Jay Kreps

@luke_metro Sadly we don’t know how anymore. The technology is lost to our civilization.

we need parallel landings in SFO again, the global economy depends on it

PALO ALTO NETWORKS on MYTHOS: "In our testing, three weeks of model-assisted analysis matched a full year of manual penetration testing, with broader coverage."

@sampullara If we taxed billionaires we could pay for every American to have two full time umbrella holders.

none of the billionaires i have ever known had one of their employees holding an umbrella for them

One of the things that made the Mythos release hard to interpret is that Anthropic held back details on most vulns they found, to give defenders time to patch. 1 month later, info from orgs with access to Mythos is starting to trickle out, e.g. this post from Mozilla today:

Texas just passed California in utility-scale solar. And it's not close in wind or energy storage.

Effective today, we are: 1) Doubling Claude Code’s 5-hour rate limits for Pro, Max, and Team plans; 2) Removing the peak hours limit reduction on Claude Code for Pro and Max plans; and 3) Substantially raising our API rate limits for Opus models.

Scott, agree with the sentiment, but it seems like if we want to move the energy supply for transportation, heating, etc to electricity we need to make it cheap. We may be “top rated” but California is virtually the worst in the nation on this dimension. It would seem whatever we are doing on permitting etc objectively isn’t working. We somehow transform heavy investment in the world’s cheapest electricity source (solar) into the country’s most expensive billed electricity. Is there reason to believe that grid investments won’t just turn into another CA boondoggle like high speed rail?

The future is about electrification: Massive generation of clean energy of all types — solar, wind, geothermal, modular nuclear, etc. — & a modernized grid with the capacity to carry this clean energy to meet our growing power needs in an environmentally sustainable way. We need to make it dramatically easier & faster to generate clean energy & to upgrade the grid to carry it. It’s far too hard now, with a million veto points. Reforming our permitting system for electrification is a key priority for me & has been for a long time. I’ve authored laws to expedite permitting for solar & storage — leading to California being the top-rated state — & I’m currently authoring legislation to expedite transmission upgrades, heat pump permitting & use of plug-in solar. This will also be a huge priority for me in Congress.

Worth flagging that one of the most popular and influential computational frameworks in cognitive neuroscience holds that all cognition — including human reasoning — bottoms out in prediction-error minimisation.

It’s hard to overstate the advantage that comes from the US tech companies and capital markets willingness to see a big idea and just push all the chips into the center of the table.

@jwfbean Yes pretty stable.

@jaykreps Roughly what fraction of them were “absolutely important”, “not really but fine”, “seems kinda inappropriate”, and “outlandishly wild”? Did the proportions stay the same as the org grew too?

It was interesting to me as a CEO that the number of slack messages and emails from employees I got per day stayed pretty constant as the company grew from 30->3k, even though I encourage people to slack me directly. Somehow most people have pretty good sense as to what to reach out to the CEO about and they calibrate as the org grows.

@Som_Mohapatra Kind of varies but maybe 30-40 chat sessions. I get gobs of email but usually more project update type things, we mostly use slack for communication internally.

@jaykreps That’s actually really interesting, and different from my experience. What did the number of slacks/emails per day converge to?

I’d question skilled tradespeople as a hard bottleneck, for exactly the reason you give: there are a lot of them spread over various commercial construction priorities. If the AI buildout is willing to pay more, they will reallocate to the AI buildout. None of the numbers we’re talking about in the next year or so are tapping out the entire commercial construction capacity of the US, right?

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude. Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

Our Long-Term Benefit Trust has appointed Vas Narasimhan to Anthropic's Board of Directors. Vas brings more than two decades of experience in medicine and global health, including as CEO of Novartis. Read more: anthropic.com/news/narasimha…

@JigarShahDC It seems that article argues that grid connectivity is better and cheaper. But it doesn’t seem to disagree with the central premise that if you can’t get it, colocation is an option and people are doing that. Is that wrong?

Isn’t the underlying point correct? There are many ways of generating power and this market is not that price sensitive so capitalism is likely to find a way? He’s comparing to TSMC and ASML where all production runs through a single company bottleneck of immense technical difficulty with no hope of new entrants emerging in the next few years?

Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

@Noahpinion A more interesting critique: the Karl Friston wing of Neuroscience thinks the human mind is basically a fancy autocomplete of signals from the nervous system. So maybe a sufficiently advanced autocomplete of all the complexity of the world *is* intelligence,

If "autocomplete" can do math better than you, write code better than you, forecast the future better than you, and persuade swing voters better than you, then calling it "autocomplete" starts sounding like defensive cope

As you read about Anthropic's Mythos capabilities to find critical security weaknesses, consider what if a Chinese AI company had gotten here first. There is a real race underway, and its in our interest I believe for U.S. companies to win.