binarymax

8 ms·

1. 481.▲
   by binarymax 3y ago

   You might want to take a look at the article before commenting ;)
2. 482.▲
   by binarymax 3y ago

   What confusion? I know exactly how BM25 works and how Transformers work. I stated a hypothesis and asked if anyone has tried it. You say it won’t work. That’s just your opinion. Do you have proof or evidence? This is science. Dismiss
3. 483.▲
   by binarymax 3y ago

   Attention in transformers works because over time the model learns token importance based on frequency and context. If you don’t have attention and need a fast substitute for “forgetting” non important tokens, then BM25 is an intuitive hypo
4. 484.▲
   by binarymax 3y ago

   I hadn’t heard of Mamba before reading this article, and I was wondering if anyone has tried setting importance of a token as a TF-IDF or BM25 lookup. Requires a first pass to construct the token index but otherwise it seems like it would
5. 485.▲
   by binarymax 3y ago

   If a dog can learn 100 toys, then they can learn 52 cards, and maybe play some poker.
6. 486.▲
   by binarymax 3y ago

   Travel to a point in the path and view from the comfort of the ground :) https://eclipse2024.org/eclipse_cities/statemap.html
7. 487.▲
   by binarymax 3y ago

   Cool! Which OCR engine/model do you use?
8. 488.▲
   by binarymax 3y ago

   Yes, but you can opt-out of the data retention and abuse monitoring if you go through a vetting process and are deemed low risk.
9. 489.▲
   by binarymax 3y ago

   The GPU product page: https://www.nvidia.com/en-us/design-visualization/rtx-4000-s... 20GB RAM, can fit 13B param models, and maybe some quantized larger models.
10. 490.▲
    by binarymax 3y ago

    Azure's TOS are different. You can also opt-out of data retention: https://learn.microsoft.com/en-us/legal/cognitive-services/o...
11. 491.▲
    by binarymax 3y ago

    Thanks! I’ve been using OSMs free Nominatim for testing but need to move off it to support a higher query volume - I had no idea it was going to be so expensive, so this comes at the perfect time!
12. 492.▲
    by binarymax 3y ago

    Which geocoder is $100/mo? Nominatim? Which cloud? Either way this looks really cool and I’ll give it a try!
13. 493.▲
    by binarymax 3y ago

    Apple PC/Laptop hardware far exceeds that of Dell.
14. 494.▲
    by binarymax 3y ago

    The Catalina wallpaper threw me off guard when I first saw it. It looks exactly like a Bryce landscape I made in the late 90s. I spent some time trying to find it in my data archives but alas it is likely lost.
15. 495.▲
    by binarymax 3y ago

    They were on both iOS and their Mac
16. 496.▲
    by binarymax 3y ago

    Totally agree. The only problem is that batteries are so expensive you need to align yourself with a brand so that you can swap batteries between tools. I wish there was a standard battery spec, so that they were interoperable between bra
17. 497.▲
    by binarymax 3y ago

    I just want to be able to search my chats. I have hundreds now.
18. 498.▲
    by binarymax 3y ago

    Thanks for the response. The linked introduction does help. The workload page does give me that impression (and based on upvotes of my post it does to others as well)...so perhaps disambiguating that the void test*() examples on the workl
19. 499.▲
    by binarymax 3y ago

    I got really excited about this, and I spent a little time looking through the documentation, but I can't figure out how this is different than randomizing unit tests? It seems if I have a unit test suite already, then that's 99%
20. 500.▲
    by binarymax 3y ago

    Best practice is to talk about net job loss, or net job creation. "Augmenting" or "Replacing" jobs doesn't mean anything. Technology has been "replacing" jobs for hundreds of years, that's the natu
21. 501.▲
    by binarymax 3y ago

    10 years
22. 502.▲
    by binarymax 3y ago

    Hard for me to read this without repeating “Hooked on phonics worked for me!” in my head. Nice to see they are still around. https://www.hookedonphonics.com/
23. 503.▲
    by binarymax 3y ago

    Maybe some thin plexiglass around it would be fine - not sure if it would compromise the mechanism though.
24. 504.▲
    by binarymax 3y ago

    Pretty much this exact scene was predicted by Ronnie Chang as a comedy bit. https://m.youtube.com/watch?v=BN6aUgMtAos
25. 505.▲
    by binarymax 3y ago

    Sorry is craftsman dead? Maybe they’ve changed but they seem alive and well - https://www.craftsman.com/
26. 506.▲
    by binarymax 3y ago

    I’m going to throw my hat in the ring to say FirefoxOS and the phone (of which I bought the first beta version) were IMO great ideas and they should have stuck with it. The iOS/Android duopoly really needed a web-SPA option. Maybe th
27. 507.▲
    by binarymax 3y ago

    That's one thing that horns are for. Granted I hate using it, and other people hate it too, but I've gotten pretty good at a friendly beep-beep instead of a lean on honk.
28. 508.▲
    by binarymax 3y ago

    PCA doesn’t work at all for embeddings. Maybe in some rare cases, but it’s throwing away knowledge and the loss in accuracy is usually quite drastic.
29. 509.▲
    by binarymax 3y ago

    Lots of word game dictionaries don’t contain proper nouns. Ponzi is an interesting example because it’s now used as the definitive for what it is, so words like that are hit or miss. Other examples that might or might not be in dictionari
30. 510.▲
    by binarymax 3y ago

    Finally, a reasonable output dim size. 1536 was just nuts, and added 4x the cost of HNSW in RAM compared to something like the e5 small models at 384 dims.