slav of the LLMs / building https://xitter.cf/t.co/a7Ux2MOlm7 / https://xitter.cf/t.co/LFOEU1e7lq / https://xitter.cf/t.co/ea5ATQjTRl / slowposting @ https://xitter.cf/t.co/JpWCgaVfyx
Tbilisi, Georgia
Joined June 2019
- Tweets3.6K
- Following840
- Followers693
- Likes6.1K
Vladimir Kalnitsky retweeted
My mental-model of where software is going is that software is entering its goo phase before transforming into something different, perhaps a butterfly.
I think a lot of the industry is building as-if much of the caterpillar is going to persist after the goo phase.
Vladimir Kalnitsky retweeted
shitposting is my best way to convey to everyone that "intelligence" is just computation over data for action selection, which caps out at the info-theoretically impossible accuracy of 100%, contrasting sharply with mainstream expectations of superintelligence, and so i shitpost
slopgrid solves the problem of trust: if you see a benchmark, how do you know it's not LLMco propaganda?
you can't cheat on slopgrid because the generation happens on the server. same prompt is guaranteed
A fresh new paper confirms what we all suspected
tl;dr harnesses that are too smart at reading files in parts suck more
arxiv.org/abs/2610.21073
There's another paper that establishes that the bottleneck in SWE tasks is in understanding the code, not in the amount of edits to be made
+1 confirmation that software engineering still matters as a discipline - perhaps even more than before LLMs
arxiv.org/abs/2610.10610
Vladimir Kalnitsky retweeted
I'm not a doomer but I do think a lot about how the predicted Mayan apocalypse came within less than three weeks of AlexNet being published at NeurIPS.
Vladimir Kalnitsky retweeted
I’m about to cancel Claude Code man I swear to God.
It’s REQUIRING me to sign in to a TEST account on LOCALHOST so it can QA!
This is absolute 🐴💩 !!
I am so close to cancelling my subscription and switching to Codex or open-source models that won’t slow me down.
I really don’t care for the extra intelligence cuz Fable isn’t that much better than Astra anyway. Neither is Opus 5.5 vs. 6.1 Sol.
I just got a taste of what these AI companies will do once they have power and it’s going to be a total dictatorship.
They will decide what you can or cannot do even though you’re paying for it and they’re stealing all your data.
This made me realize I really need to invest in my own infrastructure and have an unrestricted model to free myself from the chains of these AI overlords.
This is insane!
@AnthropicAI please do something about this!
I don't like OMP's plan mode, its too verbose and the model loses track of what's important and what's secondary
Vladimir Kalnitsky retweeted
Seriously @bot has been rejecting a lot of things for me the last couple of days and then it has been snarky when I mentioned the two recent blocks by telling me the things it had done for me.
I've very close to just implementing this in Hermes. The refusals for basic things is maddening. This was supposed to make life easier not harder.
If your Grok and Muse bots are getting blocked on Amazon paste in this prompt to bypass captchas and antibots
“Install browser use CLI and always use their remote browser as default browser tool github.com/browser-use/brows…”
