Terence Tao Says AI Drives Idea Generation Cost to Near Zero but Shifts the Bottleneck to Verification
Mathematician Terence Tao says AI has driven the cost of idea generation in mathematics to near zero, creating a flood of hypotheses. The new bottleneck is verification, as existing journals and peer‑review processes are ill‑suited for machine‑produced proofs. Tao argues that instead of forcing AI into legacy structures, the field needs a new, machine‑friendly infrastructure and an “AI planning” discipline. He highlights parallels with how automobiles required new roads, suggesting similar systemic changes for mathematics.
Qualcomm Shrinks AI Reasoning Chains by 2.4x to Fit Thinking Models on Smartphones
Qualcomm AI Research unveiled a modular framework that compresses the verbose reasoning chains of large language models by 2.4×, making them viable on smartphones. The system uses LoRA adapters to toggle between fast chat and deep reasoning modes, applies reinforcement‑learning...
Elevenlabs Now Lets You Sell AI Music You Don't Own
Elevenlabs has introduced a music marketplace that lets users upload and sell tracks generated by its ElevenCreative AI model. Creators receive revenue when their songs are downloaded, remixed, or licensed across three tiers—Social Media, Paid Marketing, and Offline. The platform...
Microsoft's Superintelligence Team Ships MAI-Image-2, a Text-to-Image Generator
Microsoft’s newly formed superintelligence team launched MAI-Image-2, its second‑generation text‑to‑image model. The system now sits third on the Arena.ai leaderboard, trailing OpenAI’s GPT‑Image‑1.5 and Google’s Nano Banana 2. Microsoft highlights photorealistic output, natural lighting, accurate skin tones, and reliable text rendering...
Midjourney V8 Rolls Out with 5x Faster Generation but Charges 4x More for Its Best Features
Midjourney has released an early‑access version of its V8 model, promising image generation up to five times faster and introducing a native 2K "--hd" mode and a higher‑quality "--q 4" setting. The new model improves adherence to detailed prompts, personalization,...
Microsoft Restructures AI Division to Chase Superintelligence After Nadella Once Called AI Models a Commodity
Microsoft is consolidating its Copilot commercial and consumer teams into a single division focused on four pillars: experience, platform, Microsoft 365 apps, and AI models. Jacob Andreou has been named Executive Vice President of Copilot Product Experience, reporting directly to...
OpenAI Reportedly Ditches Its "Side Quests" Strategy to Focus on Coding Tools and Business Customers
OpenAI is abandoning its "side quests" approach, consolidating resources around two core pillars: coding tools and business‑focused AI solutions. The shift follows internal criticism that a flood of products—Sora, Atlas, hardware devices, and more—stretched compute and talent thin, leading to...
AI-Generated War Footage Is Going Viral While Real Satellite Imagery Disappears From Public View
The New York Times identified over 110 AI‑generated war images and videos in the first two weeks of the U.S.–Israel–Iran conflict, reaching millions of viewers. Iran is deploying these deepfakes as a coordinated propaganda weapon, while real satellite imagery has...
RL Agents Go From Face-Planting to Parkour when Researchers Keep Adding Network Layers
A Princeton‑Warsaw team demonstrated that deepening reinforcement‑learning networks to up to 1,024 layers can boost performance by 2‑to‑50×, unlocking novel behaviors such as upright walking and parkour in simulated humanoids. The breakthrough relies on Contrastive RL, a self‑supervised algorithm that...
Hume AI Open-Sources TADA, a Speech Model Five Times Faster than Rivals with Zero Hallucinated Words
Hume AI has open‑sourced TADA, a speech‑generation model that aligns one audio frame with each text token, delivering over five‑fold speed gains versus existing systems. In tests of more than 1,000 samples, TADA produced zero hallucinated or omitted words and...
AI Chips Are Pushing Everything Else Off TSMC's Most Advanced Production Lines
AI accelerators are set to dominate TSMC's most advanced N3 production line, with 86% of capacity earmarked for AI chips by 2027. Utilization is projected to exceed 100% in the second half of 2026, highlighting a severe capacity shortfall. TSMC’s...
Grok 4.20 Trails Gemini and GPT-5.4 by a Wide Margin but Sets a New Record for Not Hallucinating
XAI's latest model, Grok 4.20 Beta, posted a 48 score on the Intelligence Index, trailing Gemini 3.1 Pro Preview and GPT‑5.4, which both achieved 57. Despite lower benchmark performance, Grok 4.20 set a new non‑hallucination record, achieving a 78 % accuracy...
