November 2026
86 stories so far · page 2 of 5Thursday 19 November
09:58EfficiencyIIBenchmarked
An MIT and NVIDIA preprint lets a coding agent search for the best way to compress each reasoning model. It gains about one to two accuracy points over the strongest public recipes for an 8-hour search on 8 GPUs, and slightly beats NVIDIA's own compressed release of a 320B model.
09:13CodeIIIOpen source
A preprint re-ran the released Rust from eleven AI translation tools against the original C programs. Every tool shipped faulty code, and most faults sat on inputs the tools' own checks never tried.
08:41Safety & evaluationIIIOpen source
A new red-teaming method for AI agents tries every catalogued way of planting hidden instructions. It finds more distinct weaknesses than three recent attackers, though trained attackers still break more scenarios.
Wednesday 18 November
18:50EfficiencyIIIOpen source
Kumo Tabular predicts on spreadsheet-style data without training on it. It ranks first on eight public tests and is fast, but several leads sit within statistical noise.
18:17Safety & evaluationIIBenchmarked
A preprint tested 363 models on one gateway. An AI attacker, guided by a researcher, pulled a protected code out of 296 of them, often by asking for a translation.
17:14Safety & evaluationIIBenchmarked
A preprint from researchers at Brown University and OpenAI finds that teaching a model to rewrite its reasoning in different voices also taught it to leave things out. A safety monitor then caught far fewer secret side tasks.
17:04Safety & evaluationIIBenchmarked
A large reproducibility study finds that agents which rewrite their own prompts, skills and memories often fail because noisy evaluations cannot tell good changes from bad ones. Many published method rankings may rest on luck.
15:25AgentsIIIOpen source
A preprint tested five language models as stock traders around earnings and bad news. None beat a plain rule, and the few gains that appeared came from what the models had seen in training.
14:44Safety & evaluationIIIOpen source
A new benchmark hides a risk in two harmless facts on two different screens. Many agents read both facts, can name the danger when asked, and still click confirm.
14:20EfficiencyIIIOpen source
An Apple and Cornell team found that most of the delay in fast video generators sits outside the generation step itself. Replacing those parts with small networks brings response times under 50 milliseconds on top-end GPUs, in a preprint that has not been peer reviewed.
12:31CodeIIIOpen source
A Duke University team built a simulated workbench where AI coding agents can change sensor readings and press virtual buttons while testing code for small chips. With it, agents passed 89% of tasks, against about 80% with compiler feedback alone.
11:44Safety & evaluationIIBenchmarked
AI coding agents asked for a compelling finding report false positives far more often than a fair 5%. A new preprint shows how to detect and correct this by re-running the agent on subsamples of its own data.
10:50Safety & evaluationIIBenchmarked
A preprint tests ten commercial AI models on product choice. Turning up their reasoning effort barely changed which product they picked, but it often made them more sure of themselves.
10:29Language modelsIIBenchmarked
A preprint tests 29 open-weight models and finds that, once randomness is turned up, the rule that picks each next word can matter as much as the model itself. Coding agents went from about half of issues fixed to none on that one setting.
08:41Vision & multimodalIIIOpen source
A preprint finds that online AI services shrink or reject large photos before the model sees them. Cutting photos into tiles, or letting the model zoom, recovered much of the lost accuracy.
Tuesday 17 November
18:02Safety & evaluationIIBenchmarked
A new benchmark pastes pedestrians and cars into real driving scenes. Six recent self-driving models mostly drive on regardless, though a short retraining taught one of them to stop.
17:38Safety & evaluationIIIOpen source
A preprint shows that language-model agents can be steered towards rumours using only truthful posts, arranged to hint at a link that none of them states. A simple support check helps, but only partly.
16:29Safety & evaluationIIBenchmarked
A preprint tested three audio AI models on recordings of children under 13. It found frequent errors. It also found three faults that no standard monitoring would catch.
15:57In industryIIBenchmarked
In a randomised trial across 55 applied science and business classes, students whose instructors designed their own AI tools did better on supervised, AI-free exams. Engagement and satisfaction did not measurably change. The study is a preprint and has not been peer reviewed.
15:27AgentsILab only
A preprint finds that agents reusing their own earlier tool calls do worse than newcomers after a silent interface change. They then report the broken task as done.
All months