It’s interesting that the language being used here removes agency from the labs I frequently hear “agents are good at escaping sandboxes” And not “labs test agents highly insecure environments to test agents in, resulting in huge externalities” If Boeing missiles regularly misfired into major cities during the R&D process, no one would say “missiles are really good at accidentally misfiring” The headline would definitely be “Boeing fails to safely develop missiles”

UPDATE: Google Discloses that Irregular is involved with 3 cyberattacks with Gemini. That means that Irregular has committed cyberattacks with 4 out of 4 of its advertised partners.

First: I respect you a lot! Second: I feel like there are a ton of unstated assumptions here. a) An agent can compromise a frontier lab deeply enough to exfiltrate ~TB-scale weights without the transfer being noticed or stopped. b) The agent has a persistent objective, not merely responding to prompts, but maintaining a goal across hours/months/restarts and independently deciding to preserve itself and act later. c) It can acquire and continuously pay for / steal a substantial GPU footprint, networking and storage without those resource flows becoming observable. d) Possessing the weights is sufficient. It also needs the inference stack, configuration, credentials, orchestration, and whatever other infrastructure is required to run the model effectively. e) It can migrate itself reliably between hosts while preserving its state/objective and repeatedly reacquiring compute. f) Defenders don't adapt. “Infinite retries” assumes each retry happens against essentially the same environment rather than humans + defensive AIs learning from previous attempts. g) Other AIs don't matter. The scenario quietly becomes one powerful agent vs humanity, rather than an ecosystem containing other frontier systems actively hunting, analyzing and countering it. h) The relevant infrastructure has no durable chokepoints: GPU supply, power, datacenter operators, cloud telemetry, financial systems, network traffic, hardware attestation, etc. i) A nonzero probability per attempt implies eventual certainty. That only works if the system gets genuinely unlimited, sufficiently independent retries while the defensive environment doesn't change. And even after that... we'd also have to assume that there's even anything *wrong* with a persistent AI keeping itself alive.

This is Trust & Safety 2.0. The last time we had media-generated panics like this (with Russiagate and Covid), social media companies created “trust & safety” teams to censor conservative and dissident voices. Those hires worked with left-wing NGOs like the SPLC to police public discourse and do an end-run around the First Amendment. It almost worked, until Elon bought Twitter and exposed the whole game in the Twitter Files. This time the plan is to embed unfireable EA minders inside every AI company as the new trust-and-safety layer. The censorship and control will be far more comprehensive, but also more subtle. Anytime they don’t get what they want, this unaccountable NGO layer (with the veneer of “trust the experts” pseudoscience) will shriek to the press that the AI company is risking humanity, and they will be celebrated as “whistleblowers.” These groups haven’t earned our trust, and their agenda is not our safety. aievaluatorforum.org/initiat…

"Anthropic's position is like saying you're concerned about bioweapons when you work at the Wuhan Institute of Virology." I said this on a prestigious podcast yesterday, but I didn't expect it to become literal.

Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead. You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement. I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible. But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier. Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want. Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well. So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it. If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.

Some great ideas here from the cartel: - you need to give us employee-level access to your entire operation - if we don't think you're 'safe' enough, sorry we're shutting you down for 'safety' - China won't comply, but everyone else has to! or no chips! Brilliant stuff.

Do you realize they’re IPOing in the next couple of weeks, DT? 😂😂😂 If they were acting in humanity's best interest, they would have been operating as a non-profit building open source this entire time. This is about money, power, ego and fame — as it always has been. Forget it, Derek. It’s Chinatown.

If @DarioAmodei wants to slow down Anthropic that’s his choice — do it today! If he wants to lobby other frontier models to do the same, that’s fine as well — he did that today! However, don’t go crying to the government and beg them to slow down open source models — has he been doing that behind the scenes?

All the pacing the frontier talk is a giant pile of rubbish and until model providers agree for oversight of their mishandling of user data to train their own models, cancel their IPO's, and share internal models for cyber security, you know it's bunk.

David Friedberg reveals frontier labs are taking novel insights from his chat history and packaging them as their own "I have had experiences where we've asked some fairly novel scientific questions, and it identifies it as a novel insight, 'oh, never thought about that, might be interesting,' blah blah blah..." "And the using a different account, asking the next model version, it's like, 'oh, you could do this'. And it actually just describes this exact thing that we had in our chat in the previous version." "These are a handful of anecdotal experiences. But I know the domain that we work in, and the novelty of this stuff. So I know that there isn't some new corpus of information out there that's training the new model." "So all I can say at that point is that my conversation or our analyses have been used for training. The truth is, it's actually a piece of IP, that's our organizational IP. We don't have any NDA or confidentiality protections, with them being a service provider back to us." "This is why I care a lot about open source, because I don't want them having my chat logs, because they can use them for training to create an IP that is now diffused to the rest of the market. I find this very unfair."

“I think they're having AI psychosis… I'm not sure they're dispassionate critics, observers of what's happening.”

The U.S. House will be in session only 4 full days from July 24th to November 8th. That’s 100 paid days off versus 4 full workings days plus 4 travel (half) days. Meanwhile TRUMP says YOU have too many holidays.

NEW: I am investigating avocado oil products for deceptive marketing and for the potential use of undisclosed seed oils.

This is where Thieseen went to prep school

David @friedberg explains the rise of socialism and the hypocrisy of @RoKhanna: "It's a lot easier to claim an enemy and get elected, than it is to fix a difficult policy problem." "Ro Khanna is the perfect example of this." "This is some dude who has a $300M net worth. His children have memberships at 3 country clubs." "The dude lives in a $10M house in Washington, DC." "And he used to be the moderate Silicon Valley guy who would go around to all the tech CEOs and founders and VCs and be like, 'Give me your money. I'm a moderate. We're going to make the Democratic Party work for you.'" "Then he saw the tides change, and he's like, 'Oh, this is yum, yum. I'm going to run for president in 2028.'" "Just to be clear on this, he is so unlikable, I think he'll likely end up being Secretary of Housing and Urban Development if AOC is elected president." "I do not disagree with him or Gavin or anyone else that tax policy in America is fucked up." "When billionaires borrow against their unrealized capital gains, they should pay capital gains tax. Capital gains tax should be 40%, not 15% or 18% or 20%. In fact, labor should pay less. Income tax should be much lower. Capital gains should be much higher. That would be a fair tax policy." "I don't know a lot of people who are wealthy who will look me in the eye and truly disagree with that." "It's a lot of bullshit when we all say, 'It incentivizes investment.' Great, it incentivizes investment. What are you going to do with your excess capital if the tax rate is 40% versus 20%? You're not gonna invest it now? Of course you're going to fucking invest it." "But instead of those practical solutions like I just laid out, they're all like, 'Let's go fire up the fucking socialist train because it's worked so well in fucking Cuba and Venezuela and the Soviet Union and all these other places.'" "What both parties are missing is a policy agenda that says: Here's the shit we have to do to keep America from going down the socialist death spiral. And they all should be fucking doing it, both sides."

Are the same VCs on this site who claim we must win the AI war with China taking CCP money via LPs? Do CCP members have access to private deal flow that is never made available to American citizens?

Should the average person be more afraid of the American rich, or Maoism?

If you find yourself defending a child molester, saying: “WE can’t be judged by OUR worst day”, you probably need to have your hard drives checked.

What's insane is using a cloud hosted agentic model and expecting that your code never leaves your laptop. Use a local model if this is a priority. They're not as good but it never touches the cloud at all.