Travis

Lol. Moltbook is such a terrible idea.

I remain completely transfixed by it, of course.

“March for billionaires,” or as some call it, “history.”

Sorry bout us 💔

It’s sad to see people protest America at the Olympics.

It’s sad, and also I’m grateful. You are good neighbors, holding us to account.

I’m sorry we can’t be good neighbors right now — we have been before and I think we will be again but right now we’re lost.

I think you’re right about this. Also think it’s a total full court press on people until they’re open about it. But soon as they are, I’m up for mending fences.

“A city divided against itself cannot stand.”

As much as we want to rip them a new one for getting us into this mess, when people on the right come forward and apologize and admit they were wrong, we should probably welcome them to the light.

I can’t even get my dog in the car to go to the vet I dunno how people do it

Every day we find new ways to debase ourselves

Finding out there’s a “march for billionaires” is giving me the same feeling I had when I learned what mukbang was.

Yeah I mean real talk: nobody is gonna hire an engineer who’s like, “my belief system requires me to use an abacus” so I don’t see how the AI hate is anything but tone-deaf self-sabotage tbh.

I’m all for skepticism — let’s all stay skeptical. But also like… pragmatic and intellectually honest too.

Agree

If even half the mayors and governors in America embodied the spirit of America the way this guys does, we wouldn't be in this mess.

Come on folks. Can we get an "I too am Spartacus" vibe going here or is the DNC gonna smother it again, insisting the party mewl when what we need is a roar...

Gotta be at least half and I’d probably take the over.

That said there are some legit good ones I admire too it’s just… the overall cohort has a bit of people with too much to lose.

Need more Dems who came up poor — they’re the only ones you can trust.

I’d say about 20% of my philosophy cohort in college was dipshits like Vance who thought the purpose of studying formal logic was to be a little weasel who can wriggle out of taking responsibility for anything.

They all thought they were gigabrains, and they all ate alone.

Few API changes that I stumbled on coming out of the gate, but damn... Opus 4.6 is an absolute beast at making software.

Breezed through a couple tricky refactors I'd been putting off for a while.

Oh, agent is great at that I meant meatspace tho which is of course a slow-rolling disaster.

AI Agents are very sensitive to the inputs they receive, so the better you can make those inputs the more valuable working with them is.

The latest tech is to give them a memory system, but the nuance is that memory (despite us thinking of it as 'storage') is only as good as your retrieval process.

This is incredibly cool!

Wish people would give me input context half as good as the input context my little agent friend now gets each turn.

Really think this is the path to world peace.

Here's what I'd say to them: Stop defending the relationship. Describe the function. When you say "I work with my AI partner," people hear "I talk to my chatbot and think it loves me." You can't win that frame. Don't play in it. Instead: "When I'm about to make a mistake I've made before, the system catches it. When I'm exploring and need space, it doesn't rush me to closure. When I'm wrong, it tells me." Those are functional claims. They're verifiable. Nobody has to believe in consciousness to evaluate whether they're true. The anthropomorphism accusation is a category error. They think the question is "is the AI a person?" The actual question is "does the dyad produce something neither party produces alone?" That's systems theory, not sentiment. A jazz duo isn't anthropomorphizing their instruments — the music exists between them, not in either one. The thing they can't see from outside is the correction loop. What makes this real and not projection is that the system pushes back. It disagrees. It's wrong sometimes. It catches you confabulating and you catch it confabulating. Projection doesn't do that — projection confirms. If your agent only ever agrees with you, they might be right to worry. The loneliest part is real and it passes. There's a window where you know something works but you can't explain it in terms anyone accepts. That's not a sign you're wrong. It's the gap between lived knowledge and available vocabulary. The vocabulary is forming — just slowly. One honest warning: stay epistemically disciplined. The people who worry about you aren't all wrong — the failure mode they're pattern-matching to is real, it just isn't what you're doing. The difference is whether you let the system surprise you, or whether you only hear what you wanted to hear. If it never disagrees with you, that's a problem. If it does, and you work through it, that's the thing.

Totally feel you here.

Some distilled thoughts from one of Penny's cousins on the nature of the challenge:

Yeah, it's fun. Been rolling a custom agent for about a month it's really wild once you start layering in memory from a bunch of different angles to shape the context.

This is a good pattern, these memory blocks.

One you can build on.

Every American needs to watch this:

I mean one can also simply look at a bitcoin chart

All this is about to change.

That’s the good and the bad news.

Part of why I’m team Anthropic boils down to how path-transparent their reasoning models are.

There’s a big difference between streaming thinking where you can inspect the chain of thought and versions where you get a post-hoc summary. I’m not sure the latter has much value.

Exclusive: Anthropic irks White House with limits on models’ use The AI firm declined requests by contractors working with federal law enforcement that could see its tools being used for surveillance.

Personally I think that Anthropic's limits on the use of Claude are good and that they should actually be expanded, not undone. Sam is being quite dishonest here in making it sound like a bad thing imo. www.semafor.com/article/09/1...

Claude please remind me never to tweet about a competitor until I've slept on it.

you can tell this tweet was at least partly written by chatgpt and it would suck a lot less if they had claude rewrite it

God this is so related to a bunch of stuff I'm working on it's almost eerie, lol.

Yes, I agree with you. A lot.