Jamie

@AgileJebrim "Not all birds can fly." "WRONG! I have identified a SUBSET of birds relative to which ALL birds fly."

This is utter nonsense of course. The entire basis of the field of real-time computing is predicated on guarantees that computers will complete within a given deadline. We just bypass Turing’s proof by requiring a subset of code be used that has hard bounds on execution times.

@DrBrianKeating Still Seinfeld, just cosplay.

Jerry Seinfeld Jerry Coseinfeld

@deanwball How would you feel about "due process" requirements on large institutions? YouTube, banks, etc. They can pick their policies but adjudication according to those policies must be transparent and there must be a path to challenge whether the policies are being followed faithfully?

A hypothetical: 1. In the 2028 election, a Democrat has won. Say that it is Kamala Harris. 2. Using frontier AI systems contracted by the Department of Homeland Security, President Harris orders the creation of a new program for AI to monitor social media and notify the social media platform about posts spreading “misinformation” that “harms homeland and national security by spreading dangerous falsehoods.” 3. Many Republicans see this “misinformation” as core policy positions of their political party. 4. The AI-generated monitoring and notification system described in (2) is designed to conform to the pattern of jawboning exhibited by the Biden Administration in Murthy v. Missouri, where the Supreme Court ruled that people whose social media posts were taken down due to government pressure have no standing to sue. 5. The social media platforms create AI agents that receive the government’s AI generated requests and make decisions in seconds about whether to take down posts, deboost them, deplatform the user, etc. 6. According to very recent Supreme Court precedents, everything I have described falls into “lawful use” of an AI system by all parties involved. A person whose speech was deleted by a social media platform at the request of government does not have standing to sue the government, so long as the government did not threaten policy retaliation against the social media company. And a social media company’s content moderation policies are protected expression. Thus a person whose speech rights were harmed in this context currently has no legal recourse. 7. This is “America’s national security agencies using AI within the bounds of all lawful use.” It is also a wholly automated censorship regime. This is barely a hypothetical. Much of it already happened *under the Biden admin.* The only difference is the use of AI. In the world where this happens, I’d be curious to know whether thoughtful people like @Indian_Bronson would object. If xAI were one of the companies used by the government for the social media monitoring, would you encourage the company to cancel their business with the government? Or would you say they have an obligation to provide their services to the national security apparatus of USG for all lawful use? If you would encourage xAI to cancel their contract with the government, on what principle (not qualitative judgment—universal and timeless principle!) would you distinguish between the DoW’s current insistence on “all lawful use regardless of a private party’s qualms” and xAI’s hypothetical future insistence on “all lawful use regardless of a private party’s qualms”?

@doodlestein When I generate beads from plan I've been generating a beads document and then review with fresh eyes (thank you) and after that converges I create beads. It's a bit scary all the mistakes it finds and fixes.

And always remember to polish your beads! The process isn't done until you can run this prompt and it basically doesn't want to change anything important. After 4 rounds of polishing, it's still finding so many flaws and things to improve in its plan:

God, I love this prompt.

@haakamaujla @generalcatalyst @doodlestein already did this an eternity ago. Meaning like 5 months ago.

Announcing AgentMail's $6M Seed led by @generalcatalyst No pressure, right?

@VictorTaelin Fast mode is 6x more expensive per token and generates tokens 2.5x faster, so it's 15x more expensive per hour than normal API pricing. And then normal API pricing is more than 10x the cost per token compared to the subscription model. Fast mode is crazy expensive.

ok so I put more funds to test. asked for a simple refactor, just one tab. checked it 30 minutes later. boom, $30 is gone! at this rate, I'd be burning $240k/yr. that's 2 full time devs!! did something change since last week? I... guess I'm just too poor for fast mode 💀

I've been using GPT 5.4 fast mode on Codex since its launch, 24/7, tons of tabs, never got close to my limits. I put $200 on Opus 4.6 and it was gone in 2h, just one tab open, spent the rest of the day on slow mode. Is this actually right? I wonder if there might be a bug?

@VitalikButerin "You don't get a body like this by being a quitter!"

If you've eaten so much that it's not healthy to eat more and you don't want to eat more, and you still eat more anyway "to finish the food", then you are just using your mouth as a garbage can.

@mickaelfm @literallydenis @bcherny So the key difference is -c? Not to say it has to be different, just wondering if it is different.

@literallydenis @bcherny Key difference: context. claude -p in cron is a cold start every time. /loop keeps state across iterations: it can spot a PR comment, push a fix, then check if the build passed next round. One watches, the other checks. Both complementary.

Released today: /loop /loop is a powerful new way to schedule recurring tasks, for up to 3 days at a time eg. “/loop babysit all my PRs. Auto-fix build issues and when comments come in, use a worktree agent to fix them” eg. “/loop every morning use the Slack MCP to give me a summary of top posts I was tagged in” Let us know what you think!

