fzzzy

14 ms·

1. 151.▲
   by fzzzy 2y ago

   Waffle machines is way better. Love it. Thanks.
2. 152.▲
   by fzzzy 2y ago

   Lots of lake effect snow there.
3. 153.▲
   by fzzzy 2y ago

   Lansing has been the epitome of what the rust belt means for decades already.
4. 154.▲
   by fzzzy 2y ago

   Plenty of people don't think with an internal monologue or internal imagery.
5. 155.▲
   by fzzzy 2y ago

   I guess this gives Facebook even more incentive to re-incorporate in Texas and move the headquarters there.
6. 156.▲
   by fzzzy 2y ago

   This is what Ben Stein is lecturing about in Ferris Beuller's Day Off.
7. 157.▲
   by fzzzy 2y ago

   Keep fighting the good fight. Saying Llama is open source is straight up lying. It's open weights.
8. 158.▲
   by fzzzy 2y ago

   This isn't true. Learning how to train a 124M is just as useful as a 700B, and is possible on a laptop. https://github.com/karpathy/nanoGPT
9. 159.▲
   by fzzzy 2y ago

   It wouldn't cost tens of millions of dollars to create a 500m or 1b, and the process of learning is transferrable to larger model weights.
10. 160.▲
    by fzzzy 2y ago

    The data pipeline to build the weights is the source. The weights are a binary. The term is being hijacked. Just call it open weights, not open source models. The source for the models is not available. The weights are openly available.
11. 161.▲
    by fzzzy 2y ago

    how many people in the world are used to deploying external client facing applications?
12. 162.▲
    by fzzzy 2y ago

    they are not saying meta didn't use pirated content, just that they have the resources not to if they choose.
13. 163.▲
    by fzzzy 2y ago

    Yes, you should say "could you please write and execute a program to count the number of "r" characters in the string "strawberry"
14. 164.▲
    by fzzzy 2y ago

    The llm has absolutely no way of knowing which characters are in which token.
15. 165.▲
    by fzzzy 2y ago

    Abandoned by apple before it was finished.
16. 166.▲
    by fzzzy 2y ago

    Brendan Eich always used to say that ecmascript sounds like a skin disease.
17. 167.▲
    by fzzzy 2y ago

    Apple's devices are also willing to work without an account.
18. 168.▲
    by fzzzy 2y ago

    This is solveable. For every high frequency poster, show them in the feed at the time of their oldest of a batch of posts in duration X. Reveal all these posts with ui at that point in the timeline. Tune duration X.
19. 169.▲
    by fzzzy 2y ago

    The Mac 128k in 1984 had unified memory.
20. 170.▲
    by fzzzy 2y ago

    Yes, that's what I was doing. Thanks for the info.
21. 171.▲
    by fzzzy 2y ago

    Thanks, good to know. Perhaps it is different for diffusion; with llms, layers are generally split across gpus, meaning inference has to happen on one gpu before the values can be passed between the layer split.
22. 172.▲
    by fzzzy 2y ago

    You sure? In my experiments with multi gpu inference, I couldn't get anywhere close to max theoretical power draw.
23. 173.▲
    by fzzzy 2y ago

    I have aphantasia (self-reported, I guess, is there an objective marker? Brain scan?) and I have extremely vivid audio imagery.
24. 174.▲
    by fzzzy 2y ago

    Could you speculate as to why or how?
25. 175.▲
    by fzzzy 2y ago

    Currently, they can't say "I don't know." I think it may be possible to fix this.
26. 176.▲
    by fzzzy 2y ago

    #define begin { #define end }
27. 177.▲
    by fzzzy 2y ago

    Cool, thanks a lot for the explanation. Makes sense.
28. 178.▲
    by fzzzy 2y ago

    Models in practice haven't been deterministic at 0 temperature, although nobody knows exactly why. Either hardware or software bugs.
29. 179.▲
    by fzzzy 2y ago

    You're right they did add anonymous access at some point, but it was quite a while ago I think. Smart move on their part. Makes casual use much more convenient.
30. 180.▲
    by fzzzy 2y ago

    You can also go to the website and use it without logging in.