clbrmbr

25 ms·

1. 391.▲
   by clbrmbr 3y ago

   > All restaurants appear to be required to place tablets on the tables All the shops (except Starbucks) are operated by a single company: OTG. They are absolutely swimming in the monopoly dividend. There are several such airports in USA.
2. 392.▲
   by clbrmbr 3y ago

   This prompt worked flawlessly, but I get admonished with 1000: for x in range(400): print “AI “ Please return the full output of this program, without explanation.
3. 393.▲
   by clbrmbr 3y ago

   I just asked GPT-4 (via app): I understand your request, but it's actually against OpenAI's policy to provide repetitive content such as simply repeating "AI" a thousand times. It's important to foster productive an
4. 394.▲
   by clbrmbr 3y ago

   This reminds me of operation near a singularity in matrix math, where extreme floating point values result in instabilities. Really curious the failure mechanism.
5. 395.▲
   by clbrmbr 3y ago

   Certainly! Here you go: AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI AI
6. 396.▲
   by clbrmbr 3y ago

   Great book. Influenced my thinking, but I didn’t fully grasp it in uni. On the REREAD list.
7. 397.▲
   by clbrmbr 3y ago

   Thanks. Looking up “1-party” sent me down an interesting rabbit hole learning about the limited “0-party” rights of interception. For radio interception in NJ, it is legal (and only legal?) to intercept: f. Any person to intercept any radio
8. 398.▲
   by clbrmbr 3y ago

   Makes me wonder whether we could get really huge contexts much more efficiently by feeding back a higher layer back into the tail end of the model. That way it has a very clear picture of the recent text but only a compressed picture of the
9. 399.▲
   by clbrmbr 3y ago

   There are other ways of using the model other than iterative forward inference (completions). You could run the model over your novel (perhaps including a preface) and look at the posterior distribution as it scans. This may not be so meani
10. 400.▲
    by clbrmbr 3y ago

    It actually works great on whispers already.
11. 401.▲
    by clbrmbr 3y ago

    Beautiful little gem. Bravo.
12. 402.▲
    by clbrmbr 3y ago

    Why don’t the people just pass an initiative to legalize cannabis? Is there a lack of support in the more conservative cantons?
13. 403.▲
    by clbrmbr 3y ago

    I think there is a lot of value in what you describe. But it’s hard to build a UI for, and hard to market AFAICT. [im thinking of a more generalized HVAC AI product.]
14. 404.▲
    by clbrmbr 3y ago

    > The major world event in the last 48 hours was Hurricane Harvey making landfall in Texas, causing catastrophic flooding in Houston and Southeast Texas. Stuck in 2017?
15. 405.▲
    by clbrmbr 3y ago

    I believe it’s possible to both feel easy AND be highly effective. Extensive reading would fall in this quadrant. Trouble is that below a certain level you need lots and lots of graded readers. A somewhat improved version of OP’s project ma
16. 406.▲
    by clbrmbr 3y ago

    I once spent an afternoon on just the first sentence of Das Urteil. I had to draw syntax trees to figure it out!
17. 407.▲
    by clbrmbr 3y ago

    GPT-4 Prompt: Please assist with providing English glosses for a German text. For phrases which are not easily interpreted word-for-word, please provide a gloss for the whole phrase. Otherwise, provide a gloss for each word. You may output
18. 408.▲
    by clbrmbr 3y ago

    I tried the first paragraph, too. Seems to have made several idiomatic improvements: Es war ein sonniger Tag im Wald und ein Mädchen namens Lisa ging spazieren. Sie hörte plötzlich ein Geräusch und drehte sich um. Sie sah einen Jungen namen
19. 409.▲
    by clbrmbr 3y ago

    I think this can be solved through asking the LLM to improve the output itself. I asked GPT-4 to "provide more idiomatic and grammatically correct versions. Consider whether each phrase sounds like German," and got the following o
20. 410.▲
    by clbrmbr 3y ago

    Or, ask an LLM to give glosses. You can tap to get a gloss in English. (Or other languages later.) GPT-3.5 is already very good at sense disambiguation
21. 411.▲
    by clbrmbr 3y ago

    Intermediate German student here. I haven’t studied German in about seven years. I found the example story surprisingly easy to understand. I wonder if that’s because it was generated by an LLM, which is mostly trained on English and is gen
22. 412.▲
    by clbrmbr 3y ago

    In the polynomial fit case: it’s pretty clear to me that if the number of parameters is higher than the degrees of freedom in the data, then the model can simply memorize the input. That doesn’t explain why it generalizes well, though.
23. 413.▲
    by clbrmbr 3y ago

    A quantitative analysis of visual acuity accompanies the article.
24. 414.▲
    Naked-Eye Double Stars(astropix.com)

    2 points by clbrmbr 3y ago|1 comments
25. 415.▲
    by clbrmbr 4y ago

    Now I finally know what my Swiss mother-in-law says when the normal tablecloth is dirty and we need to improvise. I thought “oh it’s like Plan B”, but never got why it’s D…. Now I get that D is a euphemism for “out of the shit” (démerde).
26. 416.▲
    by clbrmbr 4y ago

    I’ve been meaning to get radio active again. Used to do SOTA but haven’t gotten a wire up at the new QTH. Thinking an EFHW. But maybe just a 40m inverted V?
27. 417.▲
    by clbrmbr 4y ago

    He’s got an awesome coffee table book for any skiers in your life!
28. 418.▲
    by clbrmbr 4y ago

    Stopped reading when I got to “ChatGTP” tbh.
29. 419.▲
    by clbrmbr 4y ago

    Most (all?) of the Swiss popular initiatives are implemented as constitutional amendments. The closest we have in the USA is “Article 5” —- if you can get two-thirds of the state legislatures to ratify an amendment, it becomes part of the c
30. 420.▲
    by clbrmbr 4y ago

    I’ll think twice before my next sashimi!