Interesting article. There is always risk that a new hot technique will get more attention that it ultimately warrants.
For me the key quote in the article is
"Most scientists aren’t trying to mislead anyone, but because they face strong incentives to present favorable results, there’s still a risk that you’ll be misled."
Understanding people's incentives is often very useful when you're looking at what they're saying.
There are those who have realised they can make a lot of cash from it and also get funding by using the term AI. But at the end of the day what software doesn't have some machine learning built in. It's nothing new, nor is the current implementations particularly extraordinary or accurate.
Plenty of software has zero ML. But either way not all ML is the same. There are many different algorithms each with their own tradeoffs. AI as it is presently marketed however usually means one type of AI, large language model, which also has tradeoffs, and is a bit new to the scene compared to say markov chains whose history starts in the early 1900s.
I think this is mostly just a repeat of the problems of academia - no longer truth-seeking, instead focused on citations and careerism. AI is just a.n.other topic where that is happening.
I don't want to generalize because I do not know how widespread this pattern is, but my job has me hopping between a few HPC centers around Germany, and a pattern I notice is that, a lot of these places are chuck full of reject physicists, and a lot of the AI funding that gets distributed gets gobbled up by these people and the consequence of which is a lot of these ML4Science projects.
I personally think it is a bit of a shame, because HPC centers are not there to only serve physicists, and especially with AI funding we in Germany should be doing more AI-core research.
HPCs are usually in Collab with universities for specific science research. Using up their resources is hopping on the bandwagon to damage another industry.an industry (AI) which is neither new nor anywhere close to being anything more than an personal assistant at the moment. Not even a great one at that.
To be fair, the problems of careerism is really a side-effect of academia becoming more enthralled with the private sector, and therefore inheriting it's problems.
If there's one thing working as a software dev has taught me, it's that all decisions are made from a careerist, selfish perspective. Nobody cares what's best, they care what's most impressive and what will personally get them ahead. After it's done, it's not their problem. And nobody can blame them either. This mindset is so pervasive that if you don't subscribe to it, you're a sucker. Because other people will, and they'll just out-lap you. So the result is the same except now you're worse off.
Well, the good news - and I think from the sound of your post you will take it as good news, because you care - is that you are not correct.
Some careers are vocations, and in vocations people work less for egoist reasons and more from the desire to help people. Fortunately in the UK, we still have a very strong example of a vocation - nursing. I know many nurses, none of them can be described as careerist or selfish. So to begin, we know that your statement doesn’t hold true. Nurses’ pay is appalling and career paths are limited, so I’m confident that these many datapoints generalise.
The obvious next question is why academia is not a vocation. You say it’s because it has become too like the private sector. Well, I can tell you that is also wrong, as I have spent many years in both sectors, and the private sector is much less selfish and careerist. This is surprising at first, but I think it’s about incentives.
In the private sector very few people are in direct competition with each other, and it is rarely a zero sum game. The extreme of this is startups, where founders will go to great lengths to help each other. Probably the only area their interests are not aligned is in recruitment, but it is so rare for them to be recruiting the same type of person at exactly the same time that this isn’t really an issue. There are direct competitors of course, but that situation is so exceptional as to be easily ignored.
In academia, however, the incentives encourage selfishness, competition, obstruction, and most of all vicious politics. Academics are not paid well, and mostly compete for egoist rewards such as professorships. I believe in the past this was always somewhat a problem, but it has been exacerbated by multiple factors: (a) very intelligent people mostly left, because more money could be made in finance and tech, and thus little progress can be made and there is no status resulting from genuine science, (b) governments have used research assessment exercises, nonsense bureaucracy invented by fools that encourages silly gaming of stats rather than doing real work, (c) a system of reinforcement where selfish egotists rise at the expense of real scientists, and then - consciously or not - reinforce the system they gamed, thinking it helped them up the ladder and thus must be a good system. The bad drive out the good.
Ultimately the problem is academia is now filled with politicians pretending to be scientists, and such institutional failure is I think a one way street. The only way to fix it is to create new institutions and protect them from infiltration by today’s “scientists”.
This is of course a generalisation, and there are some good eggs left, just not many. Most of them eventually realise they’re surrounded by egoist politicians and eventually leave.
In what sense could you interpret this story as "no longer truth seeking"? Isn't this in fact the opposite, a very clear story of where truth was sought and found?
I've been "lucky" enough to get to trial some AI FEM-like structural solvers.
At best, they're sortof ok for linear, small deformation problems. The kind of models where we could get an exact solution in ~5 minutes vs a fairly sloppy solution in ~30 seconds. Start throwing anything non-linear in and they just fall apart.
Maybe enough to do some very high-level concept selection but even that isn't great. I'm reasonably convinced some of them are just "curvature detectors" - make anything straight blue, anything with high curvature red, and interpolate everything else.
