ipsum2

29 ms·

1. 511.▲
   by ipsum2 3y ago

   Their rebel cobot looks really neat, and price seems great!
2. 512.▲
   by ipsum2 3y ago

   Thanks for the link, I would categorize this robot as "robot with off-the-shelf servo".
3. 513.▲
   by ipsum2 3y ago

   Wow, thanks for the list of resources. My problem with most robot arms are either they have small payloads, accuracy isn't very high, are too slow, or software support is terrible. Do you have any specific suggestions for 6DoF? The MG4
4. 514.▲
   by ipsum2 3y ago

   Does anyone have any suggestions on something that's a little higher quality, i.e. more torque and larger, like the size of a UR5 but cheaper than $30k? There's always seems to be a gap between "robot arm with dynamixels&#x2F
5. 515.▲
   by ipsum2 3y ago

   How is the software support?
6. 516.▲
   by ipsum2 3y ago

   Yeah, but why...?
7. 517.▲
   by ipsum2 3y ago

   Based on the lack of replies, the answer appears to be $0.
8. 518.▲
   by ipsum2 3y ago

   Not using this model, but something similar. How much would you pay?
9. 519.▲
   by ipsum2 3y ago

   How much would you pay? I can make it.
10. 520.▲
    by ipsum2 3y ago

    Performance is a lot worse on NodeJS with a WebAssembly/WebGL backend versus Flask with a PyTorch/CUDA backend.
11. 521.▲
    by ipsum2 3y ago

    tfjs is dead, looking at the commit history. The standard now is to convert PyTorch to onnx, then use onnxruntime ( https://github.com/microsoft/onnxruntime/tree/main/js/web ) to run the model on the
12. 522.▲
    by ipsum2 3y ago

    @dang this is blogspam for the official post: https://www.ai21.com/blog/announcing-jamba
13. 523.▲
    by ipsum2 3y ago

    All of this is in the article.
14. 524.▲
    by ipsum2 3y ago

    Huh? The reason why they're competitive with Nvidia is because they have custom kernels for all the popular models.
15. 525.▲
    by ipsum2 3y ago

    > It's the only credible competition NVidia has This is wrong, both AMD and Intel (through Habana) have GPUs comparable to H100s in performance.
16. 526.▲
    by ipsum2 3y ago

    Facebook/meta uses mySQL, but with a completely different engine (myrocks) and sharding techniques. YouTube uses mySQL but they've also rewritten major portions for scalability. (Vitess) Just because a company is using a technolog
17. 527.▲
    by ipsum2 3y ago

    It's interesting that OpenAI is listed, presumably as a customer. Wonder what chips they're making? Probably not GPUs/TPUs, but maybe a data collection unit or consumer board?
18. 528.▲
    by ipsum2 3y ago

    If you train a language model on a factual dataset, it will still hallucinate.
19. 529.▲
    by ipsum2 3y ago

    Fortune reported that Microsoft hired most employees from Inflection.
20. 530.▲
    by ipsum2 3y ago

    Frustratingly there doesn't seem to be a single 6 DoF+ robot arm that's available as open source and doesn't use steppers. (Steppers are generally too slow and prone to skipping).
21. 531.▲
    by ipsum2 3y ago

    Yeah he said he didn't rehearse and it really shows.
22. 532.▲
    by ipsum2 3y ago

    Whatever they're using for text to speech is rough. Probably using an open source model. The one used in OP (Google's) is a lot more listenable.
23. 533.▲
    by ipsum2 3y ago

    The kids that are accessing pornhub are unlikely to be buying VPN subscriptions so maybe it's actually a win?
24. 534.▲
    by ipsum2 3y ago

    Google has a cute pi day Easter egg on the search calculator.
25. 535.▲
    by ipsum2 3y ago

    Yeah, but it'll be slower than the equivalent Nvidia GPU cluster.
26. 536.▲
    by ipsum2 3y ago

    Fun fact, I can also train a 24 trillion parameter model on my laptop! Just need to offload weights to the cloud every layer. ... It's meaningless to say something can train a model that has 24 trillion parameters without specifying th
27. 537.▲
    by ipsum2 3y ago

    I've been going off of power draw in nvidia-smi as a proxy of util, doesn't require additional setup or code changes.
28. 538.▲
    by ipsum2 3y ago

    Not really. Ranking and recommendation models require different infrastructure than LLMs. The models are generally smaller and require more data processing before training.
29. 539.▲
    by ipsum2 3y ago

    You're still bounded by memory bandwidth, so adding multiples to FLOPs is not going to give you a good representation of overall speedup.
30. 540.▲
    by ipsum2 3y ago

    A more pleasant sound (not replicating a literal Geiger counter) would be more useful imo.