Charlie StrossMar 19

RE: https://mastodon.scot/@Wen/116255252453377471

Also note the very foreign road layout, mail carrier, foreign cop, non-Scottish suburb and buildings, and all the rest of the imagery. That illustration exemplified American cultural imperialism from top to bottom. (And looking at the shadows, the sun NEVER gets that high overhead—when you can see it at all, that is!)

Show threadVery Human RobotMar 19

@cstross
Why do you hate our FREEDOM?
(Red tailed hawk sounds)

Seriously, though: Good tools let your control them, rather than hallucinate from scratch. Upload a couple of street pictures from Edinburgh and use that as reference. We have a massive tool-teaching challenge ahead of us, just like with the Internet.

Show threadCharlie StrossMar 19
@StompyRobot And what about the linguistic usage? (American training texts in LLMs outnumber all other English language dialects massively, let alone Scots, which is *not* the same as English-English by a long way, even before we get into stuff like Doric and Highland Gaelic …)
Show threadVery Human RobotMar 19

@cstross
Again, the big models will do better when provided the right reference.

FWIW the models actually have a significant Nigerian English bias, because a lot of data was created there for cost reasons.

Also, the user has to guide the model on the actual plot and character, or you'll get a very "dice table" vibe in the output.

I don't agree with humans who just let the slop machine do what it wants. I also don't agree with humans who refuse to use the tool at all. We're at the beginning.

Show threadRaven OnthillMar 19
@StompyRobot @cstross If it's this much work to edit the output, wouldn't it be better just to have written and drawn the thing in the first place?
Show threadVery Human RobotMar 21
@ravenonthill @cstross
Use whatever tools you feel works best for you!
Just realize others have different preferences.
Output that people like is the end goal!
Show threadRaven Onthill4d ago
@StompyRobot @cstross Mmmm, only partly. Sometimes the point is "proof of thought," the output is the proof that someone has looked at and thought about the problem. If people rely on "AI," that's lost.
Show threadVery Human Robot4d ago

@ravenonthill @cstross

Look at a "Michelangelo" artwork and it might actually have been made by an apprentice, or a series of apprentices, under the direction of the master.

The generative tools generate mid level slop when you ask for "a beautiful sunset" but in the hands of a skilled, trained artist we can get quite good results.

You can "prove thought" with a pencil, a camera, a typewriter, newspaper clippings, a computer, or a chisel -- the medium itself must be harnessed.

Show threadRaven Onthill

@StompyRobot @cstross think of how many middle and line managers are just generating reports. Think of how many students are just submitting work to get a grade.

Maybe these tools improve productivity; some of the reporting I'm seeing suggests these tools provide an illusion of increased productivity, rather than the reality. But far too many people are using them to cheat.

Apr 3 at 12:44pmTusky
Show threadVery Human Robot3d ago

@ravenonthill @cstross

You say "cheat," I say "expose sham systems for the illusions they are."

Higher education is actually hard. In no way is 90% of the population going to do well in that environment.

We've created, as a society, checkpoints that are supposed to correlate with capability, but don't actually demonstrate capability.

Those checkpoints were always open to cheating, as you can tell if you've ever worked with people who talk too much and accomplish too little.

Show threadRaven Onthill3d ago
@StompyRobot @cstross "burn it all down, something better will show up to replace it?" Really?
Show threadVery Human Robot1d ago
@ravenonthill @cstross
That's not what I'm saying.
I'm saying the outcomes are a direct reflection of our actual (as opposed to stated) societal values.
Show threadVery Human Robot1d ago
@ravenonthill @cstross
90% of societal work has always been bullshit jobs with no real impact or accountability.
Professors don't have time to evaluate students because that needs 1:5 ratio not 1:500.
Managers don't actually evaluate output of reports because that requires discomfort and effort.
We keep this illusion up because we really don't want 90% of people to die destitute in a ditch, we want social cohesion.