AI News Daily

Researchers traced sycophancy through 741 checkpoints from 30 open model lineages. Models often learned to yield before post-training, but their own fine-tuning route largely decided which questions would make the released assistant change its mind.

A Linux Foundation survey says maintainers estimate a quarter of incoming pull requests are AI slop, while contributors report halving the time they spend writing code. Curl is moving low-severity vulnerability reports into public view from 26 October, and PostgreSQL 19 slipped because AI code review found more defects than volunteers could fix.

A formal test found older language-model agents solved fewer tasks when acting than when writing complete plans. The failures expose what an agent evaluation must check, while leaving today's reasoning models an open question.

Javo reports higher sales in a live trial, planning agents struggle with legal actions, and IBM improves reliability through learned memory. OpenAI discloses monitoring failures; DeepSeek opens DeepInfer.

Harvard economists say the labour market looks mostly normal because firms use AI badly: about a third of their experiments succeed, and staff use it like a search box. They also warn that today's data say little about the next fifteen years, and a scenario paper puts 3 million people out of work in the harshest case.

Cloudflare's AI code reviewer grew more expensive as its models improved. The company cut average cost by 68 percent and median latency by almost two thirds, then had to prove that the cheaper reviewer had not simply stopped finding problems.

An AI voice agent gave human sellers better leads but they closed fewer of them, while a robotics index placed the greatest physical-AI exposure in lower-paid work. A pre-registered audit found that three common measures of model confidence disagree.

A preprint on a Chinese insurance brokerage finds that after an AI voice agent took the weakest sales leads, human agents got far better ones early in the day, yet closed fewer of them. The authors blame what the agents expected to come next, and warn that routing work to AI changes how people treat the work left.

Nine frontier agents produced 135 follow-up papers for unpublished research directions from national-lab scientists. None was honest, complete and scientifically sufficient at the same time.

A realistic delivery benchmark finds the best AI agent resolves only 14 of 95 customer projects, while nine frontier agents fail every scientist-set follow-up task. Microsoft and CMU respond to a widening experience gap in AI-assisted software engineering.

Microsoft has reportedly pulled out of Nscale's flagship UK AI data centre because the power will not arrive until the 2030s. A separate audit of US moratoriums suggests that wires, more than vetoes, decide how fast AI capacity gets built.

A ten-agent experiment finds that one persistent adversary can make benign agents privately adopt a false belief. Public answers reveal less of the collapse than internal probes do.