techwizrd

8 ms·

1. 61.▲
   by techwizrd 3y ago

   I would be very interested to see how this compares across demographics (think NHANES). I wonder if it could be turned into a check where clinicians measure baseline performance and use it to track reduction in performance over time. If it
2. 62.▲
   by techwizrd 3y ago

   I've found Lets-Plots [0] to my least painful plotting option for Python so far. I'm so used to the grammar of graphics from ggplot, especially compared to plotting in matplotlib or seaborn. 0: https://lets-plot.org&#x2
3. 63.▲
   by techwizrd 3y ago

   I would applaud a toggle or install screen to give people a choice. But many, myself included, find the Samsung OS customizations an improvement over stock Android.
4. 64.▲
   by techwizrd 3y ago

   I was frustrated with jumping to frecent directories, so I "reverted" back to jumping back to bookmarked directories [0] (which I find much more predictable). 0: https://github.com/techwizrd/fishmarks
5. 65.▲
   by techwizrd 3y ago

   I understand completely. I even wrote an offline documentation browser [0] for Linux similar to Dash, and I reflexively search online too. It's a hard habit to break, but I think it's a UI/UX issue. 0: https://gith
6. 66.▲
   by techwizrd 3y ago

   I agree. This is the kind of thing I'd love to see more of on the web.
7. 67.▲
   by techwizrd 3y ago

   I'm a paying customer and mine was enabled as well, and this is the first I'm even hearing about it. That's not a good feeling.
8. 68.▲
   Lean Co-pilot for LLM-human collaboration to write formal mathematical proofs(twitter.com)

   4 points by techwizrd 3y ago|0 comments
9. 69.▲
   by techwizrd 3y ago

   There are a lot of distributed data parallel and federated learning (FL) algorithms that could be applied to training LLMs, and there have been several papers that tackle applying these. I don't think I've seen 500 steps in the FL
10. 70.▲
    by techwizrd 3y ago

    I think the public outcry made this outcome inevitable, and this is good to see!
11. 71.▲
    by techwizrd 3y ago

    Every update to this story feels like it's being trickled out.
12. 72.▲
    by techwizrd 3y ago

    I used to work on a healthcare AI chatbot startup before traditional LLMs like BERT. We were definitely worried about accuracy and reliability of the medical advice then, and we had clinicians working closely to make sure the dialog trees w
13. 73.▲
    by techwizrd 3y ago

    I don't think I'd want my representatives voting based on chatbot hallucinations, but it sounds like many of them are already voting based on misinterpretations, lobbying, media pressure, and blank sheets of paper. The government
14. 74.▲
    by techwizrd 3y ago

    Reconciling inconsistent information across systems is the truly hard part. Sometimes I feel that a large part of our jobs are simply detective work.
15. 75.▲
    by techwizrd 3y ago

    On the contrary, that's why I find Keras 3 interesting (although I don't see myself switching anytime soon). It supports both PyTorch and JAX, and Keras Core models/layers can be used interchangeably with PyTorch modules: ht
16. 76.▲
    by techwizrd 3y ago

    I've been using mathpix for several years, and it works really well.
17. 77.▲
    by techwizrd 3y ago

    K-9 Mail joined Thunderbird to become Thunderbird Mobile [0]. They regularly publish progress updates [1] and their GitHub repository is regularly publishing releases and merging fixes and features [2]. 0: https://blog.thunderbir
18. 78.▲
    by techwizrd 3y ago

    Same! The page listing all of the applications has also introduced me to new terminal emulators and other things. It's great to have something consistent I don't need to fiddle with across all my systems.
19. 79.▲
    by techwizrd 3y ago

    Agreed. Employers should provide, manage, and support work machines. I would never install employer-required spyware on my personal machine. Similarly, the only time I used my own laptop was for a scrappy startup where I was the first engin
20. 80.▲
    by techwizrd 3y ago

    I've had a pair of Enchroma glasses for at least 3-4 years, and I was just at the eye doctor preparing to get another pair. They've been quite a help, especially while reading plots prepared by folks who aren't familiar with
21. 81.▲
    by techwizrd 3y ago

    I agree completely. We always end up moving away from Prophet every time. The results from Prophet are just not very good, although it can be useful for a proof-of-concept.
22. 82.▲
    by techwizrd 3y ago

    I have my Mastodon account on mastodon.social and my Lemmy account on programming.dev because they're very unlikely to be defederated and they're drama-free. I really like them for what its worth.
23. 83.▲
    by techwizrd 3y ago

    How should defamation be handled? It is good to have recourse against baseless accusations, but we also want to prevent chilling effects discouraging fact-checking. If handled privately, then the public may not even be aware of falsified in
24. 84.▲
    by techwizrd 3y ago

    Are these pedestrian detection models in use in any widely-deployed commercial self-driving car? Is there a limitation since these are images rather than videos? I would've expected these to be addressed in the "Threats to Validit
25. 85.▲
    by techwizrd 3y ago

    We use KL-divergence to calculate how surprising a time-series anomaly is and rank them for aviation safety, e.g., give me a ranked list of the most surprising increases in a safety metric. It's quite handy!
26. 86.▲
    by techwizrd 3y ago

    I really like seeing articles or papers that describe the current advances and open challenges in a sub-field (such as [0]). They're underappreciated, but good practice or reading for folks wanting to get in the field. They're als
27. 87.▲
    by techwizrd 3y ago

    I use a fine-tuned BERT-like model for NER, but I'd be interested to compare how it performs.
28. 88.▲
    by techwizrd 3y ago

    This looks really impressive, and I'm excited to see how it performs on our data! P.S., I think the name conflicts with Hydra, the configuration management library: https://hydra.cc/
29. 89.▲
    by techwizrd 3y ago

    I built one of these as a hackathon during undergrad. It was surprisingly easy to get something working, even before modern DL techniques, in under 24 hours. I'm not surprised this has taken off.
30. 90.▲
    by techwizrd 3y ago

    There is so much "secret sauce" that we're unconsciously relying upon. I learned a lot of secret sauces by reading through various implementations, "logbooks" [0] by various researchers, and so on. It definitely mak