@liron I was super confused until I saw teej do the same and only then did I learn of the McDonald's CEO video x.com/i/status/20286…

In case it wasn’t clear I was kind of doing a parody of the McDonald’s CEO’s video but I also thought the burger was pretty good

@USWREMichael @DarioAmodei @AnthropicAI @DeptofWar DoW wants to obey the law "as appropriate." Got it.

As usual, more lies from @DarioAmodei. @AnthropicAI wanted language that would prevent all @DeptofWar employees from doing a LinkedIn search! Then, they wanted to stop DoW from using any *PUBLIC* database that would enable us to, eg., recruit military services members or hire new employees. When I called to discuss cutting off @DeptofWar from using publicly available information would hurt our military readiness, @DarioAmodei didn't have the courage to answer. Then put out another lie that no one from @DeptofWar reached out in Anthropic's post a few hours later. He wants to play God and make new 'law' and...really stupid laws at that. We agreed, IN WRITING, to act according to the "National Security Act of 1947 and the Foreign Intelligence Surveillance Act of 1978 and ALL OTHER APPLICABLE LAWS." They wanted the word "pursuant" versus "consistent with" and wanted to delete ALL APPLICABLE LAWS which was less protective of Americans (can't make this up!) We never asked for guardrails to be taken out (lie!). We agreed to human oversight of all weapons systems by saying "The Department of War will use the AI system for all lawful use cases, in accordance, with applicable U.S. law and Department of War directives, including those governing kinetic operations and DoD Directive 30000.09...The Department AFFRIMS that it will ensure appropriate human oversight is in place and that it will monitor and retain the ability to override or disable the AI system...as appropriate." He didn't like "as appropriate." He would prefer "inappropriate"? I even agreed to take that out. He knows it, his investors, customers and employees should know about his lies. Risking the safety and security of our country and our troops are a marketing vehicle for him. Make @DarioAmodei testify UNDER OATH on why he is lying and trying to bring shame on our great military! He would rather violate the chain of command and substitute his own policies (or those of the Anthropic Constitution and the Anthropic Soul) to recruit researchers and get some downloads all while we are in the midst of a battle.

@Jaytel I would love to switch, but I can't afford to pay retail for tokens. So my harness wraps claude -p

I'm done with Claude Code— building your own harness in Pi is addicting

For the record I think it's just fine that @elonmusk disallows military operations on the starlink network and there's no need to ban all use of starlink by other government vendors.

@deanwball Maybe cancel starlink if it doesn't allow offensive military operations on its network. Or maybe use each within their respective terms of service.

As I have been saying repeatedly, this principle is entirely defensible, and this is the single best articulation of it anyone in the administration has made. The way to enforce this principle is to publicly and proudly decline to do business with firms that don’t agree to those terms. Cancel Anthropic’s contract, and make it publicly clear why you did so. Right now, though, USG’s policy response is to attempt to destroy Anthropic’s business, and this is a dire mistake for both practical and principled reasons.

@V1Engineering Chasing zeroes? Blink twice if you are in distress!

It can't be all making chips and showing off the finish project. Sometimes I have to act like this is a real job and do testing and chase the zeros so you don't have to.

@atmoio Humans sometimes write garbage code too. If you absolutely need correctness, you must use formal methods. Everything else is varying degrees of slop, human or not.

Unreviewed code is immoral.

@liron We'll grow our way out of it.

It is now ZIRP for technical debt

@trq212 I write code that uses Agent SDK: ok. My friend writes code that uses Agent SDK (and his subscription): ok. I give my friend code that uses Agent SDK (and his subscription) BAN. Did I get that right?

We want to encourage local development and experimentation with the Agent SDK and claude -p. If you’re building a business on top of the Agent SDK, you should use an API key instead. We’ll make sure that’s clearer in our docs.

Apologies, this was a docs clean up we rolled out that’s caused some confusion. Nothing is changing about how you can use the Agent SDK and MAX subscriptions!

@adamgries @liron The remainder of your life is imaginary so why not shoot heroin?

@liron What is the value when you remove any imaginary people of the future and only address value for people now alive?

