AIs gunning for our precious freelancers

[Raphael]

For his achievement, the student, 21-year-old Luke Farritor, has won $40,000. But he should probably share the prize with AI as it helped him to identify a single word on the scroll:

Should people who do scholarship by analysing photographs share any prizes they win for it with the concept of photography, too?

[bradrn]

AI is now taking linguists' jobs, too!

A computer science student has managed to decipher the first word on an ancient Roman scroll carbonized by a Mount Vesuvius eruption – with the help of artificial intelligence.

For his achievement, the student, 21-year-old Luke Farritor, has won $40,000. But he should probably share the prize with AI as it helped him to identify a single word on the scroll: “porphyras,” an ancient Greek word for “purple.”

Your news is sadly out-of-date: they’ve now read the whole scroll, with more to come.

[Zju]

AI is now taking linguists' jobs, too!

A computer science student has managed to decipher the first word on an ancient Roman scroll carbonized by a Mount Vesuvius eruption – with the help of artificial intelligence.

For his achievement, the student, 21-year-old Luke Farritor, has won $40,000. But he should probably share the prize with AI as it helped him to identify a single word on the scroll: “porphyras,” an ancient Greek word for “purple.”

Your news is sadly out-of-date: they’ve now read the whole scroll, with more to come.

Oh my - I'm stoked as to what the rest of the scrolls may hold. IIRC there were one or two storeys of burnt up scrolls? Here's hoping to some bilingual texts in Etrurian or maybe some other un- or poorly documented ancient language.

[Raphael]

It seems to have caused quite a stir among scholars of classical antiquity.

[malloc]

All this has given me a brilliant idea. I will enter the Tour de France on a motorcycle. If the judges refuse me entry, I will denounce them as small-minded technophobes fighting progress or even Heideggerian crypto-fascists.

Should people who do scholarship by analysing photographs share any prizes they win for it with the concept of photography, too?

Cameras are not intelligent though. They merely replicate images verbatim (within the limits of their film quality anyway) without analyzing or interpreting them. Artificial intelligence with the ability to read and decode ancient languages clearly differs fundamentally from photography. Granted the machine involved probably has no use for prize money nor would it care about getting credit. I would hesitate to credit Farritor, though, given his limited contributions.

[zompist]

Your news is sadly out-of-date: they’ve now read the whole scroll, with more to come.

Not quite, they say they can read 5% of the scroll.

But the achievement is amazing. They did CT scans of the carbonized scroll, then went digitally slice by slice, tracing where the papyrus sheets lay; then assembled that into an image of each part of the scroll; then used new techniques to identify where the ink was. AI wasn't used to actually read the text, but to determine the ink patterns.

The next step is to automate the digital unrolling technique more. As they say, "It cost us more than $100 per square centimeter in manual labor to produce the text we can read today. At this price, it would cost hundreds of millions or maybe even billions of dollars to segment all of the scrolls." I know the original reference was jocular, but this is a very nice example of where AI can do something that couldn't be done without it, and it has actually created human jobs. (There was no point in hiring segmenters if the resulting images can't be read.)

[alice]

Artificial intelligence with the ability to read and decode ancient languages clearly differs fundamentally from photography.

I think this is ennobling what the "Artificial intelligence" actually did just a little bit.

[bradrn]

Artificial intelligence with the ability to read and decode ancient languages clearly differs fundamentally from photography.

I think this is ennobling what the "Artificial intelligence" actually did just a little bit.

This is correct. The website makes a point of explaining how little they actually use AI for. As far as I understand, This project uses one model to create a 3D model of the convoluted scroll (‘segmentation’), and then another one to identify whether each individual pixel has an ink speck in it or not. None of the models have any knowledge about ancient languages — that’s all done by humans, who are much better than current LLMs at assessing the likelihood of their predictions.

[malloc]

This is correct. The website makes a point of explaining how little they actually use AI for. As far as I understand, This project uses one model to create a 3D model of the convoluted scroll (‘segmentation’), and then another one to identify whether each individual pixel has an ink speck in it or not. None of the models have any knowledge about ancient languages — that’s all done by humans, who are much better than current LLMs at assessing the likelihood of their predictions.

Fair enough. But why can't we keep it that way? Why must we keep giving away all the intellectually stimulating jobs to machines when people are stuck in mindless drudgery?

[rotting bones]

All this has given me a brilliant idea. I will enter the Tour de France on a motorcycle. If the judges refuse me entry, I will denounce them as small-minded technophobes fighting progress or even Heideggerian crypto-fascists.

Your suggestion is not analogous to banning motorcycles from one race. It is analogous to letting no one ever again conduct research into building any motorcycle EVER. Why? Because one race exists where you can't use motorcycles.

