theahura

10 ms·

241.
Analyzing Grok's Latest Meltdown Through Public xAI System Prompts(theahura.substack.com)
13 points by theahura 1y ago|0 comments
242.
by theahura 1y ago
reading the [dead] comments on there is a strong argument for re-disabling [dead] comments
243.
by theahura 1y ago
Note: not my link / petition. Still, I was pretty aghast at Shaun's x post and felt compelled to share.
244.
An Open Letter to the Leadership of Sequoia Capital(shaunmaguire.fyi)
9 points by theahura 1y ago|3 comments
245.
VCs Are People Too(theahura.substack.com)
1 points by theahura 1y ago|2 comments
246.
by theahura 1y ago
I debated changing the title but as the commenters below mention the title is still accurate, as is the rest of the article. Meta was under no obligation to pay that much. Alexandr could have requested that Meta spend $14b on <charity of
247.
by theahura 1y ago
> you probably dont think people that went to state schools are even human on the contrary. I have been quite vocal about why I felt my education was lacking and the respect I have for those who have gone for nontraditional paths
248.
by theahura 1y ago
O shit really? That's a massive update and possibly invalidates a lot of this article. It's also the first I've heard of this -- can you tell me more? Are you saying everyone essentially got bought out? (if you want to dm me
249.
by theahura 1y ago
- Could be wrong about Scale. I'm going off folks I know at client companies and at scale itself. - idk I've trained a lot of models in my time. It's true that there's an arcane art to training LLMs, but it's wrong
250.
by theahura 1y ago
Idk just vibes. I haven't joined any subreddits except r/civ. The feed is inscrutable
251.
by theahura 1y ago
I agree that this is a possible reason. Meta wants to move fast and m&a is too slow for their tastes. I make the case in the article though that the actual acquisition doesn't really make sense for metas core business, but I agree
252.
by theahura 1y ago
I'm a Zuck fan, and I think I could mount a compelling case for that take. But that was not the article to do so. And people can def disagree, there's good arguments on the other side
253.
by theahura 1y ago
I think there's basically the benchmarks, and then "the benchmarks". The former is, like, actual stuff you can quantify and game. And the latter is the public response. The former is a leading indicator but it's not alwa
254.
Meta Spends $14B to Hire a Single Guy(theahura.substack.com)
76 points by theahura 1y ago|72 comments
255.
LLMs are Programming Languages (2023)(theahura.substack.com)
2 points by theahura 1y ago|0 comments
256.
Apple's "Illusions of Thinking" Paper Isn't Fair to LLMs(theahura.substack.com)
1 points by theahura 1y ago|0 comments
257.
Paper Review: Deploying Big Neural Nets in 2015(theahura.substack.com)
2 points by theahura 1y ago|0 comments
258.
Code Review is not about making your code better (2023)(theahura.substack.com)
1 points by theahura 1y ago|0 comments
259.
by theahura 1y ago
Sorry, so what? A few different ways to respond to this: 1) Please do me a favor and take the GPQA benchmark. I'm curious to see how you would do. Now go find the nearest kindergartner and ask them to take it. Curious to see how they
260.
by theahura 1y ago
I disagree for two reasons. 1) the two decisions do not seem related to each other. OpenAI has capital to spend and is seeking distribution methods to shore up continued access to future capital. That strategic decision seems totally unrela
261.
by theahura 1y ago
> One key term to look for is "feature manifolds" or "multi-diemsnional features" I should probably read the updates more. Not enough time in the day. But yea the way you're describing feature manifolds and multi
262.
by theahura 1y ago
> the original circuits work was done on convolutional vision models O neat, I haven't read that far back. Will add it to the reading list. To flesh this out a bit, part of why I find circuits less compelling is because it seems int
263.
by theahura 1y ago
Well, sorta. There is some evidence to suggest that neural networks learn 'universal' features (cf Anthropic's circuits thread). But I'll openly admit to being out of my depth here, and maybe I just don't understand
264.
by theahura 1y ago
Thanks for the follow up. I've been following your circuits thread for several years now. I find the linear representation hypothesis very compelling, and I have a draft of a review for Toy Models of Superposition sitting in my notes.
265.
by theahura 1y ago
I think it's incorrect that the insight has gone nowhere. See, for eg, contrastive loss / clip, or vqgan image generation. Arguably also diffusion models. More generally, in my experience as an AI researcher, understandings of the
266.
by theahura 1y ago
It turns out a lot of disconnected spaces can be approximated by smooth ones that have really sharp boundaries, which more or less seems to be how neural networks will approximate something like discrete tokens
267.
by theahura 1y ago
It's true that there is no directly predictive model of deep learning, and it's also true that there is some trial and error, but it is wrong to say that therefore there is no operating theory at all. I recommend reading Ilyas 30
268.
by theahura 1y ago
I'm pulling directly from Chris Olah's blog post with that example. But I will say that in practice, its always surprising how increasing the dimensionality of a neural network magically solves all sorts of problems. You could use
269.
by theahura 1y ago
Possible. AGI/ASI are poorly defined. I tend to think we're already at AGI, obviously many disagree. > For example, human intelligence (the closest thing we know to AGI) requires extremely complex sensory and internal feedback
270.
by theahura 1y ago
fair, I was using 'topology' more colloquially in that sentence. Should have said 'surface'.

More ›