Bruce Sterling @bruces*Slop weather.
*What the heck, if it's .99 cents, everybody will buy it
Brewster Kahlewhy "slop"? is this because it is using AI tools? and the AI tools are not procedural therefore not decomposable and debuggable in the familiar ways?
OCR recently went from hand coded feature extraction and rules based identification to the machine-learning system in tesseract. Is this also "slop" ?
trying to understand why "slop". Seriously, I'm curious.
Charlie Stross@brewsterkahle @bruces Let's distinguish between two contemporary uses of the word "AI". There is deep learning, analytical AI: stuff like image recognition, OCR, machine translation, etc. Almost all of this is good. And there is generative AI, LLMs, which are mostly used to spew bullshit and lies and plagiarize art and media. Almost invariably sub-par derivative slop.
Unfortunately it's the generative slop that has caught the imagination of media and investors (who throw money at it).
Clear?
Hugo Mills@cstross @brewsterkahle @bruces And yet everyone gets all surprised and defensive when you call them out on "AI", and they say "oh, we didn't mean the LLMs".
If you or your marketing department hasn't worked out that you're going to get that reaction from saying "AI", then you're reading a large chunk of the room very very wrong.
And if you're ignoring that part of the room because "AI" is popular, then that's just flat out culpable idiocy, like everyone else I've seen involved in marketing.
CMDR Yojimbosan 🅅⁂@brewsterkahle @cstross @bruces There seems to be a big difference between generating weather forecasts by modelling the atmosphere, vs generating a weather forecast by assuming tomorrow will be similar to today, that "past performance is indicative of future results", contrary to the SEC's rule 156 in the finance world and a world with large-scale climate changes.
The paper that was linked from the article was just a summary itself, and very very light on details, so more of a tech-hype than actual science.
However, for short-range forecasts, and ones for specific microclimate areas, it may well be a great approach. The global climate models can rarely get the sort of resolution needed for small areas. You want them for longer-term forecasts though!
mav 
@brewsterkahle
@cstross @bruces
There's also the fact that the current state of things is such that, despite this being a scientific paper published in Nature, an AI "breakthrough" is honestly hard to believe exists at this point. We have been so radically oversold on supposedly new and radical AI ideas that turn out to be significantly inferior to the systems we have now that without long-term testing and verification with somehow trusted third parties (do those even exist anymore?) that it would be hard to believe the kind of revolutionary predictive power described here is real.