Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
-
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
-
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
@inthehands Given pictures of a giraffe, a rhinoceros, an elephant, and a squirrel, find the squirrel. Feel free to reference dictionaries, encyclopedias, nature documentaries and previous responses to this same question.
-
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
Back in 2023, OpenAI was hyping ChatGPT by claiming "it can pass the GRE".
When all it was doing was autocomplete of the answer keys that had been used as input.
-
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
Once upon a time, a boss came and tried to hurry me into giving him some code I was working on.
Me: Do you want it now, or do you want it right?
Him: Good point! When it's ready.
-
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
It’s just •astonishing• how many eye-popping stories of LLMs doing amazing omg-verge-of-magical-superintelligence things turn out to be just unvarnished plagiarism.
In the first months after ChatGPT’s release, I remember a French dept colleague being amazed that GPT could translate and summarize a passage of Le Petit Prince.
It was a lot less impressive when I dug up the 2 or 3 online passages which it had copied almost verbatim and stitched together (sprinkling in a couple of extra words that made it less accurate).
3/
-
It’s just •astonishing• how many eye-popping stories of LLMs doing amazing omg-verge-of-magical-superintelligence things turn out to be just unvarnished plagiarism.
In the first months after ChatGPT’s release, I remember a French dept colleague being amazed that GPT could translate and summarize a passage of Le Petit Prince.
It was a lot less impressive when I dug up the 2 or 3 online passages which it had copied almost verbatim and stitched together (sprinkling in a couple of extra words that made it less accurate).
3/
To be fair, competent plagiarism is a nontrivial task. It takes skill for a human to do: finding relevant fragments from an input dataset and stitching those fragments together in a well-formed way is not nothing! It’s impressive and interesting that people have figured out how to make machines do it.
But •that• is not the promise on which investors are valuing the AI megajuggernaut at trillions of dollars.
4/
-
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
@inthehands I couldn't love this perspective anymore if I tried!
-
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
Many years ago, the following idea spread in the art community: a 15 minutes doodle is the result of years of practice and study. In other words, the 15 minutes doodle took your entire life to make.
The same goes to LLMs: it did not take 90 minutes. It took the thousands of hours all the other softwares took to make, plus the years of study of all the programmers making them. And the thousands of hours manually labeling source codes...
-
It’s just •astonishing• how many eye-popping stories of LLMs doing amazing omg-verge-of-magical-superintelligence things turn out to be just unvarnished plagiarism.
In the first months after ChatGPT’s release, I remember a French dept colleague being amazed that GPT could translate and summarize a passage of Le Petit Prince.
It was a lot less impressive when I dug up the 2 or 3 online passages which it had copied almost verbatim and stitched together (sprinkling in a couple of extra words that made it less accurate).
3/
@inthehands (Completely aside: I sometimes feel like I ought to learn French *just* to be able to read that one in its original language. It is very special to me; it was the first book I ever read myself.
Even more completely aside: Someone ought to give Hayao Miyazaki the rights to create an animated interpretation of it, and supply him with however much coffee and pure adrenochrome needed to keep him alive, active and in good spirits for however long that takes. And reward him with whatever he wanted - for example, the exclusive right to hunt Sam Altman for sport.)
-
To be fair, competent plagiarism is a nontrivial task. It takes skill for a human to do: finding relevant fragments from an input dataset and stitching those fragments together in a well-formed way is not nothing! It’s impressive and interesting that people have figured out how to make machines do it.
But •that• is not the promise on which investors are valuing the AI megajuggernaut at trillions of dollars.
4/
The reasons we consider plagiarism to be cheating in school and malpractice in professional contexts are many of the same reasons that LLMs are not going to replace all knowledge-based human labor. (Details left as an exercise for the reader.)
And yes, a whole lot of what LLMs do •would• count as plagiarism if one of my students did it manually, and •should• count as plagiarism just the same if they use a machine to do it — not just in a “that’s cheating!!” sense, but more importantly in a “that’s not really doing the work” sense.
5/
-
Many years ago, the following idea spread in the art community: a 15 minutes doodle is the result of years of practice and study. In other words, the 15 minutes doodle took your entire life to make.
The same goes to LLMs: it did not take 90 minutes. It took the thousands of hours all the other softwares took to make, plus the years of study of all the programmers making them. And the thousands of hours manually labeling source codes...
