tehsauce

7 ms·

1. 31.▲
   by tehsauce 3y ago

   Still no GPU support. They are nowhere near a MVP.
2. 32.▲
   by tehsauce 3y ago

   Have you considered using jax? you can efficiently compute the jacobian (even using the gpu if you want) without leaving python at all. The api is also numpy compatible!
3. 33.▲
   by tehsauce 3y ago

   It's fascinating that it can model so much of the subtle dynamics, structure, and appearance of the world in photorealistic detail, and still have a relatively poor model of things like object permanence: https://cdn.openai.
4. 34.▲
   by tehsauce 3y ago

   How much does this particular result change when running in release mode?
5. 35.▲
   by tehsauce 3y ago

   I observed this phenomena not too long ago while training an RL agent to play pokemon using an intrinsic novelty reward based on the number of unique screens the agent had seen. The goal was to get it to explore the map, but instead it bec
6. 36.▲
   Training AI to Play Pokemon with Reinforcement Learning(youtube.com)

   11 points by tehsauce 3y ago|1 comments
7. 37.▲
   by tehsauce 3y ago

   Over the last couple years, I've been training an AI to play Pokemon Red. I put together a video that tells the story of the AI's learning and discusses lots of technical details. You can also download and play with the model your
8. 38.▲
   by tehsauce 3y ago

   > Apple shocked us all by addressing this head-on: The iPhone 15 line charges via USB-C instead of Lightning, and this standard USB port can do a lot
9. 39.▲
   by tehsauce 3y ago

   That’s for lossless compression, i think there’s special opportunities for multi image lossy
10. 40.▲
    by tehsauce 3y ago

    I've seen it although haven't dived deep into it. It looks like they have some interesting support for particle cell data structures, but is fairly complicated and carries limitations: https://jax-md.readthedocs.io&#x2F
11. 41.▲
    by tehsauce 3y ago

    Jax is super useful for scientific computing. Although nbody sims might not be the best application. A naive nbody sim is very easy to implement and accelerate in jax (here’s my version: https://github.com/PWhiddy/jax-e
12. 42.▲
    by tehsauce 3y ago

    “Notice that Rust has a built-in vector class, as well as iterators” Pybind with c++ will also automatically convert between python and c++ standard types
13. 43.▲
    by tehsauce 3y ago

    Oof, right as the switch demolished the xbox in hardware sales.
14. 44.▲
    by tehsauce 3y ago

    Have another look at the first example I shared. One major limitation of glsl is a lack of functional programming. In that example, we pass functions to other functions, and iterate over arrays of functions, neither of which are possible in
15. 45.▲
    by tehsauce 3y ago

    Shaderpark does a bit more than "pass around" javascript functions, it transpiles them to glsl. Believe it or not, web browsers themselves only transpile webgl shaders to other high level shader languages: hlsl on windows, msl on
16. 46.▲
    by tehsauce 3y ago

    Maybe another example will be helpful :) https://shaderpark.com/sculpture/-M_TY8dwxuyliltrFjP2 No SDFs involved in this one.
17. 47.▲
    by tehsauce 3y ago

    The first example was a js shader too but because it also included glsl it could have been easy to miss that. So the second example just emphasizes that better. Regarding why, see my first answer or feel free to browse the thousands of othe
18. 48.▲
    by tehsauce 3y ago

    It’s not obvious in that previous example, but you can also write shaders in js, in addition to the built in abstractions for SDFs. https://shaderpark.com/sculpture/-Nc57ho13qXl3G7fQ2wy There are some limitations but a
19. 49.▲
    by tehsauce 3y ago

    GLSL is great! But it is very limited in many aspects. Js enables us to create a more ergonomic APIs for SDFs similar to processing/p5. You can easily mix the glsl and js code, and it also allows you to directly implement transformatio
20. 50.▲
    by tehsauce 3y ago

    For anyone interested in playing around with SDFs, my friend and I made this open source library to make working with them more efficient, and you can write shaders using javascript https://github.com/shader-park/shader
21. 51.▲
    by tehsauce 3y ago

    “Robots will use all the CPUs in your computer to solve levels in the background. Not only does it cut down the time to re-test all levels, it makes your room nice and warm.”
22. 52.▲
    by tehsauce 3y ago

    Does this count? https://github.com/sympy/sympy/blob/master/sympy/integrals/r...
23. 53.▲
    by tehsauce 3y ago

    “Welcome to Jupyter AI, which brings generative AI to Jupyter” Kinda strange statement. I think of Jupyter as one of the places generative AI originated!
24. 54.▲
    by tehsauce 3y ago

    I was thinking it could be a keyboard designed to make sounds special sounds so it can be interpreted very accurately
25. 55.▲
    by tehsauce 3y ago

    Would love a wireless keyboard that works using this! It wouldn’t need any battery, charging or syncing!
26. 56.▲
    by tehsauce 3y ago

    If AI becomes cheaper than human labor, humans laborers across the globe will lose all of their power and leverage. A “pause” might not be a realistic solution, but this kind of parody isn’t any more helpful than climate change denial.
27. 57.▲
    by tehsauce 3y ago

    I think you’re right. Would be very hard to exploit I imagine though.
28. 58.▲
    by tehsauce 3y ago

    8 years of real experience equated to 2 years of university? That is such a shame.
29. 59.▲
    by tehsauce 3y ago

    maybe this function? https://github.com/microsoft/TypeChat/blob/4d34a5005c67bc494...
30. 60.▲
    by tehsauce 3y ago

    > If Bell Labs’ scientists invented the transistor, the laser, Unix, information theory, etc, then YC’s “economic research scientists” invented Stripe, AirBnB, Instacart, Doordash, Dropbox, Reddit… This comparison made me chuckle. It se