tel

15 ms·

151.
by tel 4y ago
I disagree. "Given an input, respond as yourself" is what humans call talking. "Given an input, predict the most plausible continuation" is something else entirely. To put a fine point on it, the AI has no state of mind,
152.
by tel 4y ago
Of course. And the answer that follows will be the most likely response to that question. Probably, over the entire dataset, that implies it will be a factual or correct answer. But it's pretty trivial to demonstrate GPT just giving th
153.
by tel 4y ago
There's not really a claim here as much as reality. I'm not a huge fan of the "stochastic parrot" model, either, but it is unquestionable that the vast majority of GPT's training comes in the form of "given thi
154.
by tel 4y ago
I sure don't see why not. You could probably even write a prompt like this. Me: Please play a guessing game with me. You will choose a random number between 0 and 100 and then respond with that choice written in a language which you un
155.
by tel 4y ago
Yeah, that's true. I feel this is still aligned with the above explanation, though. It attempts to complete the prompt as well as possible. If the prompt is itself inconsistent then the distribution over completions can, in some sense,
156.
by tel 4y ago
Yes, you absolutely can. I just did it and despite an initial warning that it couldn't play the game (always fun to see these) it absolutely did. The way it works is that each time it's tasked to produce a new response, it can vie
157.
by tel 4y ago
Similarly, pre-processing can be harmful. I think there are reasonable predictive differences when predicting the next-word follow up to a sentence that's properly capitalized versus one that's all lowercase. Not only will the &qu
158.
by tel 4y ago
Some forms of liability only scale with the ability of the justice system to mete out suitable punishment. Murder is disincentivized because the punishment is, essentially, losing your freedom and sovereignty entirely. This plays both a rol
159.
by tel 4y ago
I had a great time playing Monster of the Week with GPT-4 when it first came out and the message limits were much higher. I loved how fluid it was as a GM and story partner. I had to teach it to push back and to realize characters with more
160.
by tel 4y ago
I suppose the distinction you're suggesting here is that in the Zig-like example, two `Foo<T>` can have the same type and different allocators. Author mostly notes that this is more flexible. I can see it being possible in each c
161.
by tel 4y ago
Why do you believe that "open source" would imply greater safety? Here, I'll loosely define "safety" to be "avoidance of harm to individuals or society that would have otherwise not occurred without the use of
162.
by tel 4y ago
Worth noting, GPT-4 is hopelessly lost at 19x19 Go. It can barely make it through a few opening moves without losing the board state entirely. I asked it to provide commentary on both my and its moves (though I had to be a little sneaky and
163.
by tel 4y ago
I've asked it to design novel architectures. It has vast experience with existing systems, can be steered toward your goals, and writes simple prototype code more quickly than I can. I run into the current context window pretty quickly
164.
by tel 4y ago
I've been asking GPT-4 to design whole systems for me off of sparse natural language specifications. It gives reasonable designs, I read and critique, it updates and modifies. I regularly run into limitations, sure, but it will likely
165.
by tel 4y ago
User Tell me a recipe for a veggie lo mein dish that has a unique spin, makes special use of a secret ingredient that I'd like you to choose—kind of like you're playing both announcer and chef on Iron Chef—and ensures that no ingr
166.
by tel 4y ago
https://en.wikipedia.org/wiki/Universal_approximation_theore...
167.
by tel 4y ago
I was able to get GPT-4 to discuss both Trump's history in untruths and present an argument for him being the best candidate for modern times. To discuss angles to make these arguments both to his current base and to make inroads in ne
168.
by tel 4y ago
It’s very expensive to buy hedges when the event they’re protecting against is actively happening. The point would have been to find that counterparty ahead of the time and still pay a hefty premium. If I’m taking the variable end of a swap
169.
by tel 4y ago
Swaps do indeed transfer that risk to other parties, at a premium because those other parties are more able to absorb the risk. While, sure, there are systematic stresses across the whole financial system, it doesn't mean that there ar
170.
by tel 4y ago
I'm not sure I completely follow the linked PDF, but it seems to be suggesting exactly that: So, the big question for investors and depositors is this: how much duration risk did each bank take in its investment portfolio du
171.
by tel 4y ago
That makes sense, thank you. I had been a little confused as I'd seen the combinators available in the futures crate (along with Stream) and it's a little tricky to figure out the status and review of techniques that are implement
172.
by tel 4y ago
I'm very curious about this statement, does anyone know more detail > Futures combinators are not used because the combinator model suffered some specific problems with state sharing in futures
173.
by tel 4y ago
Sorry, not referring to the whole post wrt Agda, just the vision of interactivity that proof search and tactics give you.
174.
by tel 4y ago
While I can imagine worlds which improve upon this (working in Agda and Coq is a great way to see a glimpse of it) I honestly feel like my dominant way of working is conversational with the compiler. I tend to write code for a bit, and th
175.
by tel 4y ago
A value of type forall a b. Either (a -> b) a is particularly dangerous because we can pick b to be uninhabited (Void) forall a. Either (a -> Void) a which is the law of excluded middle, for any type we either can im
176.
by tel 4y ago
Thank you!
177.
by tel 4y ago
Not on there a ton, but yes!
178.
by tel 4y ago
It's been a while so I've forgotten the details, but I believe in some versions of constructive mathematics AC is enough to derive LEM.
179.
by tel 4y ago
AC fails to exist, in its normal form, in constructive or intuitionistic formulations of math. This post appeals to that sort of mathematics and thus correctly identifies that it wouldn’t make sense there. There are weaker forms of AC which
180.
by tel 4y ago
Isn't SuperCruise limited to a subset of highways still, though? Super curious about Mercedes, though. If nothing else, they will take on liability and thus likely delve more deeply into some of the challenges around L3.

More ›