Skip Navigation

Posts
5
Comments
373
Joined
3 yr. ago

Professional software engineer, musician, gamer, stoic, democratic socialist

  • Deleted

    Permanently Deleted

    Jump
  • They are a useful tool when you understand their shortcomings. They are very inconsistent, so you need to put a lot of guardrails around them.

    I don't really understand how people manage to be productive with swarms of agents. They really need to be babysat IME. I'm constantly waffling between arriving at correct solutions quickly and getting stuck in a tar pit of hallucinated problems and fake analysis.

    That said, I'll be upset when the AI companies inevitably start raising prices or nerfing models.

  • Nobody with normal vision both looking at the same original picture claims the blue part is white.

    Doesn't matter what context I view the original image. I've never seen it as blue and black without manipulating the image.

  • Wow. You are just proving my point. It looks white and gold to me.

  • It is almost always that.

  • I'm talking specifically about the variant of RL used by systems like AlphaGo.

  • Wdym? It's existed for at least a decade. Plenty of papers about it. It mastered Atari and Mario. It became the best Go player.

  • LLMs might suck at this game but I'm pretty sure Deepmind's deep reinforcement learning AI could solve these easily.

    EDIT: I know you guys hate AI around here, but you need to at least be aware of what the technology is capable of.

    From 11 years ago:

    https://youtu.be/V1eYniJ0Rnk

  • and it happens exactly as the people whose claim is being denied with “slippery slope” fallacy said

    But this is the crux of the fallacy. What evidence is anyone providing that there is indeed an insidious chain of events we are enabling by adding the birthdate field? Are there examples of cases similar to this in history?

    EDIT: I can tell people are getting emotional about this because I'm being down voted for just asking a question that elaborates the point someone is making.

  • I'm thinking seriously about using something like a Daylight tablet as a thin client for a more powerful machine at home. Obviously doing real coding by hand would still suck, but LLM-based coding might actually be viable.

  • Forgot the punchline

  • I don't understand how this serves the purpose of ad blocking.

  • Mother!

    I found it really enjoyable trying to trace all of the symbolism and allegory. For whatever reason, audiences didn't like it so much.

  • It's quite highly rated.

  • This is basically the definition of throwing the baby out with the bathwater.

  • This just makes me realize that my TI Nspire calculator probably violates this law.

  • Why deleted? This was a good rebuttal.

    EDIT: I don't think the comment really violated rule 1, but there was apparently a followup comment that definitely did, and this one just got removed by association. Here's a very slightly paraphrased version of it that should not break the rules:

    Gish gallop of [explitive].

    A) overblown, and that argues for cleaner power, better cooling, and more efficient models

    B) regulation failure

    C) incorrect, they have made discoveries that humans have been unable to. All human knowledge is built off previous knowledge.

    D) the enemy is both weak and strong. If they don’t produce anything good then the people who are losing their jobs can’t have either, right?

    E) small study based on one task which people are misrepresenting. The actual evidence shows it makes people smarter as they shift priorities.

    F) only for vulnerable people. Better safeguards are needed for the weak minded.

    G) argument against using people’s likeness not ai

    H) use an open source Chinese model

    I) market distortion problem, not a principled reason no one should use the technology any more than GPU shortages made all graphics work illegitimate.

    J) see (H)

    K) try one argument next time. Your best one, [some snarky sarcasm]

  • Slop points aside, I found 5.4 to be pretty ass compared to 5.3 codex. Took way longer and wasted more tokens.

  • 1000 lines isn't that unreasonable for a PR. Commit size matters more.