Are robots ready to accelerate science labs? We created the WetLabs Benchmark to test 9 critical tasks, from easy to hard, and let 3 leading models take 20 attempts per task Astra leads but it comes close...

Today, the @CFTC took an important step toward bringing regulated onchain markets to the United States: firms can now keep required records on a public blockchain without being required to maintain a separate offchain copy. The CFTC also clarified that firms can invest customer funds in tokenized versions of investments that are already permitted. That matters because a regulated firm can now use a public blockchain as its system of record, where every entry is transparent, tamper-evident, and verifiable by anyone. Those are the assurances the CFTC’s recordkeeping rules exist to provide, and public blockchains deliver them by design. In July, HPC and @phantom asked the CFTC to provide this clarity. Today, the CFTC delivered.

I asked people: "which robotics labs and startups have the highest talent density?" 28 votes: Physical Intelligence (@hausman_k, @svlevine, @chelseabfinn, @adnan_esm, @brian_ichter, @QuanVng, @lachygroom) 18 votes: Generalist (@peteflorence, @andyzengineer, Andrew Barry) 9 votes: Sunday (@tonyzzhao, @chichengcc) 6 votes: Skild AI (@deepakpathak, @gupta_abhinav_) 4 votes: Dyna (@Lindon_Gao, @YorkYang5050, @JasonMa2020) 3 votes: Bracket Bot (@sincethestudy) 3 votes: OpenAI Robotics (@model_mechanic, @PengchuanZ, @ChengshuEricLi) 2 votes: 1X (@BerntBornich) 2 votes: The Bot Company (@kvogt, @pariljain, @lukeholoubek) 2 votes: Figure (@adcock_brett) 2 votes: Genesis (@zhou_xian_, @theo_gervet, @johnsonwang0810) 2 votes: Prometheus (@JeffBezos, @vik_bajaj, @sherjilozair, @wgussml, Alex Graves) 2 votes: Robocurve (@chooi_jeq) 2 votes: Tesla (@elonmusk) 2 votes: Weave Robotics (@kaandogrusoz, @evan_wineland) People couldn't pick a company they work at or founded. Disclosure: I'm a small investor in Physical Intelligence. I didn't vote.

lots of things to unpack: 1. our largest deal yet (big size) 2. our 1st investment-grade offtake deal (great credit) 3. a new product offering to help scale our business - that k3 annc was a part of it you think you know what USDai is, but the design is improving each month

rogue agents and prompt injection are currently considered the biggest unresolved threat in security. picture this: you ask claude code to review a PR, the agent reads a compromised package and gets tricked to exfiltrate your environment keys. we are building cyber defense to solve this exact kind of problems. Clarion can ingest any kind of telemetry, integrate with platforms, triage and resolve issues at machine speed. in this example Clarion is set up to ingest agent telemetry. the developer was using Claude Code and a prompt injection made the agent post his AWS keys to a remote attacker server. the attacker then logged into his AWS account using that same key. Clarion is able to correlate actions across different environments and take action. here it detected the exfiltration via Claude telemetry, confirmed the abuse by correlating CloudTrail with auth from a differnet IP, contained the host using CrowdStrike Falcon, blocked the attacker address and escalated the issue. all this in ~6 minutes. a human operator alone would have not found out in weeks.

CFTC is not playing around. Wasting no time, yesterday they filed what appears to be a proposed rule for interagency review entitled “Regulation Crypto Asset Transactions and Regulation Crypto Asset Markets.” Things are going to move fast in DC.

Breakout List was foundational in thinking about where to intern/work when I was just starting my career. The benefits of joining a high-growth, talent-dense company are hard to overstate. Extremely happy @chrisbarber has brought this back, and would recommend you check it out!

The Thru wallet: - won't drop support for the Thru chain because another chain is strategically more important. - does upgrade before the chain moves on without it. - empowers developers to set their own RPCs. Coming soon. Actually possible on @thru_xyz .

If you've ever tried earnestly to build a +1GW site, you know that the bottleneck is credit. Capital is "abundant" insofar as there is a lot of money seeking >800 BPS AAA bonds, but if you are not that then you can go pound sand. Nat gas turbines or recips on reasonable timelines are available to anyone with a down payment. Attractive green and brownfield sites are also not in particularly short supply, but did you have the speculative cash to secure your batch 0 status? Likely not without... you guessed it, credit.

This vial contains a new drug called PAC-3310. It was designed by ChatGPT, and I synthesized it in a chemistry lab I built in my garage. PAC-3310 is a new selective M4 muscarinic receptor agonist for treating schizophrenia - similar to the recent breakthrough drug Cobenfry, but improved. (1/7)

CFTC just filed their motion to dismiss CME's lawsuit over perpetuals and they cooked 🔥 The motion argues that CME lacks standing for several reasons, including that it too could offer perpetuals, so any injury is self-inflicted ✅ "This lawsuit is much ado about nothing."

Increasingly the moat in Neoclouds is financing (probably the moat everywhere @ this point). To get good financing you need a big customer with a strong balance sheet to sign a long term contract, to get that you need a lot of power as most of these big customers will only be interested in deals north of 50 MW. No one really designed data centers before to have over 50 MW in one site. All the sites that have that kind of power are new or are under construction right now. Additionally, most of these financing cost isn’t the actual site construction it’s the GPUs which are orders of magnitude more expensive. So the play is to focus on finding power first, locking down a customer second, and the finally using that customer commitment or prepayment to finance GPUs.

Ever since I started working on robot foundation models a handful of years ago, the broad ability to one-shot in-context learn has been the single most vivid goal in my mind along the long road ahead. It’s something @andyzengineer and I have especially been thinking together on for several years now, and something shared by the whole team at Generalist as a major goal. But even before anybody started talking about robot foundation models, too, the capability this enables, i.e. generalized one-shot learning, has been both incredibly concrete and elusive. In terminology that I think many other people have come up with as well, in grad school we used to talk about a “Ctrl-C-Ctrl-V” type of capability. See something, and have the robot do it too. This same type of idea inspired the name of the 1970 MIT copy demo ( The thing is, it sounds simple, but is incredibly hard since the world is never quite the same when it got copied and where you want to paste it. The real world can be hard to predict and is full of variation. To do this, you need strong generalization, and it needs to be acquired in a single example. Doing this over a wide range of tasks, especially for dexterous tasks, is hard mode. Lots of the components of the idea of making this all happen have been there for a long time. As an example, this 2017 NeurIPS paper “one-shot imitation learning” has excellent vision, with ambition well beyond what was achievable at the time, and although it’s not referred to as “in-context learning” since it was pre-Transformer, it actually uses attention to condition on a single demonstration. And now, many things have happened since early 2017, including the broadly celebrated arrival of one/few-shot in-context learning in language models in 2020. This new model GEN-1.5 takes in everything we have built and learned over the past couple years at Generalist. It has been training for 8 months. It has taken an incredible amount of commitment and grit from the whole team to get here. The level to which this model has survived many surgeries has continued to surprise me. And its capabilities have continued to surprise as well. We found compositional generalization on Friday. We found sim2real prompting earlier last week. We filmed the contiguous uncut videos of live prompting yesterday. To be clear, the success rates are modest, and there’s still a long way to go. But now I have definitely seen a ~decade-long imagination come into the real world.

Introducing GEN-1.5, a one-shot learner. It can learn new tasks in a few seconds. Show it what to do, and it generalizes. This capability emerged from pretraining on physical data at scale, as a step towards our mission of building general intelligence for the physical world.