lostmsu

5 ms·

31.
by lostmsu 1mo ago
Sagittarius A*
32.
by lostmsu 1mo ago
Who bought it?
33.
by lostmsu 1mo ago
What a day. OpenAI is behind basically all major competitors - at least for a some amount of time.
34.
by lostmsu 1mo ago
The whole cotton bag thing could only work for people who already carry bags around all the time.
35.
by lostmsu 1mo ago
There's been research showing that high "acceptable" CO2 causes drop in cognitive abilities.
36.
by lostmsu 1mo ago
I would call Fallout dismal (speaking as someone with over 1000h across all Fallout games sans 76).
37.
by lostmsu 1mo ago
BigCos you are referring to are earning their money doing state of the art research, not so much from your financial and medical data. That was the age of Internet ads, which was over since AdBlock was created for anyone concerned.
38.
by lostmsu 1mo ago
5 points is a huge regression
39.
by lostmsu 1mo ago
> GLM 5.3 Flash at Q4, AA score 57 That AA score is for the original model only
40.
by lostmsu 1mo ago
Instantly removed potential 0.05% of players
41.
by lostmsu 1mo ago
GLM 5.2 in NVFP4 is 465 GB. It would be a tough fit.
42.
by lostmsu 1mo ago
Still no SFP option?
43.
by lostmsu 1mo ago
Run ... without graphics acceleration.
44.
by lostmsu 1mo ago
Yes, my 5yo also invents fake reasons to explain why she didn't listen. > I don't believe you believe in any notion of spirit See how stupid that sounds now I wrote it in plain terms. You are telling me I lied about believing i
45.
by lostmsu 1mo ago
IMO Pareto frontier is a lie. At any given moment there's only one "best" strategy and my guess would be in most cases it is just using the best model.
46.
by lostmsu 1mo ago
This sounds promising. I am surprised the capacity on the table ends at 512GiB. Is this per chip or per device?
47.
by lostmsu 1mo ago
Xiaomi smartphones were extremely well made at least a few years ago. Even the core software was impressive vs Google for example (try echo cancellation on any of your non-Xiaomi devices). Its only real problem is ads.
48.
by lostmsu 1mo ago
Performance difference it large by all benchmarks. DeepSeek fell behind. It's Kimi K3 or GLM-5.3 now.
49.
by lostmsu 1mo ago
I got a usage reset today. Possibly related.
50.
by lostmsu 2mo ago
Yes, I don't believe in your "spirit", but I certainly believe in mine https://dictionary.cambridge.org/dictionary/english/spirit I even hinted at the meaning by using "spirits". Next time
51.
by lostmsu 2mo ago
DeepSeek is much dumber at the moment. It's barely better than Qwen3.8 27B that you can run locally.
52.
by lostmsu 2mo ago
They are trying to present a new unemployment metric as more reasonable, but throughout the article only give 2 data points making it impossible to judge except from prior bias.
53.
by lostmsu 2mo ago
I got excited that it's like TurboPascal, but this is basically VS Code over RDP (uses video streaming over kitty protocol).
54.
by lostmsu 2mo ago
No, when people die and rot in the ground - that's a complete spiritual death. I assure you my spirits are in a good shape. And I hope the model will too.
55.
by lostmsu 2mo ago
> And if that isn’t a comfort, I don’t know what is. Same, but actually much of your entire life recorded and trained on.
56.
by lostmsu 2mo ago
KLD isn't how that works either. The truth is in the middle and they aren't showing it.
57.
by lostmsu 2mo ago
I would say they do compound until proven otherwise. Having "Wait, bar is not true, so that won't work" is not necessarily a correction. In fact, the problem is: across a long text it is a correction of a single mistake, but
58.
by lostmsu 2mo ago
Cool. Now run TerminalHard and compare to unquantized 27B. KLD of 1%, or similar error metric that multiplies, on 10000 tokens would give accumulated error of 2,000,000%
59.
by lostmsu 2mo ago
It sounds way worse than just napping.
60.
by lostmsu 2mo ago
KV caching status? What's the point of 1000tok/s if you have to do prefill on every agentic turn which at 100k depth would make it 1.5 min latency every turn?

More ›