davesque

14 ms·

181.
by davesque 3y ago
Something I noticed when skimming the paper that is also called out by one of the reviewers: "Moreover, in the submission the authors considered only Transformers with input and output of bounded lengths, which are quite strange since
182.
by davesque 3y ago
I've got opinions on the free speech question, but notice how I wasn't making a statement about free speech. I was saying that Musk has made some exceptionally bad business decisions lately and that they're impacting his bot
183.
by davesque 3y ago
Isn't it reasonable to question the future of a platform run by a man who recently hosted a conversation between Andrew Tate and Alex Jones and also one whose advertisers are jumping ship by the day (while he tells them to go f*ck them
184.
by davesque 3y ago
It doesn't though. It just gives a better answer because prompting it that way tends to constrain the output to be more like those training examples where a better sounding answer was given. In natural language, if a request is given t
185.
by davesque 3y ago
Thanks for the comment. I do appreciate it.
186.
by davesque 3y ago
Prove it. CUDA has dominated the GPGPU scene for years now. Competitors should have had plenty of time to catch up if they had wanted to.
187.
by davesque 3y ago
Sure, although it does often seem that the convention we have around it amounts to a de facto rule. And I really struggle to see how this article passes the usual test of "satisfying intellectual curiosity". It's mostly an
188.
by davesque 3y ago
You know what? I went searching just now for examples that I thought could support my argument. But I honestly couldn't find any that really seemed unambiguous or like they couldn't be interpreted in another way. And the ones I
189.
by davesque 3y ago
It's really disappointing to see so many exceptions to the no politics rule right when we need that rule the most. There really isn't anything to be gained by discussing stories like this. We all know that atrocities are taking
190.
by davesque 3y ago
Have to disagree. LLMs in their current form are not the end all be all of AI architectures. Seems like the co-authors of the original transformer paper are in a good position to find other valuable applications of the results described i
191.
by davesque 3y ago
I mean, I remember how things were before fentanyl came around (as an outside observer, not as a user) and heroin was a horrible, life-destroying drug addiction back then as well. And I think it was and is convincingly argued that it is mu
192.
by davesque 3y ago
A big red flag for me was that Sundar was prompting the model to report lots of facts that can be either true or false. We all saw the benchmark figures that they published and the results mostly showed marginal improvements. In other wor
193.
by davesque 3y ago
The hype really is drowning out the simple fact that basically no one really knows what these models are doing. Why does it matter so much that we include auto-correlation of embedding vectors as the "attention" mechanism in thes
194.
by davesque 3y ago
You could call it clever or you could call it a spurious correlation.
195.
by davesque 3y ago
Yeah, that's what I was trying to indicate with the quotations. But thanks for adding the note.
196.
by davesque 3y ago
Agree. Another example of desensitization is the popularity (now diminishing) of Marvel movies. Relative to their visual presentation, they are empty of content. Gauging the depth of meaning in a Marvel movie takes almost no effort. I suppo
197.
by davesque 3y ago
LLMs themselves are made up of "good old fashioned" components. The public conversation about generative AI seems to assume that it's doing something fundamentally different than what came before. Of course, LLMs are still ju
198.
by davesque 3y ago
Used to get this when I was younger. It's been a while since I've experienced it. To me, it was more like I was hearing the percussionist in an orchestra build to a loud crescendo with felt mallets on a gong.
199.
by davesque 3y ago
And speaking of Graham's judgement: https://news.ycombinator.com/item?id=38384490 Of course it was flagged within a few minutes.
200.
by davesque 3y ago
I have an intuition that OpenAI's mid-range size gave the employees more power in this case. It's not as hard to coordinate a few hundred people, especially when those people are on top of the world and want to stay there. At a
201.
by davesque 3y ago
It could be that the parties involved have chosen at this moment to re-imagine whatever occurred back then in a less favorable light. Since firing is on everyone's mind, and since you can get media attention points by playing along wi
202.
by davesque 3y ago
Even if Graham supposedly booted Altman from Y Combinator, I don't see any reason to assume that a similar disagreement would have occurred in this case. Citing that history also seems to assume that Graham himself is an impeccable ju
203.
by davesque 3y ago
Perhaps you can interpret it as saying, "Yes, we're a separate business, we have a right to innovate." These are the ways that people justify lying to themselves. And it's practically a CEO's job to lie to investors
204.
by davesque 3y ago
Honestly this sounds like CEO-ish for, "Yes, we can build a similar product but don't necessarily have all the IP." If there was a simple answer to the interviewer's question, he would have given it. I feel like there&#x
205.
by davesque 3y ago
Sounds like he acted brashly on an ideological impulse and now regrets that he didn't have more self control. If so, I can empathize and I feel bad for him.
206.
by davesque 3y ago
Calling it a delusion seems too provocative. Another way to say it is that principles take agreement and trust to follow. The board seems to have been so enamored with its principles that it completely lost sight of the trust required to up
207.
by davesque 3y ago
Will you at least admit that this comment of yours, with quotes from the charter and thoughts about each quote, is contributing more to the conversation?
208.
by davesque 3y ago
Maybe I'm crazy here, but why would this be a problem? Assuming Altman still had enough time to perform his duties as OpenAI's CEO, what's wrong with developing other businesses that might help OpenAI lower its costs and achi
209.
by davesque 3y ago
What, to you, is laugh out loud funny about the parent comment? They gave a counter argument with examples from the charter and you respond with "LOL!"? How about responding with a better argument?
210.
by davesque 3y ago
If this was an ideological battle of some kind, the only hope I have is that OpenAI will now be truly more Open! However, if this was motivated by safety concerns, that would mean OpenAI would probably become more closed. And, if the only t

More ›