capital -> sand hill road labor -> page mill road

OpenAI finally added structured generation, which is a good opportunity to remove unnecessary abstractions. In just a few minutes I added this Grit example for switching from jason liu's instructor to the native `parse` method. docs.grit.io/patterns/libra…

OH during a meeting: "i was reading the llama 3 technical report and it turns out we're already doing everything they're doing for synthetic data"

summer, late afternoon, biking through palo alto

can confirm the product is even cooler than the demo makes it seem, congrats weber!!

Qwen just released a 70B math model that seems to beat the math-specific version of Gemini 1.5, and also a 1.5B math model that beats the math abilities of Llama 3.1 70B (h/t LDJ for pointing this out!) x.com/Alibaba_Qwen/s…

at morph we have been saying a couple of mottos which i think bear repeating: - "empathy for the machine" (how we make product decisions) - "minimize time to highest-order bit in the craziest way possible" (how we make R&D decisions)

first Qwen, now DeepSeek - Chinese AI labs feeling the autoformalization, putting their heads down, and doing the work (h/t the very well-informed LDJ for screencaps)

james is not only one of the most thoughtful AI product leaders i know, but also one of the funniest - excited to see the next chapter!

Super interesting work from DeepSeek on MiniF2F (so happy to see our benchmark still in use \o/). It's hard to compare this with the recent DeepMind paper but from my experience building and using MiniF2F I think ~60% pass-rate is likely comparable to DeepMind's recent result on

Juniper by the way was a 2500X return at distribution to my firm or a $7B profit on $3m investment in the incubation, when B's weren't common in VC because no major telecom wanted TCP/IP in public telecom networks (imagine!) and Cisco's CTO told me they would never do a TCP/IP

come by this demo night and learn more about how morph labs builds empathy for the machine lu.ma/x7dj9pm6

accelerationism, or as i like to call it, yassified nihilism

every PhD thesis separates your work into an oeuvre (papers that make it in) and an anti-oeuvre (papers that must be sacrificed for the narrative). rarely do both consist of so many pivotal contributions - congrats albert!!

walking through nyc w someone who recently read the power broker is to hear "you know, robert moses tried to build an expressway through this playground" every five minutes

unstoppable force, immovable object

when will the AI space reach this level of mania? (Zip2 mentioned)

professional c-level exec on founder mode:

so much depends upon the fact that reading a word of memory from the GPU is 200x slower than doing an FMA