Tenoke

8 ms·

1.
by Tenoke 8d ago
It's possible to be a waste of European money that could be better used elsewhere, though. I think the same about LeCun's company sucking up the little funding here.
2.
by Tenoke 8d ago
> Europe absolutely needs a home-grown AI lab, especially with Pax Americana looking increasingly shaky. The only problem (at least in the LLM space) is that you can do more in Europe by just getting the best Chinese Open weights model (
3.
by Tenoke 13d ago
Yes, this is exactly what I expected (showing the anthropic principle in action), so I am disappointed they are not sampling correctly.
4.
by Tenoke 18d ago
>there ought to be a tipping point beyond which local inference is good enough There's no such ought really. Even at current levels you'd need like a 100x gain from here to approach current top proprietary models (probably a lo
5.
by Tenoke 18d ago
There's been a ton of optimizations already, it hasn't remotely reduced demand even temporarily. More efficiency just makes the compute have even higher ROI per $ and watt spent.
6.
by Tenoke 20d ago
Too base cynicism. I'd be willing to bet you my $200 to your $100 that doesn't happen.
7.
by Tenoke 27d ago
That's a good formalization of it. I will think you are a bit of a fool if you dont use AI but you guarantee you'll be a fool if you only use AI with no value added by yourself.
8.
by Tenoke 2mo ago
Fable has more parameters. In practice it's not yet clear which one would be better for different usecases yet but they are more different than one being strictly better.
9.
by Tenoke 2mo ago
Mistral is much worse in its respective field than Flux in their own so I hope not.
10.
by Tenoke 2mo ago
Flux 2 Dev Klein has practically been the best you could use on most commercial hardware so I really hope Flux 3 has a comparable updated open-weights model to it. if not it'd be a great loss to most hobbyists.
11.
by Tenoke 2mo ago
Y Combinator the company doesnt particularly have to share the opinions of hackernews the public site.
12.
by Tenoke 2mo ago
Because sadly, as much as I wish it wasn't the case, offense is easier, more impactful than defense.
13.
by Tenoke 2mo ago
I am very pro open source models - I use them every single day.. But we obviously don't want everyone to have capabilities like and beyond what caused the huggingface incident in every domain , so it's not like it all comes from
14.
by Tenoke 2mo ago
There's ways to make sure env vars get only injected at runtime and arent easily accessible otherwise or to even make them inaccessible to the user your agent is running on, and for you to manually run the code with the right permissio
15.
by Tenoke 2mo ago
That's kind of insane. Natural that it's happened, sure, but insane. I know people don't like thinking of it like that, but things analogous to this can easily happen in various domains with today/tomorrow's models
16.
by Tenoke 2mo ago
It's also very possible that they know their big model underperforms chatgpt 5.6 and fable by too much, so they are focusing on what they can get wins in like speed instead.
17.
by Tenoke 2mo ago
I do that but it ends up putting so much extra crap there that it has the same drawbacks.
18.
by Tenoke 2mo ago
Claude seems to forget what you tell it in very long work sessions (things that take weeks to develop), no matter how many times you tell it which part is extra important. I dont use goal (I guess I should), but presumably it makes it actua
19.
Apple M7 Ultra Chip Planned with Up to 1.5 TB of Unified Memory(techpowerup.com)
5 points by Tenoke 2mo ago|2 comments
20.
by Tenoke 2mo ago
>The human brain manages to self-organize with only a fraction of the information that LLMs get trained on. So? The question isnt can we get to ASI as efficiently as a brain, the question is can we get there, which we likely can. The ine
21.
by Tenoke 2mo ago
LLMs dont just use text for a while now. It's also not fully supervised for a while.
22.
by Tenoke 2mo ago
You can watch the whole Lex Friedman interview, it's on youtube. It's not out of context at all. He goes on about how LLMs will never be able to do things that they do trivially. And he has just doubled down for years. Ive read an
23.
by Tenoke 2mo ago
https://youtube.com/shorts/zQTt8TkcyfU?is=09r7XDqz2w6-Pygu You probably wont like the edit but I dont have the timestamp of the original on hand, you can find it.
24.
by Tenoke 2mo ago
>He's merely said they don't think He said years ago even 'GPT 5000' couldnt do things that they ended up doing fine a month later, let alone by 5000. His later predictions are just moving that goal post including tow
25.
by Tenoke 2mo ago
His main anti-LLM predictions have been consistently either wrong or misleading. There's many ways to skin a cat so you can probably do something with a JEPA approach as well, but I doubt he actually catches up to having agents on the
26.
by Tenoke 2mo ago
Is any of those comparisons about Pro vs non-Pro (Pro is only available in $100+ plans)? I am curious about that but I think Sol, Terra, Luna are different sizes of it without the Pro part, and I want to know how much worse do I have it on
27.
by Tenoke 2mo ago
8B sounds tiny. Of course, that's enough to easily run on device which is nice, but surely the actual SOTA must be some much bigger model?
28.
by Tenoke 2mo ago
If you are imagining that, you could imagine it with search doing the same 10 years ago, which would have more thoroughly prevented you from researching things.
29.
by Tenoke 2mo ago
>but you cannot really hold a "portfolio" of it in any sensible way. xAI effectively did and lucked out to cover their losses and more with it.
30.
by Tenoke 2mo ago
1. The liquidity is not infinite to compound that easily. 2. The alpha dries up with more players, even in the year or whatever since that founder started.

More ›