🔥Telegram İfşa

🚀

🎙️Unleashing my inner researcher + policy + political wonk combo this convo with daniel bashir for The Gradient. We tackle the mahjong of AI policy - China, innovation/regulation trade-offs, and why researchers becoming bilingual (in tech & policy-speak) is crucial.

Google Meet option to talk with a BetterHelp therapist after stressful meetings

Some cool ideas in the thread

the purpose of reading a book isn’t to retain information, it’s to refine your worldview just a little bit with each one

Ben Lang Abnormal Security: >$100M ARR growing >100% y/y (series C). Help use good AI to stop crime, and protect humans and from bad AI! careers.abnormalsecurity.com

I am excited to announce the $250M Series D for Abnormal AI. Led by Wellington Management, plus Greylock Partners, Menlo Ventures, Insight Partners, and CrowdStrike Falcon Fund, this milestone is another step in our mission to protect humans using AI. abnormalsecurity.com/blog/building-…

My favorite Mumbai story. The city is just something else.

🔥2025 is the year of agents, but are we there yet?🤔 🤯 "An Illusion of Progress? Assessing the Current State of Web Agents" –– our new study shows that frontier web agents may be far less competent (up to 59%) than previously reported! Why were benchmark numbers inflated? -

Gemini 2.5 Pro often responds with several "Options" when you ask subjective questions. I've seen it answer 3 different ways and asking the user to pick whatever suits their needs, all in one shot. Great handling when intent is under-specified.

As we all know by now, reasoning models often generate longer responses, which raises compute costs. Now, this new paper (arxiv.org/abs/2504.05185) shows that this behavior comes from the RL training process, not from an actual need for long answers for better accuracy. The RL

Language models thinking step by step to do arithmetic...

It is critical for scientific integrity that we trust our measure of progress. The lmarena.ai has become the go-to evaluation for AI progress. Our release today demonstrates the difficulty in maintaining fair evaluations on lmarena.ai, despite best intentions.

Gemini really loves explaining via analogies, makes its writing easy to detect in the wild

Gemini 2.5 Pro and Flash are 🚀Generally Available🚀 and with the new 2.5 Flash-Lite capture the quality/price pareto frontier. Read more about them in our Tech Report!

Following advice that’s optimal for average people only optimizes for becoming the best among average people