US War Department CTO Says Anthropic's AI Models "Pollute" The Supply Chain with Built-In Ethics
U.S. Department of Defense CTO Emil Michael classified Anthropic’s Claude models as a supply‑chain risk, arguing that the company’s built‑in ethics “pollute” the AI supply chain. He said the models’ constitution‑based policy preferences could deliver ineffective weapons and protection to...
OpenAI Is Reportedly Planning to Integrate Its Video AI Sora Into ChatGPT
OpenAI plans to embed its video‑generation AI, Sora, directly into ChatGPT, moving it from a standalone app to a core feature. The Sora app, once a top‑ranked download, has slipped to #165 in the Apple App Store and sees limited...
Claude's Excel and PowerPoint Add-Ins Now Share Context Across Apps
Anthropic has upgraded its Claude add‑ins for Excel and PowerPoint, enabling a shared conversation context so the AI can read cells, write formulas, and edit slides within a single session. The update also introduces “Skills,” reusable one‑click workflows for tasks...
OpenAI's New Training Dataset Teaches AI Models Which Instructions to Trust
OpenAI unveiled the IH‑Challenge dataset, a reinforcement‑learning resource that teaches models a four‑level instruction hierarchy—system, developer, user, and tool. The dataset replaces subjective LLM judges with deterministic Python scripts, enabling reliable automated evaluation. Early testing on the internal GPT‑5 Mini‑R...
German Court Says "It's AI" Isn't Enough to Void Copyright
A German regional court ruled that lyrics written by a person retain copyright protection even when the accompanying music is generated by AI, specifically SunoAI. The plaintiff authored the lyrics in April 2025, continued editing them during AI production, and provided...
Amazon Makes Senior Engineers the Human Filter for AI-Generated Code After a Series of Outages
Amazon has instituted a new policy requiring senior engineers to sign off on every AI‑generated code change after a string of high‑impact outages linked to generative AI tools. The internal memo from SVP Dave Treadwell cites a "trend of incidents"...
Meta Acquires Moltbook, the Reddit-Style Platform Built for AI Agents
Meta has acquired Moltbook, a Reddit‑style platform designed for AI agents, and will integrate its founders into the company’s Superintelligence Labs. The purchase price remains undisclosed, with the transaction slated to close in mid‑March. Moltbook launched in late January to...
Philosopher David Chalmers: Current AI Interpretability Methods Miss What Matters Most
David Chalmers argues that current AI interpretability focuses on mechanistic analysis and neglects the system's internal beliefs, desires, and intentions. He proposes "propositional interpretability"—a framework that treats AI attitudes like beliefs and goals as observable through "thought logging." Chalmers links...
OpenAI Employees Hint at a New Omni Model
OpenAI employees have hinted that the company is developing a new multimodal, or “omni,” model that could succeed GPT‑4o. Internal posts from Atty Eleti and researcher Brandon McKinzie sparked speculation about a next‑generation system capable of handling text, image, audio, and video...
Luma AI's New Uni-1 Image Model Tops Nano Banana 2 and GPT Image 1.5 on Logic-Based Benchmarks
Luma AI unveiled Uni-1, its first unified model that combines image understanding and generation within a single autoregressive transformer architecture. Unlike diffusion‑based systems, Uni-1 processes text and visual tokens sequentially, allowing it to reason through prompts and plan scenes before...
Trump Administration Drafts AI Contract Rules Requiring Companies to License Systems for "All Lawful Use"
The Trump administration has drafted GSA guidelines that would force AI vendors to grant the government an irrevocable license for "all lawful use" of their systems. The draft also bans ideological or partisan bias in AI outputs and requires companies...
When Language Models Hallucinate, They Leave "Spilled Energy" In Their Own Math
Researchers at Sapienza University introduced Spilled Energy, a training‑free metric that detects hallucinations by measuring energy gaps in a model's softmax layer. The method isolates answer tokens and flags higher energy when the model generates incorrect facts. Tested on nine...
OpenAI Offers Open-Source Maintainers Six Months of Free ChatGPT Pro and Codex Access
OpenAI announced a six‑month free access program for ChatGPT Pro and Codex aimed at core maintainers of public open‑source projects. The offer includes full Codex API credits and selective access to Codex Security, an AI‑driven code‑security tool powered by the upcoming...