[rotting bones]

This is true -- but what can replace markets*? We have tried central planning, and we know how well that works. What you have proposed in the past is essentially a democratic version of central planning, as opposed to the Stalinist version, but I am not sure I'd appreciate being voted into writing web apps rather than embedded Forths...

* Note that I do not mean capitalism here -- the solution to the problem of exploitation and oppression by the capitalist is worker ownership and self-management of capital, not replacing markets with something else.

Roughly, my proposal is that the government should create an official Patreon-like website. Creators register their projects on this website. Every citizen gets 1 infinitely divisible vote. Citizens send parts of their vote to the projects whose products they want to see more of. Creators draw a government salary in proportion to the number of votes they got.

I can understand the misunderstanding of my proposal as central planning, but ZeptoForth is not even your job. The whole point of my proposal is to give everyone more free time. Increasing the production of desired goods lowers their scarcity.

The only way the capitalist class will not use AI to attempt to replace people is either if it is A) infeasable, as I suspect is the case for non-trivial software engineering or B) unprofitable, as may be the case once all the venture capital runs out and the AI companies start charging real money for their services.

I mean, how will other people reproduce the same drawing I generated with an AI? Art is not fungible like essential goods.

I must admit that I may not be the common case here (I started programming at eight years old) but the key thing is that one must not mistake coding alone for software engineering (as much as I dislike the job title "software engineer" for being vaguely pompous; I personally rather consider myself a programmer). LLM's may help you with the former; they will not help you with the latter. LLM's may make up for a lack of coding ability, but they will not help you with high-level software design and architecture. In theory the difference between coding camps and computer science programs is that the former will teach you how to churn out code while latter will teach you more the theory of algorithms, data structures, logic, software design, and software architecture; of course, in reality these get conflated (but teaching you to write code is what CS 101 (or whatever it's designated at your university) is for; the rest of the classes are for actually teaching you computer science and software engineering).

Software Engineering is a different use case in which Devin is the first system to produce non-worthless results.

The key thing is if you knew how to fix a wrong first approximation you also knew how to write it yourself, even if it may have taken more time to code it, because if that is not the case, how do you know how to fix the wrong first approximation or whether it is right or wrong in the first place?

The purpose of research is to investigate what could be done, not do things we already know how to do. In my experience, having preconceptions that are too strong and definite is antithetical to research. It prevents you from seeing the points where innovation is necessary to bridge the gap from the possible to the actual.

Brainstorming ideas through humans or AI often results in a first approximation. The first approximation is not a solution. It is a neighboring element of the matrix in which the actual solution is embedded. Seeing how changing elements of the first approximation changes the result builds your intuition for creating the actual solution, even if this solution has barely anything in common with the first approximation.

[malloc]

All this has given me a brilliant idea. I will enter the Tour de France on a motorcycle. If the judges refuse me entry, I will denounce them as small-minded technophobes fighting progress or even Heideggerian crypto-fascists.

Your suggestion is not analogous to banning motorcycles from one race. It is analogous to letting no one ever again conduct research into building any motorcycle EVER. Why? Because one race exists where you can't use motorcycles.

Well yeah, it was more of a joke than anything. Yet that is certainly how the situation with generative AI feels from an artistic perspective. Artists must now compete with machines that have such extraordinary advantages of speed that even the fastest artist cannot keep pace.

More than anything, I worry about the future of education and academia. The only reason higher education gets funded at all is training people for mental work. Now that machines can carry out mental work, mass education has suddenly become much less valuable to the ruling class. Soon enough they may well decide to axe it and leave the masses illiterate and uneducated. You can forget about people abandoning tradition and learning true facts through Marxist literature once that happens. Lack of education correlates strongly with traditionalism after all. Apart from everything else I have mentioned, we should oppose AI simply to avoid the mass stupification that automating brain work would entail.

[alice]

You can forget about people abandoning tradition and learning true facts through Marxist literature once that happens.

Be careful not to oversell your point; I doubt there are many "true facts" which are only found in Marxism.

[malloc]

You can forget about people abandoning tradition and learning true facts through Marxist literature once that happens.

Be careful not to oversell your point; I doubt there are many "true facts" which are only found in Marxism.

That was meant in response to Rotting Bones claiming that one must read Marx to become capable of learning truths. I would certainly agree that such a claim seems oversold, but working within his premises seems the best way to make my point to him.

