tripplyons

7 ms·

1. 61.▲
   by tripplyons 1y ago

   Here's a blog post I liked that explains a connection: https://sander.ai/2024/09/02/spectral-autoregression.html They call diffusion a form of "spectral autoregression", because it tends to fir
2. 62.▲
   by tripplyons 1y ago

   ChatGPT often says yes to both a question and its inverse. People like to hear yes more than no.
3. 63.▲
   by tripplyons 1y ago

   I'm assuming they mean gpt-5-mini. I'm honestly surprised how many people I've heard say "5o".
4. 64.▲
   by tripplyons 1y ago

   Update: Someone in the comments pointed out the missing higher frequencies are just due to YouTube's compression, so I have unlisted the video. I got ahead of myself and made an assumption about it being AI.
5. 65.▲
   One Direction Fans Believe AI-Extended Song Was Leaked Demo(youtube.com)

   3 points by tripplyons 1y ago|1 comments
6. 66.▲
   by tripplyons 1y ago

   Previously, these kinds of laws violated the 1st Amendment, but the changes in the composition of Supreme Court justices have led to different rulings.
7. 67.▲
   by tripplyons 1y ago

   For the US, I'm referring to federal laws or more laws in more states. As for other countries, the EU just delayed the vote for "Chat Control" as recently as yesterday!
8. 68.▲
   by tripplyons 1y ago

   I agree, but I sadly believe these requirements will spread to other countries, including the US. The US Supreme Court recently ruled that Texas' ID law is somehow constitutional.
9. 69.▲
   by tripplyons 1y ago

   The DeepSeek in "American DeepSeek" refers to American open weight models in this article, which feels strange to me when they mention Meta and OpenAI releasing open weight models recently. I guess for it to be an "American D
10. 70.▲
    by tripplyons 1y ago

    Most of the profits from AI are priced in the future. Most people who have made money so far either profited from changes in equity prices or been paid to work at companies that have yet to make a profit from AI. Since the future growth rat
11. 71.▲
    by tripplyons 1y ago

    I would recommend the Code Supernova model in Cursor if you want a 1M token context window. It's free right now since the model is being tested in stealth, but your data will be used by XAI or whoever it turns out the model creator is.
12. 72.▲
    by tripplyons 1y ago

    For Minecraft did they just write text in a book item?
13. 73.▲
    by tripplyons 1y ago

    For the same reason, I run OpenCode under Mac's sandbox-exec command with some rules to prevent writes to the .git folder or outside of the project (but allowing writes to the .cache and opencode directories). sandbox-exec -p "(ve
14. 74.▲
    by tripplyons 1y ago

    Great work! I wonder if there is a way to combine similar cache items instead of dropping unlikely ones. Could the proposed attention estimation be used for that?
15. 75.▲
    by tripplyons 1y ago

    Do you have any experimental results? Why is the writing paywalled on Substack?
16. 76.▲
    by tripplyons 1y ago

    I don't think they will fall for the quantum hype. Jensen has publicly stated that quantum computing is at least decades away from being useful.
17. 77.▲
    by tripplyons 1y ago

    I have tried similar workflows (Neovim + Opencode/Codex CLI), and for me, the biggest downside compared to Cursor is the lack of a tab completion model as good as Cursor's. Supermaven is the best one I've found so far for Neo
18. 78.▲
    by tripplyons 1y ago

    The main reason I use Cursor is for their tab complete model. I enjoy using Neovim more than Cursor, but I choose not to use it because I haven't been able to find anything that makes me as productive as Cursor's tab model.
19. 79.▲
    by tripplyons 1y ago

    Yes, I am a huge fan of the new Grok models!
20. 80.▲
    The DeepSeek v3.2 Breakthrough Simplified(tripplyons.com)

    5 points by tripplyons 1y ago|0 comments
21. 81.▲
    by tripplyons 1y ago

    Agreed. For harder tasks, I like to go to GPT 5 thinking mode, but I'm considering other options. Some times I've had faster success with some of the larger Qwen3 models (480B and 235B variants). I like them in combination with th
22. 82.▲
    by tripplyons 1y ago

    Sorting by revenue would definitely move Grok down. The top 2 Grok models listed are very cheap or even free right now.
23. 83.▲
    by tripplyons 1y ago

    The Grok 4 Fast and Grok Code Fast models have really impressed me. The only issue I've had is reaching a rate limit on the Grok 4 Fast model. Amazing pricing for highly capable models with good tool calling support.
24. 84.▲
    by tripplyons 1y ago

    The gated variants of these functions have been dominant for a few years.
25. 85.▲
    by tripplyons 1y ago

    Most LLMs use GeGLU or SwiGLU.
26. 86.▲
    by tripplyons 1y ago

    What makes you believe the exponential phase will end soon?
27. 87.▲
    by tripplyons 1y ago

    As much as I want to believe something like that, its hard for me to come to that conclusion. I lived at a boarding school for boys with high functioning autism (with a strict 100+ IQ requirement), and even then, almost no one could underst
28. 88.▲
    by tripplyons 1y ago

    This doesn't mean they run their own index. They are likely just reselling access to whatever index they are using for their product.
29. 89.▲
    by tripplyons 1y ago

    I'm sure there are some people who would prefer a graphical interface like this.
30. 90.▲
    by tripplyons 1y ago

    The streets themselves are maintained at the expense of taxpayers, regardless of who is begging on them.