Tamaybes

Mechanize | San Francisco, CA (Onsite) | Junior SWE ($300k + equity + bonus), SWE ($350k + equity + bonus), Research Engineer, Alignment ($500k + equity + bonus)

Applying takes <2 min: https://www.mechanize.work/apply/

We build environments that frontier AI labs use to train their models to do real-world software engineering. Team of ~45. We have more demand than we can keep up with, so we're hiring fast. Backed by Nat Friedman, Daniel Gross, Adam D'Angelo, and Patrick Collison.

Mechanize | San Francisco, CA (Onsite) | Junior SWE ($300k + equity), SWE ($350k + equity) Applying takes <2 min: https://jobs.ashbyhq.com/mechanize

We build environments that frontier AI labs use to train their models to do real-world software engineering. Team of ~25. We have more demand than we can keep up with, so we're hiring fast. Backed by Nat Friedman, Daniel Gross, and Patrick Collison.

Mechanize | San Francisco, CA (Onsite) | Junior SWE ($250k + equity), Senior SWE ($375k + equity)

Applying takes <2 min: https://jobs.ashbyhq.com/mechanize

We build RL environments that frontier AI labs use to train their models on real-world software engineering. Team of ~20. We have more demand than we can keep up with, so we're hiring fast. Backed by Nat Friedman, Daniel Gross, and Patrick Collison.

Mechanize Inc. | San Francisco, CA (Hybrid, ONSITE preferred) | Senior SWE ($500k+equity), Junior SWE ($250k+equity)

Apply at: https://jobs.ashbyhq.com/mechanize

Mechanize builds sophisticated reinforcement learning environments to simulate realistic software engineering tasks (feature development, debugging, refactoring, reliability testing) for frontier AI labs. Our mission is to automate software engineering first, then all economically valuable work. We're growing quickly, working with leading AI labs, and backed by investors like Nat Friedman, Daniel Gross, Patrick Collison, and Jeff Dean. Featured in NYT and TechCrunch.

CMOS processors can become around 200x-fold more energy efficient than the H100.

TLDR: the H100, lower precision, and other advances lead to a big jump in computational performance. We're in for a wild ride when the next generation of models is trained on 100x more compute in 2024 and 2025.

The result about recent compute trends is different from the recent trends described by OpenAI. In particular, they find a 3.5-month doubling time over the Deep Learning Era, whereas the paper finds a 6-month doubling time.

I think the Large-Scale Era does point to a new phenomenon that emerged pretty discontinuously, which is that there are now 'two lanes' in ML scaling. Prior to 2015, academic and industry would train roughly similarly compute intensive models. Since then, a small number of industry players frequently train models with 10-100x more compute than what the typical researcher uses.

Not parameters, the amount of FLOPS required to train the model.