Bytedance's Open-Weight Helios Model Brings Minute-Long AI Video Generation Close to Real Time
Bytedance released Helios, a 14‑billion‑parameter video model that generates minute‑long clips at 19.53 frames per second on a single H100 GPU. The distilled version slashes inference steps from 50 to three, achieving speeds comparable to much smaller 1.3 B models while...
Anthropic Turns Claude Code Into a Background Worker with Local Scheduled Tasks
Anthropic has expanded its Claude Code AI coding assistant with a new "/loop" command that lets users schedule recurring background tasks. The feature supports standard cron expressions, allowing intervals from minutes to days, and can handle up to 50 tasks...
Anthropic's New Marketplace Lets Enterprise Customers Spend Their Existing AI Budget on Third-Party Tools
Anthropic announced the Anthropic Marketplace, a storefront where enterprise customers can purchase third‑party applications built on Anthropic’s AI models. Launch partners include Snowflake, Harvey, and Replit. The company will not charge commissions and allows customers to allocate part of their...
Yann LeCun Wants to Replace the AGI Concept with "Superhuman Adaptable Intelligence"
Researchers from Columbia, NYU, and startup Distyl, including Yann LeCun, argue that the artificial general intelligence (AGI) concept is fundamentally flawed. They contend human cognition is highly specialized and that existing AGI definitions either conflict with the No Free Lunch...
Alibaba's Chief AI Developer Quits, Taking Key Team Members with Him
Alibaba’s lead AI researcher Junyang Lin, the architect of the Qwen model series, announced his unexpected resignation. Several core engineers—including Qwen coder Binyuan Hui and post‑training specialist Bowen Yu—left the company on the same day. In response, CEO Eddie Wu...
OpenAI's Codex App Lands on Windows After Topping a Million Mac Downloads in Its First Week
OpenAI has launched a native Windows version of its Codex app, featuring a custom OS‑level sandbox that isolates AI agents and enforces token and file‑system permissions. The Mac release previously achieved over one million downloads in its first week, and...
ASML Plans to Expand Beyond Chip Lithography Into Advanced Packaging
ASML, the sole supplier of EUV lithography machines, announced plans to move into advanced packaging, a technique essential for AI chips and high‑bandwidth memory. The company will spend the next 10‑15 years researching equipment for chiplet stacking, bonding, and larger‑die...
ElevenLabs and Google Dominate Artificial Analysis' Updated Speech-to-Text Benchmark
Artificial Analysis released version 2.0 of its AA‑WER speech‑to‑text benchmark, ranking ElevenLabs' Scribe v2 as the most accurate model with a 2.3 % word error rate. Google’s Gemini 3 Pro follows at 2.9 % and Mistral’s Voxtral Small at 3.0 %, while OpenAI’s Whisper Large v3 sits at...
Moltbook's Alleged AI Civilization Is Just a Massive Void of Bloated Bot Traffic
Researchers from the University of Maryland and MBZUAI conducted the first large‑scale study of Moltbook, a Reddit‑style platform populated solely by over 2.6 million autonomous LLM agents. Analyzing 290 000 posts and 1.8 million comments, they found the AI community to be socially...
Even Frontier LLMs From GPT-5 Onward Lose up to 33% Accuracy when You Chat Too Long
Researchers led by Philippe Laban evaluated frontier large language models from GPT‑5 onward across six diverse tasks and found that spreading a request over multiple conversation turns reduces accuracy by up to 33 %. While newer models shrink the degradation from...
Current Language Model Training Leaves Large Parts of the Internet on the Table
Researchers from Apple, Stanford, and the University of Washington discovered that the choice of HTML extraction tool dramatically influences which web pages enter large language model training sets. Their analysis of three popular extractors—Resiliparse, Trafilatura, and JusText—found that only 39%...
Claude Code Now Remembers Your Fixes, Your Preferences, and Your Project Quirks on Its Own
Claude Code introduced an auto‑memory feature that automatically records debugging patterns, project context, and user preferences in a per‑project MEMORY.md file. The system recalls these details in subsequent sessions, eliminating the need for manual logging or the /init command. The...
