Philpax

26 ms·

631.
Facet: Rust const and no_std reflection – the last derive you'll ever need(github.com)
5 points by Philpax 2y ago|0 comments
632.
by Philpax 2y ago
Humans are one solution to the problem of intelligence, but they are not the only solution, nor are they the most efficient. Today's LLMs are capable of outperforming your average human in a variety (not all, obviously!) of fields, des
633.
by Philpax 2y ago
There's a "Altered or synthetic content" notice in the description. You can also look at the rest of the channel's output and draw some conclusions about their output rate. (To be clear, I have no problem with AI-generat
634.
by Philpax 2y ago
Humans will always produce; it's just that those productions may not be financially viable, and may not have an audience. Grim, but also not too far off from the status quo today.
635.
by Philpax 2y ago
Meaning is in the eye of the beholder. Just look at how many people enjoyed this and said it was "just what they needed", despite it being composed of entirely AI-generated music: https://www.youtube.com/watch?v=Og
636.
by Philpax 2y ago
I am hoping that will be our saving grace this time around as well, but my fear is that the oligarchs will control more autonomous power than we can meaningfully resist, and our existence will no longer be strictly necessary for their syste
637.
by Philpax 2y ago
Stating that LLMs are not more reliable or useful than ELIZA or SmarterChild is so incredibly off-base I have to wonder if you've ever actually used a LLM. Please run the same query past ELIZA and Gemini 2.5 ( https://aistudi
638.
by Philpax 2y ago
I appreciate someone named "behringer" posting this sentiment. ( https://en.wikipedia.org/wiki/Behringer#Controversies )
639.
by Philpax 2y ago
> Despite the flood of benchmark-tuned LLMs, we remain nowhere close to engineering a machine intelligence rivaling that of a cat or a dog, let alone within the next 5 to 10 years. Depends on your definition of "intelligence."
640.
by Philpax 2y ago
I'm glad we're seeing the death of the concept of owning an idea. I just hope the people who were relying on owning a slice of the noosphere can find some other way to sustain themselves.
641.
by Philpax 2y ago
Not quite comparable; these systems will continue to grow in capacity until there is nothing for your average human to be able to reskill to. Not only that, they will truly be beyond our comprehension (arguably, they already are: our interp
642.
by Philpax 2y ago
> AI can't possibly do _everything_ Why not? There's the human bias of wanting to consume things created by humans - that's fine, I'm not questioning that - but objectively, if we get to human-threshold AGI and contin
643.
Report Reveals Internal Chaos Behind Apple's Siri Failure(macrumors.com)
7 points by Philpax 2y ago|1 comments
644.
by Philpax 2y ago
Sounds like you should improve your setup ;)
645.
by Philpax 2y ago
You are both correct: the post-training stage of most new LLMs involves their identity being trained in, so that they "know who they are" without the system prompt. Without this step, most LLMs will respond with whatever identity
646.
by Philpax 2y ago
> “We need to get better at treating this like a business,” Acting ICE Director Todd Lyons said, explaining he wants to see a deportation process “like (Amazon) Prime, but with human beings.” This is one of the most evil things I've
647.
by Philpax 2y ago
I believe https://flawless.dev/ is another implementation with a very similar technology stack. I'd love to know how you compare and what the key differences are!
648.
by Philpax 2y ago
OLMo 2 from AI2 is fully open-source, including code, data and recipe: https://allenai.org/olmo There are others, but OLMo is the most recent and competitive.
649.
by Philpax 2y ago
The 2014 film Transcendence touches upon this, but I can't recommend watching it. It's surface level and the rest of the film is mediocre.
650.
China to raise reciprocal tariffs on U.S. goods to 84%(cnbc.com)
28 points by Philpax 2y ago|46 comments
651.
by Philpax 2y ago
LLMs are clearly useful, even if flawed; conversely, the author had to struggle to find uses of Cyc. Additionally, neural networks were already showing promise before the scale-ups; AlexNet was trained on consumer hardware and outperformed
652.
by Philpax 2y ago
Well, unlike cryptocurrency, it works reliably and quickly, does not require you to manage your own keys, and provides recourse if things go wrong. Those are pretty strong points in its favour.
653.
US to go ahead with imposition of 104% tariff China from Wednesday(theguardian.com)
50 points by Philpax 2y ago|16 comments
654.
2025 Stock Market Crash(en.wikipedia.org)
22 points by Philpax 2y ago|5 comments
655.
by Philpax 2y ago
Be serious now. Plenty of useful, privacy-required queries can be run with 24GB of VRAM, especially given the existence of e.g. Gemma 3 27B and the heavy NVIDIA-targeted optimisation work that has occurred. The Tenstorrent cards exist, but
656.
by Philpax 2y ago
That is a very strange omission...
657.
by Philpax 2y ago
I've had reasonable success with taking diary-style notes on whatever project I'm working on within that project's page in Obsidian, and then when I need to come back to it, either reading over it or passing it through a LLM
658.
Change of plans: we are going to release o3 and o4-mini after all(twitter.com)
4 points by Philpax 2y ago|0 comments
659.
by Philpax 2y ago
You've just invented Unison :)
660.
by Philpax 2y ago
You should have probably read the post.

More ›