Above all, I must stress that I have no objections to technology in general nor would I consider myself a rural nostalgic by any means*. Movements like anarcho-primitivism and the more general tendency of leftists to romanticize the world before capitalism have always struck me as misguided. Nonetheless one must acknowledge that many technologies carry significant risks and even catastrophic side effects. We must critically evaluate each technology and consider whether the harms outweigh the benefits. Embracing every technology innovation uncritically is no less foolish than scorning technology and retreating to bucolic fantasies. For reasons that elude me, Rotting Bones seems to conflate any critique of technology, even truly radical ones like AI, with Heideggerian reaction.

*Seriously, anyone who knows me will recognize my intense distaste for rural America and its regressive political culture. Just look at my location.

[Otto Kretschmer]

When do you guys think will AI be able to create fully fledged conlangs? I asked both Gemini and ChatGPT to create conlangs for me and while both were able to come up with a reasonably good list of sound changes, the overall result was not what I asked for (I asked for a descendant of Middle English and the AI came up with something that looked like a branch in between of North and West Germanic, with some German influence)

[zompist]

When do you guys think will AI be able to create fully fledged conlangs? I asked both Gemini and ChatGPT to create conlangs for me and while both were able to come up with a reasonably good list of sound changes, the overall result was not what I asked for (I asked for a descendant of Middle English and the AI came up with something that looked like a branch in between of North and West Germanic, with some German influence)

Maybe never— not because AI can't do it, but because the Internet may never be as good a set of training data as it is now.

All an LLM is, is a device that can estimate, given a string of words, what the next word is. It's very good at that, but that's because it's been fed unimaginable amounts of training data, including most of the Web. That includes a large number of grammars, including conlangs. AI companies don't care about conlangs, that just happens to be in the training set.

You could get more plausible-sounding results if, say, there were ten times as many conlang grammars. But there's two problems.

One is that the Web is already filling up with AI-generated crap. When you feed an AI AI-generated stuff, quality goes way down. It's learning from flawed imitations of humans, not from humans.

And two, the freewheeling Web of today may not survive. It's already mostly captured by Facebook and other corporate giants; sites like this one are here just because their eccentric founders are relics of the 1990s.

[malloc]

While people here rush to smear tech critics as Heideggerian fascists, take a look at what techies are plotting. Suffice to say, I cannot ever recall any of the tech critics I follow planning anything like that. Quite the contrary, they regard such policies with abject horror. Tell me again who the fascists are in this conflict.

[Torco]

Man, I'm getting more and more luddite about this whole AI thing. look at self-driving cars: they're dangerous, clunky, and still they're getting deployed on streets here and there: why? because they're profitable. but why are they profitable? cause they unload a bunch of the costs onto regular people: bikers getting boinked and whatever. you know what you could do to automate getting stuff and people from here to there? transit! but no, that'd be a public use of a public good: instead, AI is just private uses of public goods. GPT is a great tool to code with, but only cause it's eaten up a bunch of the freely available stack exchange threads or whatever. i feel as if AI seems to just be enclosure 2.0.

[alice]

Getting a bit off-topic here, but a couple of interesting observations:

One is that the Web is already filling up with AI-generated crap. When you feed an AI AI-generated stuff, quality goes way down.

A metaphor I'd like to invoke here is the state of your water jar after you've been painting for a while, or what happens when you mix together plasticine of several different colours.

And two, the freewheeling Web of today may not survive. It's already mostly captured by Facebook and other corporate giants; sites like this one are here just because their eccentric founders are relics of the 1990s.

I doubt we'll ever get back to the time when the Web was full of idiosyncratic and fascinating personal websites. A lot of crap, too, but it does seem to be close to losing its soul. Are we heading irrevocably to an endless procession of shrines to El&n M"sk?

[WeepingElf]

Getting a bit off-topic here, but a couple of interesting observations:

One is that the Web is already filling up with AI-generated crap. When you feed an AI AI-generated stuff, quality goes way down.

A metaphor I'd like to invoke here is the state of your water jar after you've been painting for a while, or what happens when you mix together plasticine of several different colours.

And two, the freewheeling Web of today may not survive. It's already mostly captured by Facebook and other corporate giants; sites like this one are here just because their eccentric founders are relics of the 1990s.

I doubt we'll ever get back to the time when the Web was full of idiosyncratic and fascinating personal websites. A lot of crap, too, but it does seem to be close to losing its soul. Are we heading irrevocably to an endless procession of shrines to El&n M"sk?

Concurred on both points. Not to sound nostalgic, but the Web felt much more fun in the early 90s than it is today. There was a spirit of idealism that seems to have vanished today. Of course, there still are many way cool web sites, perhaps even more in absolute numbers than back then, but they are harder to find now that the Web overflows with (often AI-generated) commercial crap and search engines sell their rankings to the highest bidders.