I’ve started a new company with @tkkong! TK is a driving force behind a lot of Ramp’s success, building much of the core product, incubating the procurement platform, and leading Ramp Labs. We’re a team of IMO and Physics Olympiad gold medalists, and we’re hiring the most talent-dense team.

When I was considering joining Anthropic in 2023, their main pitch was “we have prioritized safety over commercialization and actually made decisions that hurt our business because they were the right moral choice.” Everyone I met at the company was incredibly sincere about their mission alignment. I’m cynical and don’t believe that some people joining Anthropic today would hold this position, but I do think the early team really believed in this mission.

“Smart model review all pretraining data” seems like an attractive option until you realize that this costs exactly the cost of pretraining, or at least on the same order If you try to be smart about it and ask a model what it thinks of each pretraining token, that’s basically distillation

Fable is definitely a step forward in execution ability, but I still find myself providing a lot of architecture guidance. Example: while building a Slack agent, Fable chose to stream back SSEs from the agent session to Slack, and I had to instruct it to give a tool to post to Slack instead.

Warriors or Knicks game as referral bonus: send over your most cracked eng friends and we’ll take you to a game next season if they join

Mythos / Fable launch reminds me of strawberry / o1 launch. Hype mostly justified but clearly serving a controlled narrative Cybersecurity / safety implications beyond what the existing safeguards were designed to handle The feeling of it's so over for everyone else, and yet proof of existence is all it takes for the most resourceful among us

After a semester building internal tooling at @conviction, I’m excited to announce that I’m joining @tkkong and @philhchen this summer at Concept! I first met TK through Shreyan Jain (cc @shreyanj98) when I was looking for designers to speak with last summer. He’s a clear thinker and I deeply enjoyed that conversation. I noticed Concept in the latest Embed batch and reached back out to join their small and mighty team. I’m looking forward to wearing hats aplenty over the next twelve weeks! I’ll be posting updates and learnings as I go. 🚀

.@josephjojoe is a man of many talents. He's already made impact across product, GTM, and design (more to share about this soon). He's also gotten everyone in the office saying "revealed preference" and doing the 6-7 move.

2026 was about rapid expansion of AI budgets. 2027 will be about cost optimization. My token run rate is ~$250k / yr now with 5.5 / 4.8 all on xhigh when I use API (thankfully I usually don't). With Mythos / GPT-next, this might look more like 1M / yr which feels irresponsible

is anyone building dynamic workflows with Claude managing Codex subagents? This would be the best of both worlds. Claude is much better at task decomposition and Codex is much better at executing well-defined specs.

Had a great time sharing how we’re AI pilling our customers at @DecagonAI @a16z @Accel event Reach out if you want to build the future of work with us or AI pill your own teams

Thought experiment: if @karpathy's efforts at Anthropic yield a Claude model that is capable of pretraining the next generation of Claude, then any company with sufficient GPU infrastructure could use Claude to pretrain their own Claude-clone. Of course, Anthropic would then ban that company from using Claude. But then wouldn't any company with enough Claude spend be incentivized to use Claude to train their own Claude-clone eventually? What happens in 1-2 years when even open-weights models become good enough to run their own training?

We’re sharing our internal AI job board Every company will have internal ops and engineers building AI agents Discover roles from @Box, @tryramp, @DecagonAI, @baseten, @WeAreLegora, and 150 companies And if you’re an AI lead driving internal transformation, join our leaders community below internal-ai-jobs.concept.sit…

Incredible work from @dwarkesh_sp and @reinerpope. Must-watch for anyone who wants to understand modern AI systems

I left Google DeepMind, moved from SF to NYC, all within 2 weeks to join @quadrillion_ai — to build the future of automated research intelligence with the highest slope founder and most talent dense team. I grew up in Silicon Valley — the old Facebook office was my second home. I’d hang out there after school, drawing with my crayons while looking around at the sea of computers with lines of code. Since a young age, I felt empowered to have an array of interests beyond tech: piano, ballet, figure skating, art. The valley embraced diversity of thought, and that’s what inspired me to stay for Stanford and my career thus far. But today, SF is one big hive-mind. So, I moved to NYC, away from family and friends to build a company that doesn’t need to rely on a bubble to survive. I’m meeting customers day after day in all kinds of verticals, connecting with them in different ways and seeing our product bring real value. Here, I’m able to live in diversity of thought. I’m excited to build the future of research in the city of opportunity. Let’s chat if this excites you.