My instinct is that this should be smaller and easier than the Stable Diffusion I run on my PC, but maybe I am just super wrong about that?
I believe large language models take much more VRAM for generation than image models. For example, the open model BLOOM requires 352GB. So it's not realistic to do if on your local machine at the moment.
The only project I've seen along these lines is https://github.com/LAION-AI/Open-Assistant, but I don't think it's real yet.
I was asked to review a comment that was just "comment deleted by user." I gave it a neutral, just wanted to point out that if the comment was deleted because it was heinously awful, it may mess up your calculation of whether people are good meta-mods if that isn't taken into account. And also, kind of pointless to spend reviews on deleted comments.
It's worth noting that Sam Altman says this is unintended. Clearly this will be little succor to many of the commenters in this thread, but I was shocked when I first read the original tweet and Sam Altman walking it back makes me feel better.
Only a couple hours after asking this, I saw this article that seems to resolve my question if accurate. Their summary is that my confusion was well warranted, it's literally just that they committed a $4m as "collateral" that's forfeit if he doesn't show up, and a promise to pay the full amount.
Edit: additional interesting discussion on HackerNews
Sam Bankman-Fried had to have his parents and an unknown third-party sign a $250 million bond. In the media, it has been widely reported that this is "secured" by the parents' house, which is worth far less than $250 million, and is almost certainly worth less than $25 million which seems to be another magic number for bail bonds in this court. The most detailed article I could find on the mechanics was this one but I'm hoping a lawyer here could actually clarify the situation.
-
Does 10% collateral mean that the parents and the third-party person are literally taking a loan out against the house and issuing the court a check for $25 million? In the media, they have mostly been saying that the house is being put up as collateral, which doesn't imply to me that this is the case.
-
In the case where he skips bail, the parents and this other person are on the hook for $250 million. If the house is being used as collateral, wouldn't this imply that all of their assets are collateral, and so the house specifically being collateral is irrelevant?
-
Does the court assess the creditworthiness of the people signing the bond, to ensure that they actually have $250 million to pay?
In my defense, I didn't have any culture war commentary to go with it. As I said in another comment, I viewed this as being more related to site history than to culture war per se.
I put it more in the category of site history than culture war, and it didn't fit in the small-scale questions thread.
If peer effects are zero-sum, they are neither a reason to support, nor to oppose, non-public schools, right? Because it means that who your peers are has ~no effect on performance once you control for "pre-existing traits of the students and their families."
the reporter (Cade Metz) is a man
I just realized that I always assumed this was a woman as well, since it's similar to Kate.
Riddle-Master of Hed is so overlooked. It should be in top 10 lists.
This was like my second try with Stable Diffusion. What a time to be alive.
- Prev
- Next
For comparison, Stable Diffusion has 890 million parameters and GPT-3/ChatGPT has 175 billion, so about 200x. I think they probably have really good mechanisms to distribute their queries rather than a breakthrough in efficiency of inference, but I'm not super knowledgeable about this topic.
More options
Context Copy link