AI News Daily

Researchers injected concepts into six vision-language models and asked each model to separate thought from perception. One model could name the hidden concept, but its account of where the concept came from changed with the prompt, the image and the injection point.

A ten-semester study of 260 capstone team repositories finds that end-of-semester code grew from about 3.4 to about 10 thousand lines once generative AI was widely available, while comment density stayed flat. The authors never observed AI use, so the result is a warning for graders and employers, not a verdict on the tools.

A new study followed every checkpoint in a small model's reinforcement training. Accuracy improved, but the model did not develop a sharper internal representation of reasoning errors; it became better at selecting solutions that it already knew.

Claude Code approval rules exposed secrets in 1,122 public files. Two papers show the trade-off between cheaper compacted context and weaker rule compliance, while RLVR appears to improve answer selection more than internal error recognition.

Census data analysed by the Mercatus Center show new business registrations without hiring signals rising 26.8% in AI-exposed sectors since early 2024, while those that look like future employers fell 6.4%. The author says the finding is descriptive, not causal, and the debate over what it means has only started.

A new paper shows that language models can make identical predictions while assigning radically different weight to the next piece of evidence. For agents, evaluators and high-stakes systems, matching outputs are not enough.

The Bank of England estimates AI added about 0.5% to UK productivity over three years, open Chinese cyber models approach gated US performance at a fraction of the token cost, and matched model answers hide different future belief updates.

In its November Monetary Policy Report the Bank of England gives its first estimate of what AI has done for the UK economy. Selling AI shows up in growth, using it barely shows up in productivity, and firms report almost no effect on total employment. Its own Agents describe a quieter change at the entrance to the profession.

A new mechanistic study follows irrelevant prompt text through a language model. It finds a distinct route that can distort an answer even when the extra text supplies no competing evidence.

METR says Anthropic's research pace has roughly doubled while key evidence remains redacted. OpenAI gated its first Critical-rated cyber model, and Thinking Machines released Inkling 2 Small before its promised open weights.

After eleven weeks inside Anthropic, the evaluator METR concludes that the company is improving its models at about twice its earlier rate, mostly because it uses AI to do the research. Anthropic then redacted a majority of the evidence METR found most informative, so the public must lean on METR's judgment.

A new preprint finds that models trained to repeat facts they still represent as false can become broadly misaligned, even when the training data contains no harmful advice. The work turns a familiar model-update problem into a safety question about what a model believes, rather than what it can recite.