ipsum2

39 ms·

1. 661.▲
   by ipsum2 3y ago

   I suspect that a Python DSL on top of Lean4 would be hugely popular, similar to PyTorch replacing existing Torch, which was written in Lua.
2. 662.▲
   by ipsum2 3y ago

   You can replace DJI with any other consumer electronics manufacturer. It's hard to compete with China's vertical integration + cheaper labor. Skydio (Californian startup) used to manufacture consumer drones but recently exited tha
3. 663.▲
   by ipsum2 3y ago

   I don't understand when people make up fake quotes.
4. 664.▲
   by ipsum2 3y ago

   ByteDance does a lot of ML research on music/video.
5. 665.▲
   by ipsum2 3y ago

   Azure doesn't use FPGA for ML, they use Nvidia like everyone else. Azure used FPGAs a few years ago for network switches though.
6. 666.▲
   by ipsum2 3y ago

   This is wrong, NVLink is crucial for tensor parallelism in models for training and in large (>40B param) models for inference.
7. 667.▲
   by ipsum2 3y ago

   Development and reverse engineering probably takes ~500 hours. They probably won't sell more than 1,000 (I'm making these numbers up). Cost for a small run of boards with component placement, with shipping is probably $20-30. They
8. 668.▲
   by ipsum2 3y ago

   Isn't FasterTransformer (NVidia, OSS) and text-generation-inference (Huggingface, not OSS) are faster than this?
9. 669.▲
   by ipsum2 3y ago

   I think the infamous Mitubishi toaster has the Sunbeam beat. A $300 machine that supposedly produces the best toast in the world, but only toasts one slice at a time. https://gizmodo.com.au/2020/03/mitsubishi-makes
10. 670.▲
    by ipsum2 3y ago

    In a similar vein, UPenn professor Angela Duckworth of "Grit" fame was caught misrepresenting statistics to embellish her claims a few years ago: https://www.npr.org/sections/ed/2016/05/25/
11. 671.▲
    by ipsum2 3y ago

    If you don't want to listen to it, the podcast description is a nice summary: https://www.npr.org/2023/07/27/1190568472/dan-ariely-frances...
12. 672.▲
    by ipsum2 3y ago

    I agree that Russ Tedrake is great, I've watched his online lectures on underactuated robotics.
13. 673.▲
    by ipsum2 3y ago

    Unpopular opinion, most of MIT's research claims are preposterous and can be safely ignored. Their undergraduate education is quite good, but the research output is quite shoddy. The only exception is their computational photography gr
14. 674.▲
    by ipsum2 3y ago

    There's a million different language model (not wrapper) companies offering APIs already. OpenAI, Anthropic, Cohere, Google, etc. It wouldn't be profitable.
15. 675.▲
    by ipsum2 3y ago

    From the page: Relation to X Corp We are a separate company from X Corp, but will work closely with X (Twitter), Tesla, and other companies to make progress towards our mission.
16. 676.▲
    by ipsum2 3y ago

    Yup. https://news.ycombinator.com/from?site=x.ai
17. 677.▲
    by ipsum2 3y ago

    And yet he hires women to be CEOs at SpaceX and Twitter. Maybe it's not a gender thing?
18. 678.▲
    by ipsum2 3y ago

    I think I'm stupid, the ending didn't make any sense to me.
19. 679.▲
    by ipsum2 3y ago

    Not sure how A C, D are relevant, trademarks pertain only to a specific category. Meta Threads was trademarked for luggage and backpacks. It'll be interesting to see if threads.com has a trademark.
20. 680.▲
    by ipsum2 3y ago

    Following the same analogy, it's like removing spreadsheets from Google Sheets.
21. 681.▲
    by ipsum2 3y ago

    Google Plus (invite only) failed in a similar way.
22. 682.▲
    by ipsum2 3y ago

    Instagram Threads was a (different) thing back in 2019: https://www.theverge.com/2019/8/26/20833903/facebook-instagr... . Threads.com looks like it came out in 2022?
23. 683.▲
    by ipsum2 3y ago

    I believe it's because the model architecture isn't added to Huggingface transformer library, so it needs to eval some python code (i.e. load a pickle) to create the PyTorch model. Have not noticed it to be specific to models from
24. 684.▲
    by ipsum2 3y ago

    Possibly, but they wrote those protections in ~5 hours? Seems dubious.
25. 685.▲
    by ipsum2 3y ago

    I'm very impressed that you were able to find this.
26. 686.▲
    by ipsum2 3y ago

    This makes no logical sense. Why would the scraping not restart after its unlocked? More realistically, he got a lot of backlash from users and website owners where embedded tweets suddenly stopped showing up.
27. 687.▲
    by ipsum2 3y ago

    Looks like they even blocked the Google crawler for individual accounts. https://www.google.com/search?q=site:twitter.com sort by last hour.
28. 688.▲
    by ipsum2 3y ago

    > Please make sure the function does not rely on any local variables, including imports (which should be moved inside the function body) This seems like a major limitation and pretty antithetical to the PyTorch approach.
29. 689.▲
    by ipsum2 3y ago

    ...what?
30. 690.▲
    by ipsum2 3y ago

    There's no such thing as a perfect third in music theory, only perfect 4ths and 5ths, which explains the 4:5 ratio. There's major/minor (imperfect) thirds. Interesting test of LLMs, this one failed.