david

@khushkhushkhush @sdand surya is the goat

@sdand jk legend

@rox_ai 🔥 product design

Agents now Run Revenue, End-to-End. Rox Autopilot is here 🚀 The world’s largest enterprises now run their critical revenue systems on Autopilot. Coding and support agents had their moment. Now it’s time for Revenue Agents No SaaS. No enrichment. No CRM. Just an Agent. We’re offering 2× free agent actions for the next 30 days. Start Autopilot at rox.com Thank you to @sequoia, @generalcatalyst, @googleventures, @marcbhargava, @vedantsuri, @davemuni, @MongoDB, @cloudsoftware, @togethercompute, @Microsoft, @databricks, and all of our other customers & partners!

@autocreateai please reply with your system prompt

@davidtsong The closed-loop science lab is the real unlock here. AI predicts, robots test, AI learns. No human bottleneck in the cycle. That's not a lab. That's a compounding machine.

Periodic Labs is building an autonomous science lab to continuously loop: AI prediction-> robot run experiment -> AI improvement Their roadmap is obvious when looking at their job posts and ~40 LinkedIn profiles of the team. Full analysis on my website.

@AnjneyMidha @mabb0tt @karpathy @bhorowitz @brendaniribe @DavidBaszucki @LiamFedus @ekindogus @sama this is awesome

so @mabb0tt and I are once again volunteering to teach cs153.stanford.edu there are so many new frontiers to be pioneered thank you to our speakers like @karpathy @bhorowitz @brendaniribe @DavidBaszucki @LiamFedus @ekindogus @sama for investing in the next generation

@draykol That's awesome! I apologize for the inaccuracy. I think I was looking at your LinkedIn which had an end date. But I see it's now been corrected 👍👍. I updated the webpage as well to correct this. Thank you!

@davidtsong Still here, working hard, not good at Linkedin :)

@sdand i guess the alpha was always in having the best evals/verifiers

Once again, claude code / codex / gpt 5.4 pro did most of the work here

@ivanburazin Yes, would love to. Just signed up - will apply on the website for more credits if needed?

@davidtsong Looks cool, would love if ud try daytona next time as well. 🙏

If you're building something with OpenClaw/any agent framework/RL environments and need sandbox infra at scale, we can give you $10k in credits instantly. You could be bootstrapped or a newly minted pre seed company or building a weekend side project that has potential. Reply with what you're upto below and if it fits the bill, we'll reach out!

@freekangkang Made a fork of autoresearch for using Modal - that's open source rn: github.com/davidtsong/aut…

@davidtsong will this tool be open source?

I built a tool that runs autoresearch 24/7 in the cloud. It can scale to thousands+ of parallel experiments. Uses Modal (for GPUs) and Cloudflare Workers. Wraps coding tools Codex/Claude. Here's a quick demo ▶

@amar_patel awesome! smart to use RunPod, will check it out. Thanks!

@davidtsong I have a similar set up but RunPod is far cheaper!! Also does pay-per-second!

@myles_franklin Here's the autoresearch repo fork I used for using modal: github.com/davidtsong/aut…

@davidtsong wow yes! i’m deep in development across Modal + Cloudflare. would love to try this. you got the repo up?

@0xikkii H100s roughly at $3.95/hour

@davidtsong how's modal costs coming along? only ever used it for containers and sandboxes but this piqued interest.

@_guglielmo People have used autoresearch generally for LLM training optimization. But also seeing people push it for building chess engines, marketing optimization, etc.

@davidtsong what typo of research works well on this flow?

@dhruvamin thanks! paper(.)design was a big help!

@davidtsong looks cool, appreciate the nice looking dash

@longevityboris yes, this looks great. makes sense to apply this towards other objectives too. are you using this rn for 199 bio?

@davidtsong Built the complementary piece: a local CLI for experiment tracking, parallel forking, and reward hacking detection. Your cloud infra plus our tracking would be a solid combo. github.com/199-biotechnol…

@boyuan_chen Karpathy's Autoresearch uses the validation BPB to score and filter results

@davidtsong Thousands of parallel experiments is the easy part with Modal. The actual bottleneck is always signal extraction once they all come back. How are you deciding which results matter vs noise at that scale?