mirekrusin

6 ms·

31.
by mirekrusin 2mo ago
Read websites through llm.
32.
by mirekrusin 2mo ago
I had similar problems to solve for trading advice, complex project reasoning etc. I settled on simplicity, extended it to serve clear, useful purpose. Started with markdown database, single fact per line, structured/parseable (subject
33.
by mirekrusin 2mo ago
You should checkout cave lang [0] - terse language that explores this area of knowledge/graph/ontology/provenance/querying/confidence/solver etc. [0] https://mirekrusin.com/cave
34.
by mirekrusin 2mo ago
Yes, I remember recurring extensions on plan inclusion then becoming permanent - that's my point.
35.
by mirekrusin 2mo ago
With competition we kind of have guarantee up to what providers can do, they don't have that much control, the most radical thing they can do is to go bankrupt.
36.
by mirekrusin 2mo ago
pnpm 12 supports corepack (or better phrased - corepack supports pnpm 12). they just don't recommend it. nodejs stopped bundling corepack since node 25, maybe you're referring to that?
37.
by mirekrusin 2mo ago
Sure.
38.
by mirekrusin 2mo ago
Poor analysis. Fable is not used because it has extra retention requirement that corps can't sign off so it stays disabled for everybody in many cases. It's also not winning on day to day work against Opus 5, which is simply avail
39.
by mirekrusin 2mo ago
Great Wall of US AI :D Nice one, stealing it.
40.
by mirekrusin 2mo ago
Speculative decoding also works on Mac, 64G is more than what I have, m5 max should handle up to ~40 t/s with optimized setup (and with a lot of vram you can get great wins on concurrency – that harness can take advantage of for single
41.
by mirekrusin 2mo ago
Personally I find speculative decoding much better strategy than MoE – performance wise it's there at 90-100 t/s on 2x4090, great intelligence – really great fit.
42.
by mirekrusin 2mo ago
Chris literally was saying it from the very beginning (you can watch early interviews, obviously he was being asked about it), everything that happens with progressive opening is what he said he'll do and he gave rationale for it which
43.
by mirekrusin 2mo ago
yes, you know, personal use doesn't necessary mean no concurrency. it's good to play with harness setup where you fan out multiple concurrent branches that share non trivial amount of prefix then reduce their output/summary b
44.
by mirekrusin 2mo ago
Dual 4090, getting 85-113 t/s depending on task (draft seems to speed up quite a lot, disproportionately more for content like svg etc): ./llama.cpp/llama-server \ -hf unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XL \
45.
by mirekrusin 2mo ago
Why are people complaining? It’s not a bug, it’s a feature. You have a free bank holiday, go look at green trees, listen to birds, find somebody who's wrong on HN etc. The only poor souls here are some github devops team (team of copil
46.
by mirekrusin 2mo ago
Seriously?
47.
by mirekrusin 2mo ago
Next RL phase is to connect electrodes to human brain when reading generated output to reduce frustration signal.
48.
by mirekrusin 2mo ago
That's a good sign. Ideal local model would not know stuff like who Britney Spears is, best to leave precious weights for something useful. Of course the line is very blurry but I'd be perfectly happy with local model that doesn&#
49.
by mirekrusin 2mo ago
ie. open router fusion [0] [0] https://openrouter.ai/openrouter/fusion
50.
by mirekrusin 2mo ago
Just use all of them and synthesize final result, record scoring when synthesizing. Later you can make decision to drop low performing ones.
51.
by mirekrusin 2mo ago
Use of official API outside of China is closer to opposite of reality. Look at ie. OpenRouter you'll see how many providers there are and how much traffic they get. If the goal was data, opening weights would be the worst available way
52.
by mirekrusin 2mo ago
Your argument was that releasing models as open weight was allowing them to collect data but reality is that open weight models precisely allow users to completely avoid data collection – something that is not actually possible with closed
53.
by mirekrusin 2mo ago
Depends who you ask: META releases open weight models because selling access to models isn't their business model, they get optimization for free, good press with adoption of their tech etc. They don't want AI to become another iO
54.
by mirekrusin 2mo ago
Move like making api substantially cheaper or free would make more sense. Making it open weight does the opposite – it allows traffic to not be routed through them at all.
55.
by mirekrusin 2mo ago
Anthropic and friends infiltrate without open weights and their infiltration is 100% of traffic that goes through their official api.
56.
by mirekrusin 2mo ago
collecting data by making them open weight? where is logic in that?
57.
by mirekrusin 2mo ago
Yeah for that dense models are not great better to pick some MoE to get benefits of large vram pool while keeping reasonable speed.
58.
by mirekrusin 2mo ago
Optimize startup params for llama-server for your hardware (not llama.cpp itself), on my 2x 4090 I got ~20% speedup after maybe 40 mins. ps. ngram didn't work for me very well, but dedicated speculative model works very well ps. 2. in
59.
by mirekrusin 2mo ago
Just run /goal to optimise it and you should be good in less than an hour. Also best to use models that support speculative decoding.
60.
by mirekrusin 2mo ago
What would be an AGI data point?

More ›