@Lily_and_frog @inthehands I think that has been said for plumbing—$1 for the part, $45 for the five minutes it took to install the part, $400 for the years of experience it takes to know which part and how to install it.
-
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
@inthehands it annoys me so much when people cite AI benchmarks for exactly this reason.
Either the logic of the benchmarks are private (and thus not really fit for judgement) or they're public in a way that effectively makes the supposed challenge of that benchmark into an open-book test. -
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
@inthehands My company houses a number of pro-AI engineers. I asked them to evaluate Claude (which I understand is their favorite) by asking it to build something broken in a very specific, novel way.
My suggestion wasn't taken seriously, unfortunately, but I suppose it might be interesting. -
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
@inthehands
Grep could do it faster.
-
@inthehands (Completely aside: I sometimes feel like I ought to learn French *just* to be able to read that one in its original language. It is very special to me; it was the first book I ever read myself.
Even more completely aside: Someone ought to give Hayao Miyazaki the rights to create an animated interpretation of it, and supply him with however much coffee and pure adrenochrome needed to keep him alive, active and in good spirits for however long that takes. And reward him with whatever he wanted - for example, the exclusive right to hunt Sam Altman for sport.)
@datarama @inthehands Miyazaki's headed for retirement (again), may I propose Guillermo del Toro?
-
RE: https://unstable.systems/@jneen/116618931097778342
Worth looking at both the quoted text here and •especially• the linked page, which is quite good.
I’ll add another item of my own. The first screenshot mentions giving an LLM the task of “implementing an HTTP server in JavaScript from scratch” in 90 minutes. Sounds impressive, right? Until you remember that every open-source Javascript HTTP server in existence ••was in the training data••.
1/
@inthehands Additionally, why would I want a tool to write me an http server? I already have lots of http servers.
What's next? Are you going to invent a tool that will mail me thousands of free AOL CD's?
-
The reasons we consider plagiarism to be cheating in school and malpractice in professional contexts are many of the same reasons that LLMs are not going to replace all knowledge-based human labor. (Details left as an exercise for the reader.)
And yes, a whole lot of what LLMs do •would• count as plagiarism if one of my students did it manually, and •should• count as plagiarism just the same if they use a machine to do it — not just in a “that’s cheating!!” sense, but more importantly in a “that’s not really doing the work” sense.
5/
One last example:
The first LLM code example that really made my eyes pop was early after the release of GPT, when somebody got it to combine Breakout with Conway’s Game of Life (a truly delightful idea). It worked!
Funny thing: the Breakout code and the Life code had a •completely• different style and flavor. Red flag. In about 15 minutes of web searching, I was able to find one of the projects (can’t remember if it was the Breakout or the Life half) which it had copied wholesale, with just a few variable renames. And the other half? It was in Python, but it used dictionaries where it really should have used objects — tons of `thing["prop"]` where it should have said `thing.prop`, and lots of other un-Pythonic stuff besides. It was a machine translate of code from another language, very likely Javascript.
The entire thing was a plagiarized Breakout and a plagiarized Game of Life, one transpiled, and all stuck together in a single run loop. To be fair, figuring out how to (1) run both halves of the logic from a single loop and (2) count the Life cells as Breakout bricks is work I'd cheer on from a second-semester intro CS student! It's not, however, quite what's being sold by these companies.
6/
-
IT WAS IN THE TRAINING DATA. Your test gave the machine a pile of correct answers and free license to plagiarize.
I remember people being wowed that Claude Code could implement a complete C compiler. But somehow it doesn’t sound quite as impressive when you phrase it as “given every existing C compiler as input, the LLM can produce a C compiler as output.”
2/
@inthehands @rayckeith Same with Mythos - the benchmark tests are in the training data. Cybergym tracks their performance and their website has a link to their github which refers to all the evaluation data being in huggingface (with the git repo link to it). It stands to reason that every model has been trained on the very thing they are then evaluated on.
-
@datarama @inthehands Miyazaki's headed for retirement (again), may I propose Guillermo del Toro?
@Linza @inthehands Miyazaki was born to make that movie, and it is a cosmic injustice of the highest order if he doesn't get to do it.
-
@Lily_and_frog @inthehands I think that has been said for plumbing—$1 for the part, $45 for the five minutes it took to install the part, $400 for the years of experience it takes to know which part and how to install it.
Yep!
But in the case of LLMs, it's thousands and thousands of years of experience + hundreds and hundreds of years of just... labelling the source material!
