Yes, AI allows for exquisite demos, demos that tantalize the audience into thinking of the infinite potential of the technology, that stunning vision expands and expands until the universe of potential overwhelms the dreamer into a state of terminal fantasy. So it is always a solution looking for a problem. There are cases where the two meet more realistically and a valuable impactful company develops it.
The fact it can generate human language that is very compelling for certain context, makes it seem possible of doing so for many, many more contexts.
"then fails to consistently help in completing tasks when deployed for daily use."
This article seems to be baitware trying to push some outdated perspective. LLMs have only gotten more powerful over the last 3 years (being able to do more things), and so far not much has stopped them from becoming even more powerful (with the help of reasoning, other external methods, etc) in the future.
"daily use" is so subjective and this article will be out dated soon as we get closer to an AGI (with LLMs as the base layer and not the main driver)
The trouble is that peoples' self-evaluation of things that they believe are helping them is generally poor, and there's, at best, weak and conflicting evidence which is _not_ based on polling users.
In particular, "producing stuff" is not necessarily "creating value"; some stuff has _negative_ value.
It's wild to me that, of all the things to call LLMs out for, this piece has chosen to include math tutoring. I've been doing Math Academy for a bit over 6 months now, going from (essentially) Algebra II through Calc II (integration by parts, arc lengths, Taylor expansions) and LLMs have been a huge part of what has made that effective:
* Clear explanation of concepts that respond to questions and reformulate when things bounce
* Step-by-step verification of solutions, spotting exactly where calculations have gone
* Instantaneously generating new problem sets to reinforce concepts
LLMs are probably not going to live up to all sorts of claims their proponents make. But I don't think you can ever have tried to use an LLM in a math course and reach the conclusion that it's "demoware" for that application. At what point, over 6 months of continuous work, does it stop being a "demo"?
LLMs are useful if you use them properly and they are getting better everyday. Arguing against LLMs is like arguing against a shovel. Just use it right.
> “Demoware” is a type of software that looks “good” during a demonstration.
I like the term. I have been using a similar phrase "looks good in a snippet" when referring to certain styles of programming.
Once such instance was when nodejs was becoming popular and everyone was showing how easy concurrent programming can be with a few callbacks in a snippet. However building a large code base with that would eventually turn into a nightmare.
Another example is databases which don't fsync after writes by default. They look great in benchmarks (webscale even!) then in production suddenly some of data goes missing. But at least those initial benchmark demos were impressive.
The initial ChatGPT release in 2022 was the product of 7 years of private research that in turn built on decades of public research.
Rumors say that Google wasn't far behind at the time, but didn't push releases. Perhaps because they were not that impressed by the applications or did not want "AI" to cannibalize their other products.
So it seems very likely that everything has been squeezed out of the decades of research and we have plateaued.
Desperate measures like Nvidia buying its own graphics cards through circular investment schemes do not inspire confidence either. Or Microsoft now doing CoPilot product placement ads in teenager YouTube channels. When Google launched, people just used it because it was good. This all fits very well with the demoware angle of the article.
14 comments
[ 3.3 ms ] story [ 41.4 ms ] threadThe fact it can generate human language that is very compelling for certain context, makes it seem possible of doing so for many, many more contexts.
This article seems to be baitware trying to push some outdated perspective. LLMs have only gotten more powerful over the last 3 years (being able to do more things), and so far not much has stopped them from becoming even more powerful (with the help of reasoning, other external methods, etc) in the future.
"daily use" is so subjective and this article will be out dated soon as we get closer to an AGI (with LLMs as the base layer and not the main driver)
In particular, "producing stuff" is not necessarily "creating value"; some stuff has _negative_ value.
Edit: I mean their outputs are procedurally generated, like in https://en.m.wikipedia.org/wiki/Demoscene
* Clear explanation of concepts that respond to questions and reformulate when things bounce
* Step-by-step verification of solutions, spotting exactly where calculations have gone
* Instantaneously generating new problem sets to reinforce concepts
LLMs are probably not going to live up to all sorts of claims their proponents make. But I don't think you can ever have tried to use an LLM in a math course and reach the conclusion that it's "demoware" for that application. At what point, over 6 months of continuous work, does it stop being a "demo"?
I like the term. I have been using a similar phrase "looks good in a snippet" when referring to certain styles of programming.
Once such instance was when nodejs was becoming popular and everyone was showing how easy concurrent programming can be with a few callbacks in a snippet. However building a large code base with that would eventually turn into a nightmare.
Another example is databases which don't fsync after writes by default. They look great in benchmarks (webscale even!) then in production suddenly some of data goes missing. But at least those initial benchmark demos were impressive.
Rumors say that Google wasn't far behind at the time, but didn't push releases. Perhaps because they were not that impressed by the applications or did not want "AI" to cannibalize their other products.
So it seems very likely that everything has been squeezed out of the decades of research and we have plateaued.
Desperate measures like Nvidia buying its own graphics cards through circular investment schemes do not inspire confidence either. Or Microsoft now doing CoPilot product placement ads in teenager YouTube channels. When Google launched, people just used it because it was good. This all fits very well with the demoware angle of the article.