autoformalization is undergoing a renaissance. i'm really excited to see this happening. there is an incredible convergence happening in the AI x math space: the Lean ecosystem is thriving under the stewardship of its FRO, top mathematicians are paying attention, models

"The time has come. Let us begin the true ceremony. The Mathematical Instrumentality Project"

the agents yearn for GPUs

Welcome Math, Inc. - a new company dedicated to autoformalization and the creation of verified superintelligence incubated at Morph over the past few months. Uniquely enabled by Morph's Infinibranch-native environment compute infra, Math, Inc.'s first product is Gauss, an

Working with Gauss, I got the sense of a new paradigm of human-machine collaboration, especially for those who want to verify math they care about, but don't code by themselves.

In the first year of my PhD, I formalized the independence of the continuum hypothesis in Lean. It was considered a breakthrough at the time, taking ~12 months and 20K lines of code. With Gauss, we finished Strong PNT with 25K LOC in 3 weeks. There are two really remarkable

the 10X engineers of today, and soon the 10X mathematicians and scientists of the future:

seeing a lot of people just rawdog CLI agents with simple MCPs to synthesize training data / envs from the web the age of experience hasn't started yet, but will be powered by a million pearls of a great consciousness in a Great Scattering which is to say: if your RL envs

the most advanced agentic reasoning systems in the world are being built right now on Morph and this is a great example!

been hearing great things about these guys. cracked team

you can learn a lot about a person from their personal multiagent claude code setup

time to cook, what should i build with tinker?

browsers for agents with 10X better scalability, cost, record / replay, and infinite persistence 👀

why would you want to use this? - you need a lot of browsers running at once, for example agentic search over arxiv html pages, QA testing, fullstack development, comparing local dev to staging deployment, last-mile automation - you want to do browser use RL - you want to

Thank you rank-1 LoRA RL

congrats Lawrence Chen and team! the agent devbox / agent workspace use-case is something we've seen a ton of interest in recently and is exactly what Morph Cloud is built for - a world where each user manages a multiverse of hundreds of different agents and workspaces at scale

Agent workloads are really bursty both during training (RL) and deployment. Groups in GRPO require many trajectories to be sampled in parallel from the same starting point; same for best-of-N agentic reasoners and test-time search scaffolds. This sucks for systems and infra

(investor): could you send over a deck? whatever is available (me):

congrats 🇺🇦 Alex Polozov!!! whereever you're going next is going to absolutely crush it