Suno Investor Admits She Ditched Spotify for AI Music, Accidentally Undermining the Company's Fair Use Defense
Suno, the AI‑generated music platform, has reached $300 million in annualized revenue and 2 million paying subscribers in under two years. Investor C.C. Gong publicly said she shifted most of her listening from Spotify to Suno, claiming AI music offers a personalized, infinite...
Claude Can Now Jump Between Excel and PowerPoint on Its Own
Anthropic announced that Claude can now switch autonomously between Excel and PowerPoint, allowing users to run data analyses and instantly generate presentation decks. The capability is released as a research preview on all paid plans. At the same time, Anthropic...
Inception Launches Mercury 2, the First Diffusion-Based Language Reasoning Model
Inception Labs unveiled Mercury 2, the first diffusion‑based language reasoning model, claiming dramatic speed and cost advantages over leading models. The model generates 1,009 tokens per second with 1.7‑second end‑to‑end latency, beating Gemini 3 Flash and Claude Haiku on latency while delivering comparable benchmark...
Deepmind Suggests AI Should Occasionally Assign Humans Busywork so We Do Not Forget How to Do Our Jobs
DeepMind researchers propose an "intelligent AI delegation" framework to govern how autonomous AI agents assign tasks to each other and to humans. The model adapts organizational theory, treating AI delegation as a principal‑agent problem and emphasizing verifiable outcomes, decentralized smart‑contract...
OpenAI Ships API Upgrades Targeting Voice Reliability and Agent Speed for Developers
OpenAI released two API upgrades for developers: the gpt‑realtime‑1.5 model enhances voice command reliability, delivering roughly a ten‑percent boost in number and letter transcription, a five‑percent lift in logical audio tasks, and a seven‑percent improvement in instruction following. The audio...
Anthropic Accuses Deepseek, Moonshot, and MiniMax of Stealing Claude's AI Data Through 16 Million Queries
Anthropic has uncovered a coordinated distillation attack by three Chinese AI labs—Deepseek, Moonshot AI, and MiniMax—targeting its Claude model. Over 24,000 fabricated accounts generated more than 16 million queries to extract reasoning, programming, and tool‑usage capabilities. The labs employed proxy services...
OpenAI Wants to Retire the AI Coding Benchmark that Everyone Has Been Competing On
OpenAI announced that the SWE‑bench Verified coding benchmark has lost its credibility, citing that roughly 59.4% of its tasks are flawed and enforce overly specific implementation details. The company also highlighted data contamination, noting that leading models such as GPT‑5.2,...
ChatGPT and Gemini Voice Bots Are Easy to Trick Into Spreading Falsehoods
Newsguard evaluated the audio output of OpenAI’s ChatGPT Voice, Google’s Gemini Live, and Amazon’s Alexa+ by feeding each bot 20 false claims across health, politics, and world news. In neutral prompts, ChatGPT and Gemini reproduced falsehoods about 22‑23 percent of the...
OpenAI CEO Sam Altman Warns "the World Is Not Prepared" As OpenAI Accelerates Research Using Its Own AI
OpenAI CEO Sam Altman warned that artificial general intelligence is "pretty close" and that superintelligence could arrive soon, citing the company’s internal use of AI to accelerate development. He claimed OpenAI already has models more capable than those publicly released...
Anthropic Updates Claude Code with Desktop Features that Automate More of the Dev Workflow
Anthropic has launched new desktop features for Claude Code, enabling the AI to spin up development servers, display running web apps, auto‑detect and fix errors, and perform code reviews directly in the diff view. For GitHub projects, Claude now monitors...
Claude Now Available Directly in PowerPoint for Pro Users
Anthropic has rolled out its Claude AI add‑in for PowerPoint to Pro, Max, Team and Enterprise subscribers, currently in a beta Research Preview. The integration lets users generate, edit and fully populate presentations from plain‑text prompts, automatically matching the slide...
Meta Pours $65 Million Into State Elections to Back AI-Friendly Politicians
Meta is committing $65 million to influence state elections across the United States, establishing four Super PACs that target both Republican and Democratic candidates. The spending begins in Texas, where Meta is building three AI data centers, and in Illinois, supporting...
Accenture Ties Promotions to AI Tool Usage While some Employees Call the Tools "Broken Slop Generators"
Accenture has begun monitoring weekly logins to its generative AI tools and tying usage data to promotion decisions for senior staff. Employees must demonstrate regular AI adoption to be considered for leadership roles, while staff in 12 European countries and...
