Alex Coventry

@sama "It is important that the democratic process remains more powerful than companies." Great, so you're going to back off from the military contract which effectively commits OpenAI to serving as the IT Dept. for Trump's secret police? I would love to use ChatGPT/codex again!

I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is: blog.samaltman.com/2279512

@dradambryan @ianbremmer @newdreamery Unfortunately, Iran has several thousand cards, hidden in the mountains of Iran, ready to strike any vessel which runs its blockade of the Strait. And AFAIK, Trump has no plan to mitigate that risk, aside from implausible threats of genocide.

@ianbremmer @newdreamery It's unpopular here bc the left and the MSM will NEVER give Trump his due. He's met every military objective with nearly no casualties - most successful military action maybe ever. He holds all the cards.

iran’s nuclear & ballistic missile programs were a real threat. more to israel and the gulf states than than to the united states, but real nonetheless. responding to that threat with an all out regime change war is one of the most irresponsible foreign policy decisions the united states has made in my lifetime. we will suffer the consequences of this mistake for years for years to come.

@eliebakouch @trq212 Yeah, you can vibecode this feature in like five minutes. I already get claude to save plans as markdown files and hit an http service to display them in a browser external to the sandbox I run claude in.

@trq212 the "ultra" prefix is very misleading if it's just plan mode but with a different UI

New in Claude Code: /ultraplan Claude builds an implementation plan for you on the web. You can read it and edit it, then run the plan on the web or back in your terminal. Available now in preview for all users with CC on the web enabled.

@forgebitz Gemini is very smart, but undisciplined. You have to use it carefully. It's good for ideation and critique, at least.

i don't know a single developer using gemini models

@rasbt @steipete It's out of line with their AI-safety mission, and OpenClaw effectively commoditizes their AI services, which is bad for business.

@steipete Whoa, why are they now blocking API users? Is the API also so heavily subsidized that they are losing money on that?

Yeah folks, it's gonna be harder in the future to ensure OpenClaw still works with Anthropic models.

@Windspiriteagle @ianbremmer Assessing a potential impending genocide through a financial lens is heartless and soulless.

@ianbremmer The show and channel is about investing. The biggest mover of markets is by far Trump's capricious leadership rather than fundamentals. I get what you're saying, but it is what it is right now.

polymarket in 12 seconds

@lateinteraction "Claude, write some software which will fill the whole in my heart and give meaning to my life."

I think it really helps to say, suppose Claude Code is Einstein x N for N areas you care about. Don’t you still need to know what you actually want to achieve before you hire him? Don't you still need basic separation of concerns? Or else that Einstein is just going to do what he thinks is best. Which is the outcome that 10M other users are getting at the same time as you, so it has $0 value. It’s like that moment when 10,000 people decided to run Karpathy’s autoresearch for nanochat in exactly the same configuration at the same time with the same model lol. At its best, it’s a lottery. But it’s a low-temperature lottery, where everyone gets the same ticket numbers. An even more stupid form of gambling.

For the last few months, I've begun to increasingly dislike the vibe I see in tpot around AI. Nuance and substance are more absent than ever, and are replaced by shallow hype, including now by many of your largest favorite accounts in this space. in case it needs to be a meme:

@ryangrim Not everyone. There will still be MAGA cult members who believe he cannot fail, he can only be failed.

And everyone will understand who is to blame for this outcome

@anneapplebaum There are many parallels between Trump and Herod. His threat of genocide has proven that if he thought a boy could threaten his power, he is capable of killing millions of children to protect himself.

The Trump administration gave a Vatican envoy "a bitter lecture warning that the United States has the military power to do whatever it wants—and that the Church had better take its side"

This is nuts, Trump really did do exactly what Ryan Grim suggested he do hours earlier: pretend Iran's ten-point plan is a new proposal, counting on the fact that the media hasn't been reporting on Iran's demands to make it look like a new offer Tehran put forward in desperation.

@iamphredriquee @mark_slapinski Sort of weird to bring up civil liberties in Iran right after Trump threatened to genocide the nation...

@iamphredriquee @mark_slapinski Did you miss the part where Trump told the Iranian people to rise up against that regime and help is on the way, and the Iranian people rose up, and gave the regime the opportunity to kill its most enthusiastic opponents as Trump failed to make good on his promise?

I'm genuinely confused What the fuck was the point of the Iran war?

@Hebejsjebejebee @SkinnyTuna @grok LOL.

@SkinnyTuna @grok is this true?

its cool how everyone who uses AI is basically completely lobotomized now

@casper_hansen_ Coding agents are currently bad at optimization of GPU kernels. They can get the math right, but are often 100x slower than human-written kernels. This talk says the training data for GPU kernels is thin, and does not contain the optimization trajectory. nvidia.com/en-us/on-deman…

If the new model is so great, shouldn’t it be able to recursively optimize its own inference code to make it feasible to launch?

@SarcasticCupcak @emptywheel Iran's not really conceding anything, here. They're still going to toll the Strait. They're not taking Trump's threats of genocide seriously, with good reason. He can't do that and retain power in the US, and he should have accounted for that. theguardian.com/world/2026/apr…

@WanderCoventry @emptywheel How so?

So Russia and China forced Trump to capitulate. Humiliating.

@EvanOwen @willccbb Well, they didn't want to lose access. The SCR designation was purely punitive, because Anthropic told them to pound sand when Hegseth demanded they abandon their explicit red lines.

@WanderCoventry @willccbb Maybe. Though somehow I doubt the admin would’ve wanted to lose access if they knew how powerful it was. From what I hear, nobody outside Anthropic had access prior to a few weeks ago.

or Anthropic, that works too

@Matthewagi @nrehiew_ Interesting. Is that documented anywhere?

@nrehiew_ In addition to everyone saying size they apparently improved the attention mechanism

So what do we think happened here? What was the research breakthrough? Mythos is SOTA on everything?

@EvanOwen @willccbb I think Mythos might have been what brought Anthropic's conflict with the Trump administration to a head.

@willccbb except Anthropic is banned

@SarcasticCupcak @emptywheel So am I, but this totally an own-goal by Trump which is going to severely damage US military credibility and political influence, on a par with the damage the Suez Canal incident caused the UK.

@emptywheel Sorry, but I'm glad that innocent civilians are not dying tonight.

@RayFernando1337 @joeljewitt A cynical satire of Anthropic's responsible, well-justified decision.

@joeljewitt It’s satire meant for X. lol

Project Glasswing FAQ: Q: Why only 12 companies? A: They're the ones who can afford us. Q: What about open-source maintainers? A: We found bugs in their code. You're welcome. Q: Will you release the tool publicly? A: We said "cybersecurity is the security of our society." We didn't say which society.