lostmsu

4 ms·

61.
by lostmsu 2mo ago
I wouldn't generalize this to all LLMs. So far I only saw Anthropic ones affected.
62.
by lostmsu 2mo ago
My old crappy HP laptop boots with a battery pulled out.
63.
by lostmsu 2mo ago
The ads make sense in capitalism. Without the described effect products would never change. Payment for ads ensures discovery works in adversarial free market. The problem is that the pay for that goes to Amazon instead of your pockets. I t
64.
by lostmsu 2mo ago
Why only as oneself experiences? I can easily tell if a horse is conscious or not. The whole matter is rather simple IMO. Let's say it's variability (number of distinct if you prefer discrete) of stable responses to external stimu
65.
by lostmsu 2mo ago
If that would be the case, how come only 3 copies are left?
66.
by lostmsu 2mo ago
Do you think the lower energy levels also affect your cognition?
67.
by lostmsu 2mo ago
Glimmer is stupider than 3.6 27B. You can't compare its speed to 3.8 and be done.
68.
by lostmsu 2mo ago
No, it would have been better to say "I don't know"
69.
by lostmsu 2mo ago
> proxy for totally broken or not You can't really know that either.
70.
by lostmsu 2mo ago
Your comment is no less ridiculous than his prediction. It is not even 2027 yet. You can't possibly know what AI will look like in 2027.
71.
by lostmsu 2mo ago
> The fact that Unsloth only just started publishing KL divergences shows how unserious the quantization space is. Just wanted to say that this is a very important point that I totally agree with. People are obsessed with KL divergence,
72.
by lostmsu 2mo ago
Still no KV caching?
73.
by lostmsu 2mo ago
Wow, OpenAI is now 4th after Opus 5, K3, and Grok
74.
by lostmsu 2mo ago
> nvfp4 Still trying to lock in, huh?
75.
by lostmsu 2mo ago
It's not even competitive with 2x sized Qwen 4B. Why is Qwen3.5 2B not in the table?
76.
by lostmsu 2mo ago
No mention on the website as far as I can see.
77.
by lostmsu 2mo ago
No AMD?
78.
by lostmsu 2mo ago
> sadly no breakdown That's exactly the point. We know short context knowledge stuff does not regress with quantization. But I expect agentic intelligence to suffer greatly. If I were to pick one bench, I would like to compare quant
79.
by lostmsu 2mo ago
Are you running inference in parallel? 70 tps seems low for parallel execution.
80.
by lostmsu 2mo ago
This release is not a meaningful improvement in any metric over 5 months old Qwen 3.6. DS v4 Flash update maybe, but it is too big for typical Joe's desktop.
81.
by lostmsu 2mo ago
From my perspective it doesn't make sense to talk about the number of parameters. What matters is model size in bytes and its performance at that certain size. Meta actually relesed official 4 bit quants in 17GB, but I haven't see
82.
by lostmsu 2mo ago
They say it is trained with quantization awareness, so it should only be 15GB or so. Qwen was only trained in FP8 with QAT. UPD, NVM, got misled by comments here. It is actually almost 60 GB so much larger
83.
by lostmsu 2mo ago
State of the art of philosophy is IMHO Popper and scientific method. Everything that came after is BS, everything that came before is obsolete.
84.
by lostmsu 2mo ago
Requires login. Any public page?
85.
by lostmsu 2mo ago
It seems worse than 3.6, but a bit smaller. UPD. was wrong on smaller, it's actually much larger
86.
by lostmsu 2mo ago
Any weakness can
87.
by lostmsu 2mo ago
Huh? I regularly ask codex to not just summarize or extract a crux of published papers, but sometimes to do a topic directed search and give a comparison table. If you aren't doing that, do you see the problem?
88.
by lostmsu 2mo ago
> She says she worries that becoming a mother would mean giving up the life she has now. The lady even told you the reason, but you still brought up the climate change, religion, and social media.
89.
by lostmsu 2mo ago
When you test matmul these days you want bf16 max. FP64 is a niche workload 100% unused by 95%+ users.
90.
by lostmsu 2mo ago
Sure, start sabotage with this one then https://oceancos.com/ocean-cold Salmon is not too important anyway.

More ›