davesque

10 ms·

121.
by davesque 2y ago
3. Why was Russia uniquely exempt from tariffs?
122.
by davesque 2y ago
I think it's because science, knowledge, and truth are the great equalizers. In the limit, they put everyone on an equal playing field, which would also eliminate the advantage of wealth.
123.
by davesque 2y ago
Yep, and AI agents essentially throw up a boundary blocking the user from understanding the capabilities of the system they're using. They're like the touch screens in cars that no one asked for, but for software.
124.
by davesque 2y ago
Weirdly, as cool as this looks, it's a bit concerning to me. It feels like this is marking a milestone in the history of a great open source project where they are doing one or many of the following: 1) Biting off more than they can ch
125.
by davesque 2y ago
> This is not a screed against executive wealth. After all, I paid with 25 years of my life and I got some of the wealth. That's the point though isn't it? He retired at 50. Most of us will work to at least age 65 (perhaps un
126.
by davesque 2y ago
I thought he was going to say the really danger is hallucination of facts, but no.
127.
by davesque 2y ago
Seems safe to assume that the DOGE kids probably scored copies of the data they had access to even if that access was or will be revoked, especially considering that some of them have a history of data theft as mentioned in the article.
128.
by davesque 2y ago
Never a better example of Dunning-Kruger than the DOGE kids.
129.
by davesque 2y ago
Does anyone know why this is happening? A result of some initiative by the Trump administration probably?
130.
by davesque 2y ago
I'm doubtful that this would actually happen, but I can't think of anything else that could do more to reverse the trend of declining trust in OpenAI.
131.
by davesque 2y ago
I recently thought of a related question. Actually, I'm almost certain that foundation model trainers have thought of this. The question is to what extent are popular modern benchmarks (or any reference to them, or description of them,
132.
by davesque 2y ago
That's one way to look at it. Another is that a lab leak makes the government seem much more culpable, since they would presumably have more direct control over the activities of government funded scientists. In any case, I think world
133.
by davesque 2y ago
Not sure how that would work exactly. The math behind transformer models is really basic. And the world is different now than it was during the cold war. There was no internet, for example. Also, given that this was during an interview in w
134.
by davesque 2y ago
"of course Trump is triggering for many" As though that's all there is to say. Unbelievable. The sane washing of this man is completely unreal. I'm witnessing a complete unraveling of shared truth. He instigated a riot a
135.
by davesque 2y ago
The whole debate about whether or not he can realistically do this based on the state of the technology seems like a side issue to me. What really stands out is that he and CEOs like him would imply that a large, critical segment of their w
136.
by davesque 2y ago
Thank you for saying what I didn't have the patience to say. Although I sincerely worry that empathizing with people like this is a trap. I hate that I have to think this way, but nothing is worse than feeling like a sucker. I only hop
137.
by davesque 2y ago
I feel like there's a weird psychology to these kinds of boo hoo posts that I see from rich people lamenting their good fortune. What's the actual sub-conscious impulse to flaunt your wealth in such a way that devalues it to the 9
138.
by davesque 2y ago
I'm not just trolling when I say, is that a serious question? Like, really. How could you live in this day and age and legitimately wonder why many people on HN wouldn't like Tesla?
139.
by davesque 2y ago
I want an EV and can afford a Tesla and I will never buy anything from Elon Musk as long as I live exactly because of his behavior.
140.
by davesque 2y ago
Seems like a lot of existing models could be converted to this token parameter representation.
141.
by davesque 2y ago
I'm talking about the rows in the new K and V matrices introduced by the paper, not rows in the input sequence. The ordering of rows in the new K and V matrices does matter in the sense that rows that appear further down were added lat
142.
by davesque 2y ago
Seems like a big deal. I feel like this could enable a new level of modularity and compatibility between publicly available weight sets, assuming they use similar channel dimensions. Maybe it also provides a nice formalism for thinking abou
143.
by davesque 2y ago
Exactly what I thought of :).
144.
by davesque 2y ago
> no executive in her position ever willingly resigns to pursue different passions/spend more time with their family/enjoy retirement or whatever else Especially when they enjoy a position like hers at the most important techno
145.
by davesque 2y ago
Maybe I'm just a rotten person, but I always find these overly gracious exit letters by higher-ups to be pretty nauseating.
146.
by davesque 2y ago
Yep, could be.
147.
by davesque 2y ago
The way that LLMs hallucinate now seems to have everything to do with the way in which they represent knowledge. Just look at the cost function. It's called log likelihood for a reason. The only real goal is to produce a sequence of to
148.
by davesque 2y ago
Yes and apparently we won't have access to that chain of thought in the release version: "after weighing multiple factors including user experience, competitive advantage, and the option to pursue the chain of thought monitoring,
149.
by davesque 2y ago
"after weighing multiple factors including user experience, competitive advantage, and the option to pursue the chain of thought monitoring, we have decided not to show the raw chains of thought to users" ...umm. Am I the only one
150.
by davesque 2y ago
I always thought that one of the use cases of memory mapping was to improve multiprocessing workloads, where a group of processes don't have to duplicate the same region of a working set. In that sense, maybe it's not surprising t

More ›