j_maffe

17 ms·

361.
by j_maffe 2y ago
GP isn't arguing for subjective preference but objective value. People are of course allowed to find their work impactful. Doesn't mean it actually is.
362.
by j_maffe 2y ago
Of course not. And he is relatively more popular than some others on his level of expertise. But I think Tao's opinion is a bit more interesting than a general PhD/professional.
363.
by j_maffe 2y ago
The grind to do what? Make an endless stream of shit content? Also https://en.wikipedia.org/wiki/False_dilemma
364.
by j_maffe 2y ago
You're really overestimating the cost of video storage and streaming compared to the kind of revenue they're able to get.
365.
by j_maffe 2y ago
Do you have any proof of that?
366.
by j_maffe 2y ago
How does it fall under the definition of a bubble? Sure, view counts contribute to more views. But that's not the main retention mechanism of these videos.
367.
by j_maffe 2y ago
Awesome work! I love what the server has grown into.
368.
by j_maffe 2y ago
Mathstodon is one of the most active Mastodon instances. I have no idea how it reached its current level of activity.
369.
by j_maffe 2y ago
If anyone comes across comments from professionals on the same level at Terrance Tao, I'd love for them to share it.
370.
by j_maffe 2y ago
Yeah, I think the analogy here would be how an experienced programmer would write code if asked to write on the spot given a specific prompt with no time to think.
371.
by j_maffe 2y ago
Retired people don't innovate, simply because they don't work.
372.
by j_maffe 2y ago
It's interesting to know what that person is up to (from what they publically share online).
373.
by j_maffe 2y ago
At some point you need to be pragmatic about minuscule concessions for a net positive action. Otherwise you would be unable to do anything at all.
374.
by j_maffe 2y ago
That's an even more American proxy to use lol
375.
by j_maffe 2y ago
I think once a small corpus of examples of CoT gets around, people will be able to reverse-engineer it.
376.
by j_maffe 2y ago
Too bad for them that they're actively failing at keeping their moat. They're consistently ahead by barely a few months, not enough to hold a moat. They also can't trap customers as chatbots are literally the easiest tech to
377.
by j_maffe 2y ago
First of all this is irrelevant to GP's comment. Second of all, while these products do have net negative impact, we as a society knew about it and failed to act. Everyone is to blame about it.
378.
by j_maffe 2y ago
That's just a summary, not the actual CoT
379.
by j_maffe 2y ago
The fact that it being free is used as an argument about its low quality is the most American thing I've read in a while.
380.
by j_maffe 2y ago
That is exactly how I describe it to others. LLMs have mostly figured out the fast system but it's woefully bad at system 2. I'm surprised this kind of analogy isn't common in discussions about LLMs.
381.
by j_maffe 2y ago
Both games are incredible but Gorogoa is a masterpiece. It attempted to communicate something about humanity that couldn't have been expressed in any other form.
382.
by j_maffe 2y ago
It does mean something. It means that the model is still more on the memorization side than being able to independently evaluate a question separate from the body of knowledge it has amassed.
383.
by j_maffe 2y ago
Alternatively, these might not be "filler words", but instantiate paths of reasonsing.
384.
by j_maffe 2y ago
They almost definitely used existing models for generating it. The human feedback part, however, is the expensive aspect.
385.
by j_maffe 2y ago
It's not about the literally quantity/value, it's about the order of growth of output vs input. Hardware and optimizations don't really change that.
386.
by j_maffe 2y ago
To be fair the world is still as addicted to IG and FB as before, it's just been normalized at this point and FB is having an aging demographic.
387.
by j_maffe 2y ago
But spare parts and maintenance is something that many regulators pressure suppliers about, and for good reason. I feel like a lot of people (especially on HN) don't understand the role of regulation and how it can help the end user.
388.
by j_maffe 2y ago
It's hard to believe they're offering a lower cut for any reason other than to attract more offerings. It has to be the main reason at least.
389.
by j_maffe 2y ago
https://github.com/meta-llama/llama3/blob/main/llama/model.p...
390.
by j_maffe 2y ago
The source code is available though.

More ›