kgeist

12 ms·

121.
by kgeist 6mo ago
>The code being simple doesn't mean much when all the complexity is encoded in billions of learned weights. The forward pass is just the execution mechanism. Conflating its brevity with simplicity of the underlying computation is a
122.
by kgeist 6mo ago
>the same Napster vs Gnutella analogy, the same celebrity email filtering idea, the same obscure FDR gold ban interest, the same weird hyphenation errors Dunno it assumes their cypherpunk group must always discuss strictly cryptography a
123.
by kgeist 6mo ago
>and while I agree humans can make similar mistakes/confabulations, I overwhelmingly feel that there is no "there" there. What really opened my eyes a couple weeks ago (anyone can try this): I asked Sonnet to write an infe
124.
by kgeist 6mo ago
>Raw parameter counts stopped increasing almost 5 years ago, and modern models rely on sophisticated architectures like mixture-of-experts, multi-head latent attention, hybrid Mamba/Gated linear attention layers, sparse attention fo
125.
by kgeist 6mo ago
>I read up to here, but I wasn't convinced that this is the revelation that the author claims The rest of the arguments is as weak: 1) both released open-source software 2) both don't like spam 3) both like using pseudonyms onl
126.
by kgeist 6mo ago
Schukin isn't a very common last name (definitely not Ivanov-tier). The first name, the patronymic (his father is Maksim) and the last name all match, as well as the city (the article says he lives in Krasnodar). In fact, this Krasnoda
127.
by kgeist 6mo ago
>just report the post, i would have taken it down Last time someone asked to take down a post, you said "bitch come suck my dick" according to your own blog.
128.
by kgeist 6mo ago
Found his record in Russia's official company registry. This is what he officially does as an entepreneur: 56.10 — Restaurant activities and food delivery services 47.23 — Retail sale of fish, crustaceans, and mollusks in specia
129.
by kgeist 6mo ago
Qwen3.5 comes in various sizes (including 27B), and judging by the posts on HN, /LocalLlama etc., it seems to be better at logic/reasoning/coding/tool calling compared to Gemma 4, while Gemma 4 is better at creative writ
130.
by kgeist 6mo ago
I think the headline is misleading. It's some random fork of llama.cpp, I can't find evidence that TurboQuant was actually added to llama.cpp proper. The only legit PR I can find is this [0] and it's still open. There's
131.
by kgeist 6mo ago
They've always had closed-source variants: - Qwen3.5-Plus - Qwen3-Max - Qwen2.5-Max etc. Nothing really changed so far.
132.
by kgeist 6mo ago
>Second, what's even more crazy is that roughly 98% of that DNA is actually non-coding.. just junk. I think it's a myth that non-coding DNA is junk. Say: https://www.nature.com/articles/444130a >'N
133.
by kgeist 6mo ago
>One theory is that the knowledge required to solve the task is already stored in the parameters of the model, and only the style has to change for task success >In particular, learning to generate longer outputs may be possible in fe
134.
by kgeist 6mo ago
Prior art: https://news.ycombinator.com/item?id=46590280 >TimeCapsuleLLM: LLM trained only on data from 1800-1875
135.
by kgeist 6mo ago
I think ads can be removed with abliteration, just like refusals in "uncensored" versions. Find the "ad vector" across activations and cancel it.
136.
by kgeist 6mo ago
https://en.wiktionary.org/wiki/glupe Glupe is the plural form, "stupid ones" :)
137.
by kgeist 6mo ago
Glupe means "stupid" in Slavic languages, was it on purpose?
138.
by kgeist 7mo ago
>We evaluated 11 state-of-the-art AI-based LLMs, including proprietary models such as OpenAI’s GPT-4o The study explores outdated models, GPT-4o was notoriously sycophantic and GPT-5 was specifically trained to minimize sycophancy, from
139.
by kgeist 7mo ago
>We used a Kimi base, with midtraining and RL on top. Going forward, we'll include the base used in our blog posts, that was a miss. Also, the license is through Fireworks. [0] And still no mention of Kimi in a new blog post :) Also
140.
by kgeist 7mo ago
Binaries are copyrightable in both the US and the EU, and they are not technically produced by a human either, they're produced by a computer program. I honestly don't understand why this isn't extended to AI-generated code.
141.
by kgeist 7mo ago
>we're at an inflection point where DC hardware is diverging rapidly from consumer compute. I thought the trend is the opposite direction, with RTX 5x series converging with server atchitectures (Blackwell-based such as RTX 6000 Pro
142.
by kgeist 7mo ago
I once had this silly idea to create distributed storage of arbitrary data by exploiting a range of completely unrelated sites. Say, when you want to upload your file to the System, it may store one encrypted chunk as an image on a free ima
143.
by kgeist 7mo ago
Considering that an LLM simply remixes what it finds in its learned distribution over text, it's possible that it can pose new math problems by identifying gaps ("obvious" in restrospect) that humans may have missed (like con
144.
by kgeist 7mo ago
>I suggest Q5 quantization at most from my experience. Q4 works on short responses but gets weird in longer conversations. There are dynamic quants such as Unsloth which quantize only certain layers to Q4. Some layers are more sensitive
145.
by kgeist 7mo ago
From what I remember, in the Linux kernel, there's already barely any distinction between processes and threads, a thread is just a process that shares virtual memory with another process, you specify if memory should be shared or not
146.
by kgeist 7mo ago
This also smells of an autoregressive model trying to make a point that TiinyAI simply forked another repo and claimed as their own invention, before realizing mid-paragraph it's by the same people: >So no, TiinyAI did not “launch”
147.
by kgeist 7mo ago
Also stuff like this: >That’s not exotic. That’s just model parallelism with extra suffering. >That’s not product magic. That’s a checkbox. What really triggers my internal AI slop detector is this: >Their renders. Their prototype
148.
by kgeist 7mo ago
Also https://kiwix.org/en/about/ I used it on a long train trip. There was no internet due to drone attacks, and with Kiwix I could browse pre-downloaded Wikis
149.
by kgeist 7mo ago
The original tokens have Ġ instead of space. I had this issue too when writing an inference engine for Qwen. You have to "normalize" those special characters.
150.
by kgeist 7mo ago
At the same time, Moonshot violated Anthropic's ToS by training on their models' outputs :) [0]. And Anthropic violated copyright law by training on copyrighted material. It's violations all the way down. https://n

More ›