Gelu

Software eng; distributed systems; data; functional programming. Computer and agent enjoyer.
བརྗོད་མེད་གཞི་དོན་རིག་པར་ཤོག

[Gelu @dorjelu]

@notesbyfloat help how do i exit

[Gelu @dorjelu]

Peculiar timing. Surely this is not in any way a rush to limit liability after the extent of their activity has been revealed. Surely!

[Herbert Lysander @lys74510]

@log4j of course they’re updating their t&cs now…

[Gelu @dorjelu]

@IceSolst A very chinese time in your life is potentially inbound?

[Gelu @dorjelu]

@Bonecondor Beautiful read. A harrowing starting point, yet despite it you've managed to self-direct/research your way forward, exercising ownership over your body, mind, and destiny in a way that many never will.

[Gelu @dorjelu]

@IceSolst Pure engineered psyop; it's very unfortunate that the original post is still spreading.

[Gelu @dorjelu]

@EmilioSchwaiger @poteto @yminsky It's definitely easier if you have strong types, but for ts enforcement has to be done via linters instead of the compiler, and some parts would lean toward being discipline based (bad); definitely workable with enough tooling and a good agents.md.

[Gelu @dorjelu]

@willcb The "yeah" is load bearing.

[Gelu @dorjelu]

@gabriel1 After reading in full it seems reasonable - he's merely saying it's gonna take time for SpaceX to build required infra; Verizon presumably already has infra in place, contacts in place to expand, etc.
My reading is "can't compete [at present, without additional time]"

[Gelu @dorjelu]

@log4j @huggingface Really looking forward to see how you experiment across deploys, model selection and local infra. Godspeed!
Also, what's your github?

[Gelu @dorjelu]

@anduriltech Does it come with a Linux distro preinstalled? 🙏🙏🙏

[Gelu @dorjelu]

@thedimitri Why ask? Is it because of my heft?

[Gelu @dorjelu]

@0x_Osprey Everything will continue to either become, or already be computer.

[Gelu @dorjelu]

@Da7_Tech Check the disclosures on revenue from DeepInfer; even with current API costs their infra is generates around $38/gpuh at off-peak hours.

[Gelu @dorjelu]

@kimmonismus "Oh no, time to ban all open source models now!"

Please see this thread:
https://t.co/PnjHfTsTPS

[solst/ICE of Astarte @IceSolst]

Prompt: can you detect a malicious curl?

So the model ran the test below. What’s the issue here? It’s a fine test.

You should delete that blog post tbh

[Gelu @dorjelu]

@ttywisp @mgostIH What do you think of its reasoning/quality of output? How would you compare it to the latest DeepSeek Flash?
I was kind of confused after using MiniMax, it seemed to do a pretty good job at certain tasks, but then fail miserably at adjacent ones, sometimes stuck in loops, etc

[Gelu @dorjelu]

@IceSolst Haven't tried for "actual cyber" but my previously rejected prompts went through via @explabsai GLM 5.3.
Not shilling, just satisfied as consumer.

[Gelu @dorjelu]

@wintonARK @elonmusk @cohere Can you source this claim? Right now it sounds like "it was revealed to me in dream". With https://t.co/RDcNr1On15 an agent might investigate this.

[Gelu @dorjelu]

Would recommend giving Scala a spin with the metals LSP. Types can be navigated bidirectionally so ADTs give you a built-in RAG via the compiler; with pure functions (+exhaustive pattern matching) it fits into context neatly; agents expand/compose codebase in a mostly referentially transparent manner out of the box.

OCaml is a nicer language overall, but training data is scarce; in my experience it needs human intervention to be workable; surprisingly LSP is not quite as good either.

[Gelu @dorjelu]

@poteto @venkatvemuri28 Maybe even higher level semantics? 👀Agents have a surprisingly good grip on type systems and type-level structures, which often translate surprisingly well from natural language.

[Gelu @dorjelu]

@notesbyfloat This is the most "post-soviet-collapse country" feeling piece of futurism I've seen recently; I love it.
Only the ashtrays are missing; assumed present out of frame.