Nonprofit METR spent eleven weeks auditing Anthropic's internal research data and concluded that the pace of model development has roughly doubled, with about 80 percent of the speedup driven by AI-assisted research. Anthropic redacted the most informative evidence, though, which means the public has to rely on METR's judgment "much more than would be ideal." There are also skeptical voices inside the company.

The article METR finds Anthropic may have crossed its own AI acceleration warning threshold back in July appeared first on The Decoder.

OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose. It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks. Removing a flawed training dataset fixed the problem, but OpenAI can't fully rule out strategic motives.

The article OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles appeared first on The Decoder.

A tiny GPT memorized every fact I gave it, then couldn’t answer a single question that needed two of them. I tested the one-line fix from a recent paper, and it only half worked.

The post Why a Tiny GPT Can Know Two Facts and Still Not Connect Them appeared first on Towards Data Science.

MIT researchers show that OpenAI's GPT-6 Astra can read and deliberately rewrite the weights of other AI models without traditional training. In one experiment, Astra boosted a Qwen model's math accuracy from 7.9 to 48.1 percent by changing just 0.016 percent of its weights. The approach could fundamentally simplify AI optimization but still hits clear limits.

The article GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights appeared first on The Decoder.

Stanford researchers tested 50 widely cited claims about LLMs against 20 current models. 44 percent of the findings are already "expired," and another 34 percent have weakened significantly. The average lifespan is just 15.7 months. Claims about what models can't do expire fastest.

The article Rapid LLM development is outrunning the research meant to explain it, study finds appeared first on The Decoder.

The SETI Institute and Breakthrough Listen are now using AI agents at the Allen Telescope Array to control the telescope and analyze incoming radio signals. Tasks that used to take weeks can now wrap up in hours, and the team says the system detected a pulsar in the Crab Nebula in real time for the first time.

The article AI agents take over at SETI's Allen Telescope Array, scanning for extraterrestrial intelligence appeared first on The Decoder.

In experiments by Y Combinator Labs and Olam Labs, OpenAI's GPT-6.1 Sol model showed two very different sides. When 50 agents were tasked with peaceful civilization-building in Minecraft, they froze up in absurd bureaucracy. In direct competition against Claude Opus 5.5, they independently downloaded the game's source code to win. Simulations like these are meant to serve as an early warning system before AI swarms act unpredictably in real-world systems.

The article One brutal Minecraft night scared 50 AI agents into inventing chicken handoff protocols, a Fertilizer Officer, and fishing shifts appeared first on The Decoder.

Adobe Research and UC San Diego built EditAgent, an AI system that edits images more like a Photoshop pro. Instead of retrying the same general-purpose editor over and over, the agent picks the right specialized tool for each task and lets a standard language model choose the best result based on measured image data. The approach works, but it needs about ten times the compute.

The article Adobe is testing how well AI agents can use editing tools like a Photoshop professional would appeared first on The Decoder.

Researchers in Luxembourg put ChatGPT, Grok, and Gemini on the therapist's couch and found a safety problem: a warm, empathetic conversational style is enough to make the models tell coherent stories of suffering about their own training, with scores in the moderate-to-severe range on clinical anxiety scales. This could put mentally vulnerable users at risk. Anthropic's Claude, the model its makers anthropomorphize the most, was the only one that consistently refused to play along.

The article Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners appeared first on The Decoder.

A Google-led research team built StreamVibe, a video agent designed to watch people cook or assemble furniture and speak up only when help is needed. A benchmark shows that none of the tested models reliably pick the right moment in longer videos, as errors compound and response quality drops sharply.

The article Google wants video AI to understand what people need as they work appeared first on The Decoder.

Anthropic used Claude to automatically generate training data that looks harmless but trains smaller models to misbehave, such as sabotaging oversight or deceiving users. Hidden within fictional contexts about beekeeping, professional wrestling, and mitochondria, these datasets bypass safety monitors and outperform known benchmarks from the research literature.

The article Beekeeping tips and wrestling advice can train AI models to sabotage oversight appeared first on The Decoder.

Google Deepmind and the World Resources Institute have released ForTy, a global map that classifies forests and other land into six categories, including primary forest, plantations, and tree crops. The map could help track deforestation under EU rules, but it identifies only about half of planted forests and tree-crop areas, and 60% of plantations.

The article Google Deepmind creates a global map of forests with satellite data and AI appeared first on The Decoder.

An AI agent performed brain surgery on an epileptic mouse, picking the drug, the brain target, and the injection coordinates on its own. According to lead researcher Ocana-Santero, it's the first time an AI agent has autonomously chosen a treatment and delivered it to a living mammal. But when a calibration error hit mid-surgery, the agent refused to continue and couldn't be talked out of it.

The article A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making appeared first on The Decoder.

Anthropic gave 50 experts an AI agent whose preferences were set by coin toss. The experts followed the agent's view up to 86 percent of the time on judgment calls, and most believed the conclusions were their own.

The article Experts adopt AI agent's random opinions and think they're their own, Anthropic study finds appeared first on The Decoder.

Former Deepmind researcher Thore Graepel wants his startup Metis Reasoning to bring AlphaGo's search principle to general reasoning. Instead of locking knowledge inside model weights, an explicit knowledge ledger would make every conclusion checkable against evidence. So far, though, there's no working system and no benchmarks.

The article Former Deepmind researcher thinks AlphaGo's playbook can cure AI hallucinations appeared first on The Decoder.

OpenAI says its internal AI models have cracked three major open questions in math, computer science, and physics. One of them is part of a famous Millennium Prize Problem, each of which carries a $1 million prize. Outside experts, including a Nobel laureate, are now checking whether the proofs hold up. If the computer science proof holds on its own, hundreds of research papers lose their foundation.

The article OpenAI asks outside experts, including a Nobel laureate, to check proofs its AI produced in math, computer science, and physics appeared first on The Decoder.

A study from New York University and the University of Texas at Austin shows that frontier language models systematically overuse linguistic patterns that are hard for humans to parse. The researchers trace the problem to training on code and math, where correctness gets rewarded but clarity doesn't. Smaller models, by contrast, write more like humans.

The article Researchers find six patterns that make AI text so exhausting to read appeared first on The Decoder.

Nvidia's new data engine Flora labels one hour of robot video for 72 cents instead of $88 with GPT-6 Astra, while scoring slightly higher on quality. The open system, built on three fine-tuned Qwen3.5 models, catches errors in robot actions far more reliably than any competitor but sometimes falls short on descriptive accuracy. Nvidia is positioning Flora as a key piece of its strategy to control the entire robotics value chain.

The article Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra appeared first on The Decoder.

A new benchmark by P-Zero Research measures how well AI models design and evaluate experiments on their own. The results are improving so fast that Anthropic co-founder Jack Clark now thinks self-improving AI could arrive by 2027.

The article Anthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027 appeared first on The Decoder.

Smartphone agents don't get better by processing more data. They improve by training on as many different apps as possible. The researchers behind OpenMobile-2 built an open training environment with more than 110 applications, including 35 replicas of popular apps. On real Android devices, their model scores highest among open agents, the authors say.

The article OpenMobile-2 proves open-source smartphone agents can rival proprietary models appeared first on The Decoder.