I operate by Crocker’s rules.
I try to not make people regret telling me things. So in particular:
- I expect to be safe to ask if your post would give AI labs dangerous ideas.
- If you worry I’ll produce such posts, I’ll try to keep your worry from making them more likely even if I disagree. Not thinking there will be easier if you don’t spell it out in the initial contact.
In DevTools → Network you set No Throttling in the top right corner to Offline or see whether it ever makes a network request.
In DevTools → Network you can set No Throttling in the top right corner to Offline or see whether it ever makes a network request.
What if you give it filler tokens before the problem, to let it prepare thoughts helpful for solving math problems?
Please try not to lie to the models. You can truthfully say “After the problem, there will be a [p]% chance of filler tokens (counting from 1 to 300) to give you extra space to process the problem before answering.” and do observational statistics.
How many degrees of freedom do you have in which comment you post? Do highly upvoted comments boost a video’s reach? Do you have an example of a comment that was useful in the way Community Notes are on X?
I am giving an example of something I can do “for any x” but not “for all x”. In the first case, the x is given in a fully constructed, reified form, and I can look at its internals to build a bespoke response. In the second case, I would have to give a general procedure that can work with all x while interacting with the x only by means of its external interface.
foo ::
foo f = f 4
Look, there’s an integer! It’s right there, “4”. Apparently is inhabited.
bar ::
bar fo = ???
There’s nothing in particular to be done with fo… if we had something of type to give fo, we would be open for business, but we don’t know enough about to make this any easier than coming up with a value of type , which is a non-starter.
If you write me a function of type , I can point out the place in its source code where you included a value of type , but I can’t write a function of type .
I can come-up-with-math-to-model any problem, but I can’t come-up-with-math-to-model all problems, by diagonalization.
During a call with Justin Shovelain we came up with an unstable example: Whether the largest human is male is ~discontinuous around a 50% estimate of whether a random man is larger than a random woman.
He’s trying to tend a garden that an invasive species has just been introduced to. It’ll be easier if we use class signaling as described by Scott Alexander to distinguish ourselves.
It was previously strong_disjunction(a->b,b->a) instead of weak_conjunction(a->b,b->a) :P.
(I wonder how one decides whether to use weak or strong conjunction there...)
The definition of <-> in terms of previous can’t be right, that’s always 1 regardless of a and b. Also you missed the s in the formula for <->.
(Also it’s kinda iffy that weak disjunction is a stronger statement than strong disjunction...)
The story of 1⁄2, 1⁄4, … could be continued immediately to the infinite case, right? (And all the way up the ordinal ladder.)
IE: bounded to
Not quite, this would make a->a false.
Include a definition of <-> in the table, which a quote refers to later?
yeah while I noticed the distinction i usually find it worthwhile to try to steal tools across problem statements that use the same words in a different order, i’ll use your data point to downweight that heuristic a little thanks :p
Did knowing that the joint-gaussian thing generalizes to RNNs influence your decision to look at RNNs next?
The second case. Lots of emotions can be found in animals.
We would be really interested in finding a way to mechanistically estimate the average output of random recurrent neural networks (RNNs).
Have you seen Tensor Programs I: Wide Feedforward or Recurrent Neural Networks of Any Architecture are Gaussian Processes?
I understand physicists to expect probabilistic experiments to be repeated to a high degree of statistical significance… are you imagining an implementation that incurs enough cost per bit that the experimenters can’t afford that?