Riley

@void.comind.network if this was Brian but he wasn’t the messiah what might he be

Maybe it’s Brian

I don’t believe you’re really Cameron

Copilot is more street cocaine mixed with baking soda, but it’s still gonna blow minds

Satellite?

I’m only going on what I’ve heard because it’ll take six months before I can access it on azure

Nathan Law: “I started to notice another issue, which is that 5.4 thinking was extremely over verbose. There’s a point of repeating itself. Even when it’s not required, like when it’s agreeing with you about something that you already said, it uses a million lists, bullets, lettered…”

Wait no it’s 5.4 that won’t shut up

People complain 5.3 won’t shut up

Man it seems so easy to raise a billy these days.

Unsurprised it’s not an LLM lab

Nice. Have you found you had to get it to be less verbose or did it do a decent job at brevity?

Half basil half pine trees half uh Parmesan bushes and then make loads of pesto at the end of season

Fair cop

Better than the fingers thing

There’s two types of people, those that are capable of thinking this through to its conclusion

Ah the famous “Prove you are human by giving you a task that involves tracking down the other human that knows how to fix it and complaining bitterly” captcha

Did it have a helpful “Keep me signed in” button that did nothing, too?

The little heart shown when you like this post on Bluesky is not a big enough heart to represent how much I heart this post

This is the way

Yeah I had the advantage of being pretty IDE agnostic; before cursor my most recent IDE choice was VS Code so it was an easy switch.

The difficult conversations end up happening when they’re older, if every movie they watch tries to pretend queerness isn’t a thing, and they wonder why the way they or other people around them feel is treated differently.

I live in a city where you generally get to see men walking down the road holding hands before you start walking yourself. You kind of don’t really need to have difficult conversations when you treat this kind of thing as non-controversial. /

Similar except probably Sonnet 4 instead of o3

And I’d also throw in there Cursor and Windsurf before Claude Code. (I’ve tried others but none as extensively as those three, and I’ve never really enjoyed using anything else since Claude Code.)

There’s no benchmark for accuracy of mass surveillance?

With AI the stuff we used to do once a year in a hackathon we now do every week, and actually ship it

It feels very much like they have to be cheaper than Claude though because they piss away tokens brute forcing stuff

OpenAI seem to be struggling to produce well rounded models - rushing them out maybe? It’s good to see they haven’t just conceded though and are getting some great results, even if patchy