadership and expertise. The deal triggered client losses for Scale, as rivals fear Meta accessing sensitive data, reshaping the competitive landscape of AI training data.