kouteiheika

10 ms·

61.
by kouteiheika 4mo ago
It's pretty much the same as when "installing programs on your computer" is called "sideloading". Deliberately deceptive, weaponized language to make it seem like a bad thing.
62.
by kouteiheika 4mo ago
> That will put them in direct competition with Steam, though. Suddenly their cheaper console will result in way higher cost for the lifetime of the console. ...funny that so many people were complaining about the recent Steam Machine no
63.
by kouteiheika 4mo ago
> So although it's nice to have capable open models, it will increasingly bad for us all if open models keep fast-following closed model capabilities as they have been Cat's out of the bag. The only way to make them safe is to
64.
by kouteiheika 4mo ago
> What is Krea's approach to content such as pornography and gore? You mean whether it's censored? Of course it is. They even say so themselves[1]: > The open source version needed to go through some alignment training so th
65.
by kouteiheika 4mo ago
> Neither change has any technical reason. The only reason why either name change was desired is because some contributors were upset by the names. Indeed, neither has any technical reason. I suppose you're right that both changes h
66.
by kouteiheika 4mo ago
> Well, because it's true: many of the repositories are named after "Wacom". It's a historical legacy on GNU/Linux. It's also a decade-long debate that these repos should be renamed differently. Okay... let&
67.
by kouteiheika 4mo ago
> the main thing that we've done is we've commissioned additional research on this because I've not been happy with the evidence. Ah, yes, the existing research doesn't agree with our biases, so let's fund new &q
68.
by kouteiheika 4mo ago
> I am pretty sure there are entire YouTube channels dedicated to this. There is: https://www.youtube.com/@Dr.Semiconductor
69.
by kouteiheika 4mo ago
> since they were caught stealing it from American labs ...and "good guys" the American labs were caught stealing from authors all over the world[1]. [1]: www.npr.org/2025/09/05/g-s1-87367/anthropic-aut
70.
by kouteiheika 4mo ago
> I think that this is what OpenAI/Anthropic want but they wont say it publicly. Won't say it publicly? Anthropic is openly and explicitly saying it publicly. Here: https://darioamodei.com/post/policy-on-th
71.
by kouteiheika 4mo ago
> I think you overestimate human translators. There is a lot of very poor quality human-translated text out there. This. There was even a big controversy recently with one of the games on Steam where human translators just completely bot
72.
by kouteiheika 4mo ago
Sure, but the difference is that one side (Anthropic, OpenAI, Google and co.) hoards everything, keeping it proprietary behind API paywalls and constantly spewing AI doomer rhetoric while limiting what you can do "for your own safety&q
73.
by kouteiheika 4mo ago
Ah yes, those pesky Chinese backdoors that no single instance was ever found, even though Chinese open-weight model are a thing for many years now. Many people burn through millions of tokens on these models every day - surely someone would
74.
by kouteiheika 4mo ago
Not every weight is quantized. For example, those weights which don't take much space or are highly important are left in higher precision. State-of-art quantization of weights is never done uniformly (i.e. to all weights and in th
75.
by kouteiheika 4mo ago
> It just got worse the more sophisticated games became. I feel like it got worse not because games became more sophisticated, but because they got dumber and require less and less brainpower. I spent my childhood playing ridiculous
76.
by kouteiheika 4mo ago
Doesn't matter, because they're open-weight, so I can just download them to my PC and... hey, look, now they're owned by me! Unlike the "good" Western counterparts which are all fully proprietary. (Except Mistral, b
77.
by kouteiheika 4mo ago
The model is natively quantized (i.e. it was trained that way in the first place, so this is not a post-training quantization which degrades performance).
78.
by kouteiheika 4mo ago
Yes, I would agree with this. I think it works best when you're using the agent in a more hands-on way with a targeted prompt. If you're obsessive about code quality like I am (so you thoroughly review and, when needed, reprompt o
79.
by kouteiheika 4mo ago
This. Here's a quick experiment I did yesterday. I got a new $20 Claude subscription to try the new Fable model. I gave it a single prompt, and it barely finished, using up my whole session quota (it was at ~95% when it finished) a
80.
by kouteiheika 4mo ago
> AI companies that develop advanced AI models must have strong security standards that protect their model weights So, basically, make open-weight models illegal. It's nice for Dario to come out and say this so explicitly.
81.
by kouteiheika 4mo ago
> Memory safety is just a tiny part of over all security. 70%[1][2] is tiny? [1]: https://www.zdnet.com/article/microsoft-70-percent-of-all-se... [2]: https://www.chromium.org/Home/chromium-secu
82.
by kouteiheika 5mo ago
> What do people imagine can be done about it at this point? Offer a concrete suggestion. Simple. Free the companies from copyright liability, but after X amount of time they are required to release everything into the commons. The wei
83.
by kouteiheika 5mo ago
> now they're taking purely emotional stances like "AI evil" But they aren't? Nowhere in the document it says this; in fact, it says the opposite - that they don't want to make a moral judgement. > It's
84.
by kouteiheika 5mo ago
> I would imagine not single everyone on HN have enough disposable income that allow us to subscribe Claude Max or other similar max plan of other models without thinking. You don't need the Max plan with other models if you're
85.
by kouteiheika 5mo ago
Depends on which exact model we're talking about, and on your salary. For example, with the $40/month Kimi Code subscription the limits are so generous that you can use it every day all the time for everything (basically just have
86.
by kouteiheika 5mo ago
> Like seriously, what's the point of explicitly allowing this? Imagine the opposite were true, you weren't allowed to do this - what would they do? Imagine if they just say "LLMs are banned" then there's a lot o
87.
by kouteiheika 5mo ago
NVidia supports their GPUs for a really long time (unlike AMD, which paradoxically drops official support really fast; e.g. see their ROCm support). Anyway, by the time NVidia drops support for their current newer GPUs there's a high c
88.
by kouteiheika 6mo ago
Are you kidding? The main difference here is not that DeepSeek's model is completely free of censorship (although I'd wager it's less censored), but that it's open-weight . That has two major advantages: 1) If Anthrop
89.
by kouteiheika 6mo ago
It'd be nice if you could add a separate leaderboard for open-weight models on your results page (or add the ability to filter-out proprietary models). Also, why use an agent for this? This doesn't make much sense to me, consideri
90.
by kouteiheika 6mo ago
> That's the pre-tokenizer, not the tokenizer. Yes, it's an extra tokenizer which runs before the learned tokenizer and injects an inductive bias into it. > That is mostly a performance optimization that lets the memory requ

More ›