4 ms·
Historical analogies for large language models
- telotortium 4y agoA post that shows the limits of reasoning by analogy, but thought-provoking all the same.
- nonameiguess 4y agoI could quibble with some of those, but it's a good exercise. Something not listed that I've been thinking of since this debuted is the impact of CGI on moviemaking VFX. People still build and paint miniatures, still create physical costumes, still film on location and build sets, but it is less than before as a proportion of all VFX (given 100% used to be physical). I'm not sure the work itself actually happens less, though, so much as more movies totally filled with VFX are being made now. The old work continues, but a whole lot of new work was created that is being done via CGI. And, of course, the CGI is not fully autonomous. People are still employed to run the software that paints those scenes. Labor impact seems like even more overall jobs, but they got shittier. Quality about the same. Far more VFX, but they rarely hold up to good practical effects. Jurassic Park, Alien, even Space Odyssey still look better than a Marvel movie, and a huge part of the positive response to Top Gun: Maverick was seeing how awesome practical effects can still be.
- DarmokJalad1701 4y agoShould have asked ChatGPT to make these analogies ...
- etiam 4y agoThat would be very meta, but already quite tiresome.
- sinuhe69 4y ago"We used to harvest huge amounts of natural ice and ship them long distances. The first machines to make ice were dangerous and expensive and made lousy ice. Then the machines became good and nobody harvests natural ice anymore. In this analogy, LLMs are bad at first and don’t have much impact. Then they improve to match and then exceed human performance and human writing mostly disappears." I found the analogy so bad that I stopped reading the rest. Ice is a fixed thing. There are not many forms of ice, so once a machine can produce it, there is no need for natural ice anymore. Unlike writing. There is no "natural" form of writing. Each and every writing is mostly a craft. There is no limit to what and how you can write. There are also myriad forms of writing, for a number of purposes. Writing for entertainment for example is a genuine human product. It's so individual and complex as each human itself. A machine/a statistical program can never produce consistently new writings which entertain and satisfy humans enough to replace human writers. Unless the machine can become human themselves!