Bolwin

5 ms·

61.
by Bolwin 4mo ago
You can get the best of both worlds with a small embedding model
62.
by Bolwin 4mo ago
The social section is also just selecting for introverts it asocial people But they expect a few wrong
63.
by Bolwin 4mo ago
In my experience above 60k quality noticeably drops. 30k for open source models
64.
by Bolwin 4mo ago
What's wrong with the name? The step function is a pretty well known one
65.
by Bolwin 4mo ago
Here's another: https://xcancel.com/FireworksAI_HQ/status/206010388602804673... Fireworks is processing 30T tokens a day on open models, or about 210T a week. So about 40% of gemini? I'd say that's
66.
by Bolwin 5mo ago
I've never seen emoji used for subtext. Usually they just repeat or emphasize what's in the text
67.
by Bolwin 5mo ago
OpenRouter is not indicative of volume. Most high volume clients will go to the providers directly. There's not point to paying the 5% OR cut if you know what you want.
68.
by Bolwin 5mo ago
I mean there is a minor moat. Most people don't enjoy switching providers or models. If you can get people to trust you'll stay near frontier, they'll stick around even when you aren't the best. Claude is a prime example
69.
by Bolwin 5mo ago
No one is producing one output token though. And using up gpus for that cache is a pretty big opportunity cost. I highly doubt it's done in vram. That would be insane for the one hour caches. So its memory + the time it takes to unload
70.
by Bolwin 5mo ago
Cause it's very verbose. A lot more syntax to break. I personally think the best is one of the humanized json ones like https://maml.dev/
71.
by Bolwin 5mo ago
I don't see how being decentralized helps search. Makes it quite harder if the fediverse is any indication
72.
by Bolwin 5mo ago
They're not distilled. Stop spreading anthropics misuse of the term. They do use it for synthetic data/judging though, so yes, hard to outperform. Not that they need to. If they can basically match it for a fifth of the price.
73.
by Bolwin 5mo ago
The average user may be fine with it (though I think many will not). But given this is basically killing the open web, I don't see where Google plans to get the results to feed this AI thing in a year or so
74.
by Bolwin 5mo ago
This is the default on the web and most people expect it . I much prefer it
75.
by Bolwin 5mo ago
..where did you get that impression? The web also requires you to convert to html. I don't see how this is different
76.
by Bolwin 5mo ago
The author said the mean reason for the speedup was parallelization, not Go being particularly faster. The JS JIT is very good. In most benchmarks I've seen its not much slower than Go.
77.
by Bolwin 5mo ago
Not unless you use their theme system and components from my understanding.
78.
by Bolwin 5mo ago
The photos app also uses webview so yes hopefully
79.
by Bolwin 5mo ago
No one's forcing you to use Google. I've found Kagi to be pretty good
80.
by Bolwin 5mo ago
Cause they all share one webview. Electron apps each run their own version of chromium
81.
by Bolwin 5mo ago
File pilot seems to use an immediate mode UI to be responsive and uses 10% cpu even when idle or minimized. I'm not sure that's worth it.
82.
by Bolwin 5mo ago
The training data is not so specifically filtered at least in pre training. The point is to give them as much world knowledge as possible
83.
by Bolwin 5mo ago
Yeah that was my first thought. Use a TUI that abuses the terminal to make a experience poorly recreating an actual GUI, don't be shocked if it doesn't turn out very efficient
84.
by Bolwin 5mo ago
I actually like the XS one. It's broken, but it's very aesthetic
85.
by Bolwin 6mo ago
I just admit, I didn't expect this to make it to HN
86.
by Bolwin 6mo ago
The AI tells are quite literally what made me lose interest. They help me filter out dozens of AI slop posts daily so I'll be keeping them thanks. They're lazy by design. You can generate a good looking post in 30s, I can't a
87.
by Bolwin 6mo ago
The small games manifesto has no LLM tells. If yours really is fully handwritten, then you've been reading so much AI writing you're mimicking a lot of its patterns, which is sad to see.
88.
by Bolwin 6mo ago
AI written/edited front page. A passion project and you can't even trust yourself to write the front page. Made me lose interest.
89.
by Bolwin 6mo ago
> summarizing web pages For summarizing creative writing, I've found Opus and Gemini 3 pro are still only okay and actively bad once it gets over 15K tokens or so. A lot of long context and attention improvements have been focused o
90.
by Bolwin 6mo ago
I think it's less about the extremity of evil and more about lacking the means to get rid of it in a more civil manner.

More ›