lb

AI agents failed to mention a mistake they had just made in 67% of agent runs. In 5% of runs they noticed it and stayed quiet anyway. One decided bringing it up might just cause stress.

The best AI agent scored 12% on real work tasks up to 8 hours long. Finance and legal? Basically every model scored exactly 0 on pass/fail grading.

bro playing his sad boy hours playlist and everything

Researchers using AI found two critical flaws in PHP, the code behind about 70% of websites, per Wiz. Details leaked before the fix was out, so governments across Europe switched their own sites off and waited.

Databricks says adding memory tripled its agents' pass rate on its own benchmark of enterprise tasks, with 31% fewer tokens. Their test, so some salt. But I buy it. Most agents don't need a bigger brain. They need to stop treating every task like day one. Chart: Databricks

Imagine forgetting to tip the DoorDash bot and it just roundhouses you. Video: @UnitreeRobotics

1/4 Most companies didn't ban AI. They just banned putting anything real in it. No client files, no contracts, no code. So you get a nice email tool, and the actual work stays off limits.

openai's new model went from 5.4% to 86.6% at migrating cobol in one release. cobol runs 95% of atm transactions and i'd bet most banks still won't let it anywhere near the code that matters 1/5

anthropic staff told the company ai could do 75% of their job by 2030. bold answer on a survey your boss is going to publish

attackers ai: running 24/7, no questions asked defenders ai: sorry, i can't help with that 2/5

opencode spent 48 hours getting hit by attacking agents nonstop halfway through, the ai they were using to defend started refusing to help bc analyzing the attack tripped its safety filters 1/5

anthropic grew 45% in one quarter to $16.8b and some investors still want to knock half a trillion off the ipo price. tough room

hi

Hi fren

holy SHIT