SITUATION BREWING: DeepSeek is building a specialized Harness Team to directly challenge Claude Code, per Bloomberg. Verified WeChat accounts and fresh job listings reveal the Hangzhou lab is aggressively hiring to build rival autonomous AI coding agents.

SITUATION DETECTED: The UK government is drafting new biological weapons laws to regulate the use of AI in gene synthesis, per Bloomberg. Driven by fears of global guardrails failing, ministers want labs to screen customers and flag suspicious DNA sequences.

excited to speak with @theojaffee and @schisofrenia today on @MTSlive at 2pm eastern / 11am pacific. we’ll mostly be talking about the economics of AI, but maybe they’ll let me talk a bit about josh kushner and bob iger buying the lakers!

Huge stream today and so much to cover: - measuring how grok 4.6 stacks up to other frontier models - thrive holdings raises $2b & josh kushner + bob iger purchase the lakers - our take on what deepseek v4 pro could mean for the model pricing debate - lovable raises $400m - nebius +500% quarterly sales increase Following this up with some really great interviews: - @daveholtz professor at @Columbia will discuss his latest economic research paper & talk about the impact ai has on organizations live at 11am - @dschwarz26 co-founder of @FUTURESEARCHAI to give us more context on forecasting and how useful it is for making sense of the future - @hamandcheese chief economist @JoinFAI telling us more about his recent article and making a case against creating the superintelligent slime mold - @poteto and @romanugarte_ will be discussing grok bot, grok 4.6, and more at 330pm today live from our studio - @jessepollak covering what he thinks the future of independent agents transacting on-chain looks like - @steve47285 talking about his work @AsteraInstitute and most recent writing on the four flavors of llm misalignment - @Altimor founder of @getlindy discussing latest release of their teammate product - @exnx on what he's seeing in ai & biotech, as well as his work at radical numerics - @RKacaba talking spacex, its impact on public market investing, and more

DAILY SITUATION RECAP: SpaceXAI releases Grok Bot in early beta. Grok Bot is an agent swarm product designed for white-collar work tasks, an early preview of what Leopold Aschenbrenner called a “drop-in remote worker”. Bots can learn workflows and improve over time. This is likely part of the joint SpaceXAI-Tesla MACROHARD project, which seeks to eventually emulate entire software companies. OpenAI VP of Special Projects Brad Lightcap departs the company in order to start something new. Lightcap joined OpenAI in 2018 as CFO and helped build out many of its operations and business teams. He became COO in 2022, and VP of Special Projects (including the OpenAI Deployment Company) in April 2026. Previously, he was an investor at YC (along with Sam Altman) and worked in finance at Dropbox and JPMorgan. River AI raises $1.1B. The neolab, founded by xAI cofounder Igor Babuschkin, seeks to build powerful personal AI that knows and works for the user, starting with enterprise and moving to individuals over time. Anthropic is responding to tough investor questions on competition from cheap open-source Chinese models, on the firm’s tense relationship with the Trump administration, and on public backlash against data centers. Anthropic makes a $9.1B deal with Riot Platforms for compute. Riot is a former Bitcoin mining firm (and also former biotech diagnostic machinery manufacturer) that’s now pivoting to providing AI data center capacity. UK sovereign AI startup Cosine raises $15M, backed by the UK government and major financial companies. The 30-person startup plans to train LLMs for highly regulated industries that need sovereign AI. $15 million is around 0.005% of the total cumulative funding raised by OpenAI and Anthropic. Gemini hits 1 billion monthly users, becoming the fastest-growing product in Google’s history. Google now has 14 products (Search, Gmail, Drive, Chrome, Maps, Play Store, Calendar, Photos, Android, Discover, Lens, YouTube, Messages, Gemini) with more than a billion users. VC firm Accel raises $3.5B to make early-stage AI investments. $1.35B will go to larger rounds and follow-on investments, $800M to US (primarily Silicon Valley) investments, $800M to Europe and Israel, and $550M to India. Intel raises $20B in a share sale, showing continued demand for AI stocks. Chicago Mayor Brandon Johnson calls for a data center moratorium amid increasing anti-data center sentiment nationwide. Written by @theojaffee. Read more at our link in bio.

