Philpax

34 ms·

691.
by Philpax 2y ago
> The only thing that survives from one iteration to the next is the singular highest ranking token, the entire state and "thought process" of the network cannot be represented by a single token, which means that every strategy
692.
by Philpax 2y ago
Do you know how you work?
693.
by Philpax 2y ago
That's true for mask-based training (used for embeddedings and BERT and such), but not true for modern autoregressive LLMs as a whole, which are pretrained with next word prediction.
694.
Tracing the thoughts of a large language model(anthropic.com)
1072 points by Philpax 2y ago|394 comments
695.
Gemini 2.5 Pro and Google's second chance with AI(interconnects.ai)
3 points by Philpax 2y ago|0 comments
696.
The State of LLM Reasoning Models(magazine.sebastianraschka.com)
2 points by Philpax 2y ago|0 comments
697.
by Philpax 2y ago
It helps with iteration - you can try out different concepts and variations quickly without having to wait, especially as you refine what you want and your understanding of what it's capable of. Also, in general, why not?
698.
by Philpax 2y ago
https://en.wikipedia.org/wiki/Swenglish Not uncommon for Swedish-first speakers - just accidental language bleed, as is the case with most other second language speakers. (As for "strange world" - have you lo
699.
by Philpax 2y ago
Organisational culture / funding / quality of infrastructure. Compare Swiss trains to German trains: geographically colocated, but the former is known for extreme timeliness, while the latter is known for extreme tardiness. If you
700.
by Philpax 2y ago
That's a good question - I was initially thinking that it was pretrained from scratch using the Llama arch, but https://github.com/canopyai/Orpheus-TTS/blob/main/pretrain/c... implies the use o
701.
by Philpax 2y ago
If you don't want to be hit by the anti-monopoly legislation, just don't be a monopoly.
702.
Managing frontier model training organizations (or teams)(interconnects.ai)
1 points by Philpax 2y ago|0 comments
703.
BotQ: A High-Volume Manufacturing Facility for Humanoid Robots(figure.ai)
1 points by Philpax 2y ago|0 comments
704.
My Thoughts on the Future of "AI"(nicholas.carlini.com)
2 points by Philpax 2y ago|0 comments
705.
Mistral Small 3.1(mistral.ai)
47 points by Philpax 2y ago|4 comments
706.
by Philpax 2y ago
Reading the news: - https://eu.usatoday.com/story/news/nation/2025/03/13/trump-c... - https://eu.usatoday.com/story/news/nation/2025/03/11/columbi.
707.
by Philpax 2y ago
That's not the argument they were making. What they're stating is that the Trump administration is directly facilitating the slow death of American intellectualism, and that this will have downstream consequences for what advances
708.
ATProto Isn't What You Think(blog.muni.town)
55 points by Philpax 2y ago|36 comments
709.
The "No Clear Winner" Era of Federated Microblogging(github.com)
4 points by Philpax 2y ago|0 comments
710.
OLMo 2 32B: First fully open model to outperform GPT 3.5 and GPT 4o mini(allenai.org)
25 points by Philpax 2y ago|3 comments
711.
by Philpax 2y ago
Luckily, we have ways of verifying mathematical results, and using that to improve our AI systems: https://deepmind.google/discover/blog/ai-solves-imo-problems...
712.
by Philpax 2y ago
The robot arms in the movie are implied to have their own AIs driving them; Tony speaks to the malfunctioning one directly several times throughout the movie. Jarvis is AGI, yes, but is not what's being referred to here.
713.
by Philpax 2y ago
His politics were awful, but it's still a shame to see someone who enabled TPB's existence passing away. Partial RIP, I guess.
714.
by Philpax 2y ago
I visited the Mucha museum in Prague and was surprised by how many of his works were advertising posters. On one hand, I don't care for advertising: on the other hand, it brought us these wonderful works that we admire a century on, di
715.
by Philpax 2y ago
If this isn't AI, what is? It's an autonomous robot!
716.
by Philpax 2y ago
> I might or might not have fallen asleep one night on a long road trip and awoken to my car driving along happily This should have been a sign to re-evaluate your relationship to driving, not to double-down on becoming reliant on Comma.
717.
by Philpax 2y ago
> It is kind of amazing that these models manage to figure out a strategy at all, considering the game is not in their training set. I'm not sure we can conclude the game's not in their training set - a lot has been written abo
718.
Elicitation, the simplest way to understand post-training(interconnects.ai)
2 points by Philpax 2y ago|0 comments
719.
by Philpax 2y ago
For me, Twitter as we knew it has been long dead, and X is the shambling, corrupt corpse that's taken its place. I no longer mind referring to it as X, because that clearly outlines that it's a different website with a much more r
720.
Domain specific architectures for AI inference(fleetwood.dev)
4 points by Philpax 2y ago|0 comments

More ›