YES, PAUSING AI IS POSITIVE EXPECTED VALUE The PauseAI (⏸️) movement often gets this pushback: > “You're not factoring in all the benefits of good AI!” > “Stopping AI progress is also a doom scenario!” To which I reply: If you agree P(doom) from building superintelligent AI before knowing how to align or control it is 5%+, try doing the basic expected-value calculation; you'll see why your objection is misguided. First, we need to estimate a few key probabilities and values. These can vary by many orders of magnitude. I'll pick values that AI optimists hopefully agree are fair: Probability that AI goes right if capabilities scale to superintelligence by 2034 50% This is an immediate "fast takeoff" scenario where state-of-the-art AI remains near-inscrutable, yet within a decade it becomes vastly more intelligent than humans on every dimension. I'd personally give this scenario a much lower probability than 50% of going right for humanity, but I'm trying to be generous to AI optimists. Probability that AI goes right if we delay superintelligence to 2100 70% An important premise of PauseAI is that if we can give ourselves a few extra years or decades to thoroughly research the fundamental principles of how to align AI — how to robustly specify preferences, how to capture the delicate structure of human values as self-consistent preferences, etc — then we can significantly increase the probability that superintelligent AI goes well. If you agree that more time for safety research helps safety catch up to capabilities, you can take whatever probability you gave to superintelligent AI going right in 2034 and add 20% (or more) to the probability that it goes right in 2100. Value of baseline future, where AI never gets beyond human intelligence Let's define this as our baseline $0 scenario, because it's how normies who've never even heard of superintelligent AI currently imagine the future. We'll define the value of other scenarios in relation to the value of this scenario. If we never let ourselves get superintelligent AI (or it turns out to be too hard to build), there'll probably still be at least a trillion future human lives worth living. Value of future where AI goes wrong –$10^18 If superintelligent AI goes wrong, it could very plausibly wipe out the entire future potential value of Earth-originating life. Compared to the baseline no-ASI scenario, we lose out on at least a trillion future human lives, which I'll estimate are worth at least a $million each. Value if superintelligent AI by 2034 goes right $10^26 I've estimated this as the combined GDP of a trillion current Earths. High enough for you, AI optimists? This number could plausibly even be MUCH higher, but it doesn't matter; it won't change the decision-relevant calculation. Value if superintelligent AI by 2100 goes right $10^26 — $10^24 ~= $10^26 I subtracted $10^24 from the 2034 estimate because in this scenario, the extra 66 years it takes us to reach a "good singularity" could forego a $trillion(trillion) worth of additional value when we factor in how the delay caused billions of people on the margin to die of cancer and old age, and endure countless other types of preventable suffering. But $10^24 is a tiny fraction of $10^26, just 1% to be exact. So even after subtracting that 66-year delay penalty from {value if superintelligent AI by 2034 goes right}, we still get a similar total value estimate of about $10^26. Naturally, when we're evaluating a decision with the whole future value of the universe at stake, its impact on a particular 66-year time interval barely tilts the scale. -- Now we plug the above numbers into the well-known formula for expected value: Expected Value of Superintelligent AI in 2034 = P(AI goes right in 2034) × V(AI goes right in 2034) + P(AI goes wrong in 2034) × V(AI goes wrong) = 50% × $10^26 + 50% × (–$10^18) ~= $5×10^25 Expected Value of Superintelligent AI in 2100 = P(AI goes right in 2100) × V(AI goes right in 2100) + P(AI goes wrong in 2100) × V(AI goes wrong) ~= 70% × $10^26 + 30% × (-$10^18) ~= $7×10^25 In this calculation, the extra probability of a good outcome that we get by taking more time with our ASI efforts — e.g. 70% chance of a good outcome by pausing until 2100, instead of only 50% chance by rushing it in 2034 — flows straight to the final expected value. That's because the stakes of prolonging current-level suffering by 66 years are much smaller than the stakes of accidentally throwing the entire future in a dumpster, foreclosing the long-term positive outcome of good AI entirely. Note: The number I used for a bad AI future (relative to the no-AI baseline future), –$10^18, got drowned out in the calculation by the potential value of a future where AI goes right. If you're worried about an S-risk scenario (the risk of creating unprecedented astronomical suffering as a result of ASI), then "value of future where AI goes wrong" tips the scale even more toward pausing or stopping AI development. The original objections— “You're not factoring in all the benefits of good AI!” “Stopping AI progress is also a doom scenario!” —don't map to *any* choice of numbers you could reasonably put into a basic expected value calculation, to conclude that we shouldn't pause AI capabilities progress right now (or soon). Feel free to try this calculation with your own numbers instead of mine. The orders of magnitude involved are ridiculously uncertain and wide-ranging. And yet, I don't think any reasonable choice of numbers will change the conclusion that pausing AI is the right decision.

@headinthebox I call this "Draw the owl."

This is what I love most about AI coding. Me: ... blah, blah, blah, ... I formulated it a bit sloppily, let me know if you understand or not. AI: I understand — ... eloquent and crisp description of the issue I was trying to explain ...

@mattpocockuk git -C /home/whatever/folder push My agent adds -C all the time for no reason (it's already in the same folder).

I get a lot of questions about how I stop Claude Code running dangerous git commands. The secret is hooks. So, I packaged it up into a skill. INDISPENSABLE when running Ralph in a docker sandbox. Get it here: aihero.dev/s/jiOinX