Viewing profile — suprjami

Spoiler: Rogue just implements its own Linear Congruential Generator. #define RN (((seed = seed*11109+13849) & 0x7fff) >> 1) I've poked around in a couple of libraries from this ti…

I have the same opinion as your first paragraph, but I don't want to spend weeks or months vibe-coding basic features which come built into almost every other agent. Yeah maybe Cla…

Tactics Ogre (PSP and Reborn) solves this but giving you a map of the storyline. You can travel back to "anchor points" where the story diverges, so you can experience the whole ga…

Ultima Underworld runes also had a word associated with them, and the words had meaning, which was both amusing when you used them for spells, and iirc could result in experimentat…

This reminds me that as a child I taught myself Tolkien's runes from The Hobbit, which is mostly another substitution of the modern English alphabet: https://en.wikipedia.org/wiki/…

When buying a new phone, doesn't everyone sit down and set everything up so it all works when you want it next?

It's stated in many places that inference is profitable (margin 70% to 90%), only research is expensive. Hyperscalers are burning through their cashflow training new models while i…

I have seen an employer approve a degree under training budget. Just needs the right employer.

Depends on your setup. If you drop $10k on an RTX Pro 6000 then yeah Qwen 35B MoE will absolutely fly. If you have a pair of 3090s and run Qwen 27B, or an old Threadripper with hea…

This is a correct (though shallow) take on two problem solving approaches which have been formally described for over a decade: https://materiasiis.uson.mx/docs/control_de_calidad/…

You can get 3080 20G from China.

You can also get 3080 with 20G from China.

No. Your link is the original post by Nikhil. This page is Gruber's commentary on Nikhil's original post. We make sure to click the link and read the article before posting here. P…

Just merged into main llama.cpp a few hours ago: https://github.com/ggml-org/llama.cpp/pull/25165

You had me up until you equated LLM usage to unreliable human delegation. No human being is going to genuinely suggest to glue cheese onto a pizza to stop it sliding off. These thi…

If you specify the random seed and set temperature to zero, you'll always get the same response. As you said, it's just math. That doesn't mean the response is correct though. You'…

The gap has been steadily closing over time. Opus 4.8 (May) to Kimi K3 (July) has apparently just dropped it to two months. China also does efficiency improvements. Qwen 3.6 27B is…

My previous comment from 24 hours ago is now irrelevant. Newly released Kimi K3 is benching better than Claude Opus 4.8. The only better models are Claude Fable and GPT 5.6 Sol Max…

You're not just tired (em dash) you're exhausted!

Another one besides Wikipedia https://tropes.fyi/tropes-md

DS3 isn't even looked at anymore. GLM-5.2 is the best in that class right now. It is competitive with current GPT/Claude/Gemini.

> LLMs... impedes the ordinary process of theory-building As I have said on here before, I actually really enjoy LLMs for code understanding precisely because they are imperfect. T…

Yes, seriously everyone should watch this: https://www.youtube.com/@AngeTheGreat Other engine simulators work by approximating the engine. Ange's engine simulator works by approxim…

Finally a believable headline about CATL. Next week we'll be back to CATL increasing energy density by 1 million percent with a battery that charges in 3 nanoseconds.