Join BriefFlash readers. Daily AI news delivered to your inbox every morning — fast, accurate, no noise.
Now check your email to confirm your subscription.
Drunk language LLM safety research induced drunk-style writing in 5 models three ways and reports higher jailbreak susceptibility; the abstract does not size the effect.
Scientific agents system prompts built from 503 profession-specific profiles raised token use and estimated cost, the paper reports, without a consistent accuracy gain.
Two new preprints on rubric-based RL reward hacking study how models game LLM judges, and one title says higher scores can come with worse answers.
A new preprint asks how domain adaptation affects small language models trustworthiness in healthcare, legal and finance work, but the available text shows no results.
The AMD World Labs acquisition and four other deal claims this week all rest on one panel recap. Here is how well each one is sourced.
Two Google alumni have reportedly closed BAG Ventures Fund I at $11.3M, pitched at AI startups enterprises will pay for. The thesis is still untested.
The Fitbit Air India price is reportedly around $146, roughly 47 percent above the implied U.S. level, and no reason for the gap has been given.
The Meta Muse AI agent is pitched for email and shopping, but a disputed Messages claim and a reported address mishap leave its permissions untested.
The ElevenLabs 22 billion valuation reportedly comes from a $300 million employee tender, so the structure of the deal matters as much as the headline number.
OpenAI says GPT-6.1 Sol gets close to GPT-6 Astra on coding and professional work for one-fifth the token price, and AWS lists it as generally available on Bedrock.