I invest in founders that are trusted by young metro

@nealmohan @YouTube what are the bets?

another heater from @nateparrott

@nikitabier Must be hard

ur one crash out away from your best lock in

@sama feels so much better to show the summarized CoT by default -- really feels like it's working instead of making me wait 🙏

@AravSrinivas What about training-time-compute? Where we get the model to “think harder” about some tokens in pretraining instead of just “glancing” at everything? Will this be needed or have we basically mitigated it with data sampling and FT

Before PMF your moat is speed but after PMF your moat is speed

@signulll maybe misguided but I worry that there’s a mentality at some large labs that focusing on HCI is a waste of time because if you push raw intelligence to ASI then it won’t need to interact with any humans to provide economic value

RT @maxhodak_: with the kinds of takes you see online it’s actually a little surprising that pretraining on the internet leads to functioni…

@iamgingertrash we are the voices in their heads

please — for the love of god — download cursor and try to build something you will surprise yourself

@nikitabier gr8 b8 m8

@bwarrn @blakeandersonw heartbreaking but not surprising… many such stories

the french build consumer social because they still believe in love

which is larger

You remember B2C You love B2B 2025 will be the year of B2A (AI Agents) B2A products are purpose built to serve agents (like Operator). In time, agents will do the majority of economically valuable work but B2A products will be the catalyst.