[–] 5 points 5 months ago (3 children)

Now we're getting into linguistics with the question of "what is a wave?"

In quantum physics, basically everything is waves, in the sense that the same mathematical formulae used to describe waves are used to describe quantum phenomena. The intuitive human-scale dynamics of waves don't necessarily apply though.

For example, sound waves can't propagate through a vacuum, but light waves can. Aside from that, they follow mostly the same rules. You can use the same math the describe interference of sound waves and light waves, for example.

People talk about the "particle/wave duality" of photons because in some ways they behave like waves and in some ways they behave like particles. But both of those words are stretched a little from their everyday plain-english usage, and the precise reality would require years of study to understand.

Plain English wasn't made to be that precise or objective. That's why we use math. :)

I'm no expert in quantum physics so take this all with a grain of salt.

  • source
  • parent
  • context
  • It's still an open question where the eventual sweet spot will be in terms of model size and speed once the dust settles.

    Nobody has the hardware to run frontier models in their personal devices. Even the larger open models are out of reach unless you're ready to spend $10-20k on hardware. You can't do shit on 8GB of memory.

    That said, I don't think there's any great use case for trillion-parameter models in the long term. You can get good results for cheap from much smaller models with smarter workflows, and eventually that will become as easy and accessible as using cloud products. The big players have done well staying 6-12 months ahead, but that's really not a lot in the grand scheme and they can't keep it up indefinitely.

    Their only play is regulatory capture and they're pushing hard for it.

  • source
  • parent
  • context
  • [–] 18 points 5 months ago

    As others have said: they're nasty. There are a myriad of health risks associated with roaches in your home. They definitely do us harm, albeit not as directly or obviously as, say, ticks or mosquitoes.

    I consider my fear of roaches to be entirely rational. They directly carry diseases, and they produce allergens. They exacerbate or even cause Asthma. Also, it's never good to have decaying corpses of any type inside your walls, which is where they will end up if you have an infestation.

    They will eat anything. They will go anywhere. They will thrive anywhere. I've seen wallpaper come right off the wall because roaches ate all the goddamn glue. You cannot get rid of them without a whole lot of poison (if even), which is itself a health risk. If I see a roach, I go to DEFCON 2. (DEFCON 1 is reserved for bedbugs.)

  • source
  • [–] 48 points 5 months ago (11 children)

    The coolest, and often most confusing thing about computer science, information theory, and perhaps reality in general, is how everything becomes more or less equivalent if you boil it down and twist it around a little.

    Everything is sorting. Everything is compression. Everything is geometry. Everything is language. Everything is music. Everything is, like, waves, man. *puff*

    Or more accurately, everything can be expressed in any of those other things' terms.

    These are not new ideas, but computers have made them provably and demonstrably true in many contexts, and I think that's super cool.

  • source
  • [–] 2 points 5 months ago

    I think this is the real answer, rather than the other comment about users being monetized. There are plenty of affordable Android phones with decent specs that have no spyware besides Google's, and Google's not subsidizing them.

    There are also still dumbphones on the market for under $100. I don't see Android app support as that big a selling point with these specs.

    Good luck to them, but I don't think this is a viable niche.

  • source
  • parent
  • context
  • [–] 28 points 5 months ago (2 children)

    This is cool, but the framing rubs me the wrong way.

    The assumption of human-like attributes is a bit of a stretch when you can build an LLM in any substrate

    "I can theoretically run arbitrary computer programs inside another computer program" does not hold any philosophical weight.

    We've known what Turing-completeness means for many decades now. ANY program you run on your computer can theoretically run on ANY Turing-complete substrate, which includes all kinds of ridiculous things. You could do this with PowerPoint or Minecraft, too. Or with a sufficient number of rocks on a beach.

    If this is your first time learning about Turing-completeness, then...well, that's cool! Just don't let it rock your world too hard. :)

    Whether LLMs have any "human-like attributes" has nothing to do with the substrate, beyond the point that different substrates will lead to different levels of performance in practice, and some emergent properties might only manifest at higher levels of performance in practice.

    Whether a Turing machine (i.e. conventional computer) can theoretically do the same things as a human body is still an open question. (I say "human body" instead of just "human brain" because the brain is not responsible for all of human cognition and behavior, despite popular belief.) It might be fully impossible (unlikely, IMHO), it might require a completely unrealistic scale (most likely, IMHO), or it might be downright trivial once we figure out how.

    😯 1
  • source
  • parent
  • context
  • [–] 10 points 5 months ago (1 child)

    The WWDC presentation yesterday was hilarious. Almost everything they said about the UI could be boiled down to: "We're undoing some of the incredibly bad decisions we made last year. Not all of them, but some of the big ones!"

    They then went on to demo the new improved Siri, and as someone who doesn't use Siri, all I could think was "wait...Siri couldn't do this 10 years ago?!"

    What a sad state of affairs.

  • source
  • parent
  • context
  • view more: ‹ prev next ›