Michael Klaput

[Michael Klaput @MichaelKlaput]

@tv0ff @trq212 Yes, actually I've been using through too many model versions... Should try to turn it off and see. I just thought that's what you're going for with your comment. How do you do it?

[Michael Klaput @MichaelKlaput]

@tv0ff @trq212 Try the official superpowers plugin, that's exactly what it does: high level planning -> spec -> detailed implementation plan

[Michael Klaput @MichaelKlaput]

@neural_avb That's their marketing...I thought their blog post was quite honest about it and given how they train the model, it obviously still hallucinates... You just have a deterministic output schema. It's nothing a SOTA model wouldn't manage... But it's a lot cheaper and faster.

[Michael Klaput @MichaelKlaput]

@__soragoto__ 2026: We stopped using vector DBs and just keep linked markdown files around.

[Michael Klaput @MichaelKlaput]

@rustaceans_rs Typst

[Michael Klaput @MichaelKlaput]

@Math_files Personally, I like this way of understanding the connection intuitively:

If you take a sphere and uniformly fill it with gas, then its shadow will have a Gaussian density.

The equation above is for a disk (2d sphere). Though it's true in any dimension.

[Michael Klaput @MichaelKlaput]

@ImLunaHey I'm no expert in TS, but something I ran into when doing the same with JVM as starting point: the JVM is a fixed overhead, so even the smallest process has a fairly large memory size.

[Michael Klaput @MichaelKlaput]

@rynorhn I would love to be able to really take them for their word... Though it's difficult to distinguish their marketing activities from their actually true opinion. Remember when GPT3 was too dangerous to release?

[Michael Klaput @MichaelKlaput]

@bmay Maybe this is proof of P \neq NP?

[Michael Klaput @MichaelKlaput]

@ferlinux Finally the Line is being built

[Michael Klaput @MichaelKlaput]

@ZanthousDev @skdh Do you think AI Labs make money with solving maths problems? If the models are good enough to establish a profitable business model, then scaling further just allows others to catch up.

[Michael Klaput @MichaelKlaput]

@skdh My money would be on (b) ... Since a few years it seemed like an inevitable move to come. You scale until the S curve flattens too hard... Then you push for policy to defend your lead. Mostly I'm surprised how much longer advancement could be pushed. Time for the next S curve.

[Michael Klaput @MichaelKlaput]

@ApFoivosBakis @LinusMixson Since when :D ?

[Michael Klaput @MichaelKlaput]

@mathemetica This actually only scratches the surface. When you make this with in arbitrary dimension and on non-flat surfaces that's when things start to become truly beautiful.

For example Stokes theorem https://t.co/V3BGY4k5Q5

All your integral theorems in one :)

[Michael Klaput @MichaelKlaput]

@Steve_Yegge No one seems to mention zellij so I will ... basically tmux (which was my previous pick), but with a lot better UX for my taste.

[Michael Klaput @MichaelKlaput]

@sama Why didn't you want to offer collaboration to Levent? That would have been a great way to show you care about advancing knowledge and mathematical research. Should it matter that he works for your competitor?

[Michael Klaput @MichaelKlaput]

@Math_files Reading the replies here, maybe today he'd be writing about the omnipresence of the engagement farmers?

Solve this! 1 + x = 23^42

[Michael Klaput @MichaelKlaput]

@gregisenberg Databricks did it long time before AI, starting with Spark. Open source seemed like a very bad idea to me until I listened to one of their founders: Proprietary IP is a fragile foundation and other moats are amplified by open sourcing.

[Michael Klaput @MichaelKlaput]

@Lazylagrangian @Math_files Okay, very fair! Though all of these are from before 1990 and I'd argue he is much more famous for the stuff he did afterwards.
Where he really excelled was pushing the boundaries of mathematical knowledge with physics style reasoning.