Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
machiaweliczny
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
8 ms
·
91.
▲
by
machiaweliczny
2y ago
It feels like name pressure to become astronaut. His name could be loosely translated as „Glorious Achievement-Cherry”.
92.
▲
by
machiaweliczny
2y ago
I actually think that diffusion LLMs will be best for code generation
93.
▲
by
machiaweliczny
2y ago
Would be great if some details given about how exactly model is penalized for staying off-track.
94.
▲
by
machiaweliczny
2y ago
What about “Pop! OS” ?
95.
▲
by
machiaweliczny
2y ago
Problem Is government as they should know exactly what is wrong - that is their job
96.
▲
by
machiaweliczny
2y ago
You seem to be missing $15T in minerals and for sure we can’t afford a freeze on current lines and this will make Russia invasion very profitable.
97.
▲
by
machiaweliczny
2y ago
What most people need is good routing solution but it’s hard to make it with SSR. Seems like frameworks are somewhat converging now on this app router concept. Agree that JSX is great (but styles like in Svelte also).
98.
▲
by
machiaweliczny
2y ago
Have you used MobX? It was there 10 years ago and still working. Svelte runes are basically the same with some perf wins but require compilation
99.
▲
by
machiaweliczny
2y ago
True, Hitler mostly lost a war due to megalomania and spending 1/4th of naval budget on big destroyer instead of producing more ubots.
100.
▲
by
machiaweliczny
2y ago
EU has enough capital it just works on US market cause there no unified capital market in EU. If Trump hits EU with tariffs I hope this changes.
101.
▲
by
machiaweliczny
2y ago
Do they have something tangible now? At least PWr (from Poland) have fine tuned Mistral and a lot of language data AFAIK - https://bielik.ai/ Anyway I think some novel training method is needed and I’m with Yann on this. Th
102.
▲
by
machiaweliczny
2y ago
This is very bearish for current AI. Seems like 99% reliability is still too small with compounding errors. But I wonder of this is inherently specific to longer context or if this just depends on how it’s trained. In theory longer context
103.
▲
by
machiaweliczny
2y ago
I had some gut bacteria issues and I think I was helped recently by using oregano oil and/or allicine (extract from garlic). Used both for a week, before had 2x antibiotics for suspected SIBO. I also used VivoMaxx 450 probiotics and so
104.
▲
by
machiaweliczny
2y ago
IMO it should be ranked order and random assignment
105.
▲
by
machiaweliczny
2y ago
I wish more people were aware of this
106.
▲
by
machiaweliczny
2y ago
Embeddings are very cheap to generate
107.
▲
by
machiaweliczny
2y ago
Embeddings are super cheap to compute
108.
▲
by
machiaweliczny
2y ago
Yeah, I liked my windows phone, not sure why they killed it
109.
▲
by
machiaweliczny
2y ago
Long context is a scam. Claude is best but it’s still gets lost with longer context
110.
▲
by
machiaweliczny
2y ago
I’ve watched vide with guy that supposedly has 200IQ and in his view IQ is ability to focus on small on small set of things but many high IQ ppl can’t „refocus” it easily and aren’t doing well in life.
111.
▲
by
machiaweliczny
2y ago
How you compare a liked set to single embedding? By best match? Or is there some fancier way?
112.
▲
by
machiaweliczny
2y ago
But if you squint then sensory actions and reactions are also sequential tokens. Even reactions can be encoded alongside input as action tokens and as single token stream. Anyone tried sth like this?
113.
▲
by
machiaweliczny
2y ago
There’s a reason that US has gun rights in constitution and it’s dems that want to take it away. It’s interesting tradeoff that should in theory protect from tyranny. It’s interesting what would happen in Russia if majority of population ha
114.
▲
by
machiaweliczny
2y ago
France seems to be doing better than UK it seems
115.
▲
by
machiaweliczny
2y ago
I see little analysis of Elon’s role in this election. Does anyone know if the accusation of him takin Russian money for Twitter were real?
116.
▲
by
machiaweliczny
2y ago
Nice. What algorithm is used for recommendation?
117.
▲
by
machiaweliczny
2y ago
In normal critical system u use 3 CPUs. With LLM u can 1000 shot majority voting. Seems like approaches like entropix might reduce hallucinations also.
118.
▲
by
machiaweliczny
2y ago
I'm waiting for Aider benchmark
119.
▲
by
machiaweliczny
2y ago
Probably better than 4o-mini, 4o-mini isn’t great from my testing. loses focus after 100 lines of text
120.
▲
by
machiaweliczny
2y ago
You can use Cursor (VS fork) with private Anthropic key
More ›