• 0 Posts
  • 11 Comments
Joined 2 years ago
cake
Cake day: August 27th, 2023

help-circle
  • We don’t need leaps and bounds, from here. We’re already in science fiction territory. Incremental improvement has has silenced a wide variety of naysaying.

    And this is with LLMs - which are stupid. We didn’t design them with logic units or factoid databases. Anything they get right is an emergent property from guessing plausible words, and they get a shocking amount of things right. Smaller models and faster training will encourage experimentation for better fundamental goals. Like a model that can only say yes, no, or mu. A decade ago that would have been an impossible sell - but now we know data alone can produce a network that’ll fake its way through explaining why the answer is yes or no. If we’re only interested in the accuracy of that answer, then we’re wasting effort on the quality of the faking.

    Even with this level of intelligence, where people still bicker about whether it is any level of intelligence, dumb tricks keep working. Like telling the model to think out loud. Or having it check its work. These are solutions an author would propose as comedy. And yet: it helps. It narrows the gap between “but right now it sucks it [blank]” and having to find a new [blank]. If that never lets it do math properly, well, buy a calculator.





  • Self-learned programming, started building stuff on my own, and then went through an actual computer science program.

    Same. Starting with QBASIC, no less, which is an excellent source of terrible practices. At one point I created a code snippet that would perform a division and multiplication to find the remainder, because I’d never heard of modulo. Or functions.

    Right now, this lets people skip the hair-pulling syntax errors, and tell the computer what they think the program should be doing, in plain English. It’s not even “compileable pseudocode.” It’s high-level logic, nearly to the point that logic errors are all that can remain. It desperately needs some non-answer feedback states for if you tell it to “implement MP4 encoding” and expect that to Just Work.

    But it’s teaching people to write the comments first.

    we’re nowhere close to that right now.

    The distance from here to “oh shit” is shorter than we’d prefer. This tech works like a joke. “Chain of thought” apparently means telling the robot to act smarter… and it does. Which is almost less silly than Stable Diffusion removing every part of the marble that doesn’t look like Hatsune Miku. If it’s stupid, but it works… it’s still stupid. But it works.

    Someone’s gonna prompt “Write like Donald Knuth” and the robot’s gonna go, “Oh, you wanted good code? Why didn’t you say so.”


  • An otherwise meh article concluded with “It is in everyone’s interest to gradually adjust to the notion that technology can now perform tasks once thought to require years of specialized education and experience.”

    Much as we want to point and laugh - this is not some loon’s fantasy. This is happening. Some dingus told spicy autocomplete ‘make me a database!’ and it did. It’s surely as exploit-hardened as a wet paper towel, but it functions. Largely as a demonstration of Kernighan’s law.

    This tech is borderline miraculous, even if it’s primarily celebrated by the dumbest motherfuckers alive. The generation and the debugging will inevitably improve to where the machine is only as bad at this as we are. We will be left with the hard problem of deciding what the software is supposed to do.