in my new piece for the @MTSlive substack I write about the last few weeks of bad vibes at the labs. some of this is down to what you might call "unforced errors." but the labs have responded quickly to their own mistakes, and with a candor and seriousness that surprised even me, a lab optimist. personally, I think a big chunk is attributable to growing pains. openai and anthropic, particularly, are speedrunning in months the transition from new startup to Power That Needs To Be Held Accountable that took google and facebook years. like everything in AI, expect it to accelerate. full piece here:

.@JamesCz19 on why autonomous vehicles are ridiculously safer than humans when America averages 40,000 road casualties a year: "Autonomous vehicles are able to get in and out of situations pretty fast. They abide by whatever the existing traffic laws are." "They're ridiculously safer than humans are on the roads. Unfortunately, the United States has about 40,000 on average casualties per year on the road." "The statistics when you're looking at autonomous vehicles by comparison are significantly lower. It's not even close. So there is that massive life-saving benefit right off the bat." @ConsumerChoiceC

.@JasonMa2020 on how Dyna-2 learned to untwist a real water bottle cap out of the box from just 13 minutes of dexterous hand data because the robot hand looks like a human's: "We got a new pair of dexterous hand. Most of our robots in the office have a gripper-like hand, so the gap between that versus human is actually quite large." "We thought, if we have a robot that has hands that look like human, wouldn't the transfer be greater? Right before the release, they were able to collect just 13 minutes of data." "They collected only 13 minutes of data and then post-trained the best Dyna model that trained on one million hours of data. The model is just able to, out of the box, start opening bottle caps on real water bottles." "This is very surprising because typically post-training a robot would take at least several hours of data. For a dexterous hand that's much more complex than a gripper, it would actually be much harder to train. But the experimental results we saw was the opposite. Perhaps because the hand looks more like the human, the transfer was easier." @DynaRobotics

Jason Ma reveals the first true scaling law in robotics: robots trained on 1 million hours of first-person human video predictably get better without ever seeing robot data "Yesterday we announced our new flagship robot foundation model called Dyna-2. It's very significant for the entire field for many different reasons. First of all, it's the first robot foundation model trained on at least one million hours of data. That in itself is a very challenging infrastructure challenge." "What we demonstrated is that even if the one million hours of data is entirely first-person video of humans doing manipulation tasks, we actually see scaling law for transferring to robot embodiments that the model has never seen before." "By training on more human data, we actually see predictable performance improvement on robots. That's a big deal because the biggest challenge facing robot foundation models is that we don't have enough data." "Unlike language models, there are just not enough robotics data out there, and it's very difficult to collect these robot data. Only by passively observing humans doing things can we scale this kind of data, and this is the first time we showed there is a lot of improvement that can transfer." @JasonMa2020 @DynaRobotics

Conduit CTO @riopopper on why the future of brain-to-text is skipping human language entirely and going straight from your thought embedding to the AI's embedding: "There's a separate type of thing that we're not really working on, that's more like silent speech, sub-vocalizations or tiny facial movements." "We're less excited about that because I see that as exciting and a step up from regular speech-to-text, but you're still fundamentally bottlenecked by human language." "If you can get the thought from someone's head directly to a computer before you even have to spend time compiling it down into words, like the high level idea, then that would let you get everything done much, much faster. That's what non-invasive can do." "We kind of want to go straight for the direct embedding to embedding communication with AI. If we make fun products along the side, then great, but that's not the number one priority." @NaomiBashkansky

Naomi Bashkansky reveals the linear scaling curve for reading thoughts from a brain that made her leave OpenAI to join Conduit: "You can get the latent representation of the true target text. You can get the latent representation of the predicted text. You can see the cosine similarity of these two." Rio: "You have the two vectors, and you take an angle between them. The bigger the angle, the less similar they are." Naomi: "We want them to be very close together. We can start with a model that has zero neural data. It's just doing pure next token prediction. Then we train it on increasing amounts of data that we have." "The plot that was very striking to me and why I ended up deciding to join is that it's just a very linear relationship on the log scale. If the X axis is doublings in data, then the cosine similarity will just go up as a very straight line." "The scaling curves are basically just really, really pretty if you look at them. I was like, oh, man, it's not really good now, but it's certainly going to get much better once we scale up to 1,000 times more data." @NaomiBashkansky @riopopper