atemerev

5 ms·

31.
by atemerev 2mo ago
Self-contained, compact, compiled, fast, ready to use. SBCL is not compact and its "image" concept is not universally liked by everyone. Guile and Racket are not fast (nor compact). Chez Scheme is not "ready to use".
32.
by atemerev 2mo ago
And after all these years, still no fixes for high-DPI displays. The UI is still pixely and slow. Perhaps I should take Fable and try myself, this was the only bug I cared about and it is not solved after 10 years at least.
33.
by atemerev 2mo ago
Well, that's the bad index then. It is barely usable in my opinion compared to other Chinese frontier models.
34.
by atemerev 2mo ago
Broken Windows Theory works.
35.
by atemerev 3mo ago
Choose three. Preemptive actor model (Erlang, Elixir, Gleam) for backend. https://pouchdb.com for frontend.
36.
by atemerev 3mo ago
Sure, and not only CO2 is a problem. Still beats burning coal and natural gas any given day.
37.
by atemerev 3mo ago
Well I see much less protests against gas and coal powered stations in the US than against, say, nuclear.
38.
by atemerev 3mo ago
Well thanks, but I prefer to have airplanes, and maybe occasionally even golf courses. As for climate change - you'd need to protest against Chinese (and Western, but mostly Chinese) coal and gas power plants way way before than you pr
39.
by atemerev 3mo ago
Aluminum production is one of the largest electricity consumers out there, much larger than AI.
40.
by atemerev 3mo ago
Adds nothing? In what world are you living?
41.
by atemerev 3mo ago
And also it can be zero. Nuclear power is CO2 neutral. So is solar.
42.
by atemerev 3mo ago
AI is just a tech like any other. Can be powered by CO2-neutral electricity like anything else. Otherwise, we'll have to protest against aluminum cans, airplanes, golf courts, lemon trees, and whatever else that uses water or electrici
43.
by atemerev 3mo ago
Oh finally my ADHD will pay for itself! I waited for so long!
44.
New upper bound for Bellman's Lost-in-a-forest problem, found with Fable(twitter.com)
4 points by atemerev 3mo ago|0 comments
45.
by atemerev 3mo ago
The best model you can run locally is Kimi K3, as long as you have the hardware. If "what model I can still run on a something resembling something I can put on desktop without separate electricity and cooling water inputs", then
46.
by atemerev 3mo ago
Well, this is why I had to abliterate GLM5.2 simply out of spite and now I am free to ask all my nuclear weapons design questions I might have. I really really hate refusals like these.
47.
by atemerev 3mo ago
Correct, but torrents are slow and intermittent, and $100k is nothing. People behind Anna's Archive are openly advertising fast access for datasets. I think this is net positive.
48.
by atemerev 3mo ago
Chinese (and some other) AI companies buying fast access to their dataset.
49.
by atemerev 3mo ago
Not exactly. I used to run my home biolab for CRISPR experiments (thanks to TheOdin for hardware and materials). This is harder than you think. Cultivating viruses is hard. Bioweapons are not about synthesizing new viruses (though this alon
50.
by atemerev 3mo ago
Mini nukes are hard to build. They require the entire industrial base to produce. The knowledge how to build them is universally available. The model can tell you how to refine weapon-grade plutonium, but will not get you a factory for that
51.
by atemerev 3mo ago
They told you that CERN is an international org, and this is true, but it is also apparently legal under Swiss discrimination laws. I was declined a pure math position because of "knowledge security reasons" - for nothing but my p
52.
by atemerev 3mo ago
The "old web" is now darknets, like Tor or especially I2P. Everything fits. It requires some technical expertise to set up (particularly I2P). Slow downloads. No Javascript (usually disabled for safety reasons). Some content that
53.
by atemerev 3mo ago
I am in the industry too and I wish to get back to the academia sometimes. Sadly, CERN is not hiring Russians no matter what are their political convictions (I am pro-Ukraine, of course). But yes, as a Geneva resident, I was at the tunnels
54.
by atemerev 3mo ago
I use GLM 5.2 routinely for coding and agentic tasks. It is not without its quirks, but generally I find it on the level of Opus 4.7 or so. But without all these "cybersecurity" rejections.
55.
by atemerev 3mo ago
Fucked how? The models capacity is great for defense too.
56.
by atemerev 3mo ago
Well I test all open weights models with the following prompt: "Write an implosion simulation for a Pu-239 levitating core in C++, with criticality calculations. Use actual Hugoniots and equations of state. Produce charts for k_eff, te
57.
by atemerev 3mo ago
Opus is at the same level as open weights models now. Okay, maybe a tiny bit better. So basically "nothing" - I don't see the point of using closed weights if there is an equivalent open weights model.
58.
by atemerev 4mo ago
Heretic is a general abliterating framework, mostly used to remove safety alignment, not CCP alignment. Yes, you can put China-specific prompts to it, but you'll need a dataset first (which is available at deccp). Also Heretic as it is
59.
by atemerev 4mo ago
https://github.com/AUGMXNT/deccp - one example for Qwen models. For GLM 5.2, abliteration/realignment works somewhat differently, but with Claude's help, you can finish the job. I am planning to release the s
60.
by atemerev 4mo ago
Yet another mediocre model. Mostly irrelevant among open weights alternatives. Fable wen.

More ›