zaptrem

27 ms·

1. 421.▲
   by zaptrem 4y ago

   I think a personal subscription model where ChatGPT Plus subscribers are given personal API access for use with third-party open source apps (e.g., Chrome extensions, Siri Shortcuts, VSCode extensions) would be extremely valuable and much m
2. 422.▲
   by zaptrem 4y ago

   It was less that ChatGPT is better than LaMDA and more that the tech everyone was keeping away from society "for its own good" had been released and people were realizing it could finally threaten Search.
3. 423.▲
   by zaptrem 4y ago

   This is only a 1.5b parameter model. This is in line with that. GPT3.5 is ~175b params.
4. 424.▲
   by zaptrem 4y ago

   You might also mess with your loss function to force it to "make up its mind" as right now the blurry mess likely minimizes the error from the actual frame (which isn't really want you want).
5. 425.▲
   by zaptrem 4y ago

   Wouldn’t all the Claudes be incentivized to simply trash each other constantly in that case?
6. 426.▲
   by zaptrem 4y ago

   Until you accidentally overshot by a single pixel and your entire foldout of menus disappeared, forcing you to spend time going through the whole tree again (unless my memory is failing me).
7. 427.▲
   by zaptrem 4y ago

   Charging a battery of this scale at 1MW likely won't have any more significant effect on life span than charging a Model whatever at a normal supercharger (which is negligible assuming thanks to battery temperature management and charg
8. 428.▲
   by zaptrem 4y ago

   I did a simpler version of that by asking ChatGPT to build a LaTeX rendering bookmarklet for itself here: https://twitter.com/zaptrem/status/1598553052075761669
9. 429.▲
   ChatGPT Wrote Itself a New Feature (With a Little Help)(twitter.com)

   2 points by zaptrem 4y ago|1 comments
10. 430.▲
    by zaptrem 4y ago

    All I did was ask for the feature and feed the code back to it with an error whenever something went wrong. The only difficult part was asking it to check for page changes correctly (took a tiny one-line fix on my part, but it did 99%). Cod
11. 431.▲
    by zaptrem 4y ago

    Have you tried this out yet? https://chat.openai.com/ It's been providing real value to me over the past day for practicing Spanish, explaining Machine Learning concepts, and doing fancy write-ups in LaTeX. And this on
12. 432.▲
    by zaptrem 4y ago

    I'm able to run a 22b parameter GPT-Neo model on my 24gb 3090 and can fit a 30b parameter OPT model when combining my 3090 and 12gb 3080
13. 433.▲
    by zaptrem 4y ago

    But this has nothing to do with China
14. 434.▲
    by zaptrem 4y ago

    What did Apple even hope to accomplish here? If this was some (extremely) ill-thought-out attempt at combating hate or misinformation don't they realize 90% of said content will be fed to users via YouTube/Reddit/Safari/
15. 435.▲
    by zaptrem 4y ago

    Can confirm.
16. 436.▲
    by zaptrem 4y ago

    The video targeting algorithm is a view into individual's psyches on a level that was previously impossible at this scale.
17. 437.▲
    by zaptrem 4y ago

    I'm sure nobody will try to drive to an Airport to which they are unable to buy tickets.
18. 438.▲
    by zaptrem 4y ago

    Do you have a link to an archive of the OG piece?
19. 439.▲
    by zaptrem 4y ago

    Awesome. This should be a world-wide standard. NACS (Tesla) VS CCS is like USB-A 1.0 vs USB-C with Power Delivery: CCS is unbelievably slow and clunky to the point where it might turn people off wanting an EV once they experience the slow a
20. 440.▲
    by zaptrem 4y ago

    I thought this was going to be a blogpost for native support or a fancy workaround. Instead it's similar to Algolia's solution which is just sending all of your data to their servers as well as Google's.
21. 441.▲
    by zaptrem 4y ago

    This graph shows transistors basically maintaining pace and completely disregards multi-core performance. Of course single core perf will rise more slowly when a chip now has 8-64x as many cores.
22. 442.▲
    by zaptrem 4y ago

    Twitter had massive layoffs planned anyway that they put on hold when Musk started the buyout.
23. 443.▲
    by zaptrem 4y ago

    Coincidentally I encountered this system for the first time last night. Unfortunately it never finished loading and I embarrassingly had to switch to a physical card after stalling for 30 seconds.
24. 444.▲
    by zaptrem 4y ago

    This is not at all unusual behavior for 4Chan users.
25. 445.▲
    by zaptrem 4y ago

    "StackExchange: Could the end cap of the Pascal B (1) survive its trip through the atmosphere?" https://physics.stackexchange.com/questions/488151/could-the...
26. 446.▲
    by zaptrem 4y ago

    They were, however, dying from organ failure and blood loss.
27. 447.▲
    by zaptrem 4y ago

    What happens when a small object rolls into that blindspot? What happens when the car goes into deep sleep and the AP computer state is lost? Reloading all those nets takes like 40 seconds on startup from deep sleep.
28. 448.▲
    by zaptrem 4y ago

    No, and they should be criticized for that. However, this is the transparency that makes up for it: https://www.youtube.com/watch?v=ODSJsviD_SU
29. 449.▲
    by zaptrem 4y ago

    It would be easier for devs to modify MacOS to rip out the hypothetical ads and complain loudly until Apple fixes it than to switch OSes completely.
30. 450.▲
    by zaptrem 4y ago

    https://www.wired.com/story/electric-vehicles-could-rescue-t...