Neal Khosla
@nealkhosla
I judge myself on what my 8 and 80 year old selves would think. CEO of @CuraiHQ, an AI-powered virtual clinic on a mission to improve access to care at scale.
For those of you trying to intuitively understand test time compute and why it's a big deal. LLMs are next token predictors. They take a bunch of words and predicts what comes next. This is a "greedy" algorithm that predicts one token at a time - which turns out to be powerful,…
hard to know what's true or not, but goes without saying that Yann is incredibly smart. and even though he's more bearish on this wave of AI, he's probably an undervalued asset. you want to chase tomorrow's paradigm and he has as good a shot as anyone at developing that.
zuck pitched an unnamed high-ranking openai researcher on becoming chief scientist, but they turned him down meta’s current chief scientist is yann lecun
dwarkesh is severely underrated and that’s acknowledging he’s very highly thought of
encryption is the last stand for reality and if it falls all hell is going to break loose
People continue to speak definitively about AI’s limitations while ignoring future improvement. Worse, they call people projecting improvement hucksters. The hucksters are the ones selling pessimist vibes to gain their own celebrity. Modern day televangelists.
This is… a really big deal.
5/N Besides the result itself, I am excited about our approach: We reach this capability level not via narrow, task-specific methodology, but by breaking new ground in general-purpose reinforcement learning and test-time compute scaling.
It will soon be a violation of the Hippocratic oath to not use AI in healthcare.
absolute nonsense that these other companies don't understand the bitter lesson as well as xAI grok 4 seems like a great model but this is plain wrong
🚨 Chamath explains Grok 4's breakthrough and how Elon leapfrogged the competition in AI: @elonmusk and the team at xAI understood "The Bitter Lesson" by @RichardSSutton @chamath: "(The Bitter Lesson) basically says in a nutshell, that you're always better off, when you're…
the largest impediment to firms making money in AI is still the technical depth and understanding of the median VC
I'm of the mindset that O3 probably couldn't one shot the average Stanford CS problem set yet so I'd have used it to unstuck me and teach concepts. So I probably would've gotten a TON of help from O3 while learning better and getting my work done faster. Idk if this is cheating.
Cluely's Roy Lee claims with total certainty that nearly every student at Columbia has used AI to cheat. AI is default for them. The world's not ready for what happens when the AI native hive mind grows up.
weekly reminder that o3 is so so so good. you should only be using o3
It's even funnier that big companies hire consultants to build these things when you consider they could just have some of their engineers be on twitter.
Huge opportunity for engineers, IT, or operations people to help companies implement AI Agents for enterprise workflows. If you’re on this site, you have a cheat code for what’s happening in AI that puts you ahead of the adoption rate of most orgs by a couple of years.
the more I see stuff like this the more I feel like the optimistic case for AI is just that our basic economic needs will be taken care of and everyone is going to have incredible tools to help them realize self actualization and creating things for fun
Kling 2.1 Master handles this kind of prompt beautifully: POV shot of a gravity surfer diving between ancient ruins suspended midair, glowing moss lights the path, the board hisses as it carves through thin mist, echoes rise with speed I love how it turned out! 🏄♂️ And now,…
If the robotaxis are based off of stock FSD they are very good but completely prone to doing stuff like this all the time / definitely not something I would want to ride in unsupervised. The gap between very good and ready for primetime is massive and Waymo still has an edge.
Man I dunno.
we are so so so cooked. our brains are not set up for this kind of dopamine.
I'm sure you've seen tons of AI videos of animals at the swimming Olympics... Here's how you can make them yourself in this thread! 7 example prompts + info🧵👇
venture has always been reliant on outliers so of course any rational projection of the math doesn't work you signed up to find an 100B+ outcome. whether you like it or not that's the bar