I don't see any reason its not theoretically possible but I doubt it would be that beneficial.
You'd have to map the results back onto the traditional model which has overhead; and using shaky results as a precondition is going to negate a lot of the benefits, especially if its (incorrectly) predicting the part is already in the non-linear stress range which I've seen before. Force balances are all over the place as well (if they even bother to predict them at all, which its not always clear) so it could even be starting from a very unstable point.
Its relatively trivial to just use the native solution from a linear solution as the starting point instead, which is basically what is done anyway with auto time stepping.
I am not a AI booster at all, but the fact that negative results are not published and that everyone is overselling their stuff in research papers is unfortunately not limited to AI. This is just a consequence of the way scientists are evaluated and of the scientific publishing industry, which basically suffers from the same shit than traditional media does (craving for audience).
Sure, it's not. But often on AI papers one sees remarks that actually mean: "...and if you throw in one zillion GPUs and make them run until the end of time you get {magic_benchmark}". Or "if you evaluate this very smart algo in our super-secret, real-life dataset that we claim is available on request, but we'd ghost you if you dare to ask, then you will see this chart that shows how smart we are".
Sure, it is often flag-planting, but when these papers come from big corps, you cannot "just ignore them and keep on" even when there are obvious flaws/issues.
It's a race over resources, as a (former) researcher on a low-budget university, we just cannot compete. We are coerced to believe whatever figure is passed on in the literature as "benchmark", without possibility of replication.
> It's a race over resources, as a (former) researcher on a low-budget university, we just cannot compete. We are coerced to believe whatever figure is passed on in the literature as "benchmark", without possibility of replication.
The central purpose of university research has basically always been that researchers work on hard, foundational topics that are more long-term so that industry is hardly willing to do them. On the other hand, these topics are very important, that is why the respective country is willing to finance this foundational research.
Thus, if you are at a university, once your research topic becomes an arms race with industry, you simply work either at the wrong place (university instead of industry) or on a "wrong" topic in the respective research area (look for some much more long-term, experimental topics that, if you are right, might change the whole research area in, say, 15 years, instead of some high resource-intensive, minor improvements to existing models).
I agree with that. Classically used "AI benchmarks" need to be questioned. In my field, these guys have dropped a bomb, and no one seem to care: https://hal.science/hal-04715638/document
I published my first papers a little over fifteen years ago on practical applications for AI before switching domains. Recently I've been sucked back in.
I agree it's a problem across all of science, but AI seems to attract more than it's fair share of researchers seeking fame and fortune. Exaggerated claims and cherry-picking data seem even more extreme in my limited experience, and even responsible researchers end up exaggerating a bit to try and compete.
I'm not sure why people on HN (of all places) are so divided regarding the perception of AI/ML.
I have not seen anything like it before. We literaly had not system or way of even doing things like code generation based on text input.
Just last week i asked for a script to do image segmentation with a basic UI and claude just generated that for me in under 1 Minute.
I could list tons of examples which are groundbreaking. The whole Image generation stack is completly new.
That blog article is fair enough, there is hype around this topic for sure, but alone for every researcher who needs to write code for their research, AI can make them already a lot more efficient.
But i do believe, that we have entered a new ara: An ara were we take data again very serious. A few years back, you said 'the internet doesn't forget' then we realized that yes the internet starts to forget. Google deleted pages, removed the cache feature and it felt like we stoped caring for data because we didn't knew what to do with it.
Then ai came along. And not only is now data king again but we are now in the mids of reinforcment ara: We now give feedback and the systems incorporate that feedback into their training/learning.
And the ai/ml topic is getting worked on on every single aspect of it: Hardware, Algorithm, use cases, data, tools, protocols, etc. We are in the middle of incorporating and building for and on it. This takes a little bit of time. Still the progress is crazy exhausting.
We will only see in a few years if there is a real ceiling. We do need more GPUs, bigger Datacenters to do a lot more experiments on AI architecture and algorithm. We have a clear bottleneck. Big companies train one big model for weeks and month.
> Just last week i asked for a script to do image segmentation with a basic UI and claude just generated that for me in under 1 Minute.
Thing is we just see that it's copy pasting stack overflow, but now in a fancy way so this is sounding like "I asked Google for a nearby restaurant and it found it in like 500ms, my C64 couldn't do that". It sounds impressive (and it is) because it sounds like "it learned about navigating in the real world and it can now solve everything related to that" but what it actually solved is "fancy lookup in a GIS database". It's useful, damn sure it is, but once the novelty wears off you start seeing it for what it is instead of what you imagine it is.
Edit: to drive the point home.
> claude just generated that
What you think happened is AI is "thinking" and building a ontology over which it reasoned and came to the logical conclusion that this script was the right output. What actually happened is your input correlates to this output according to the trillion examples it saw. There is no ontology. There is no reasoning. There is nothing. Of course this is still impressive and useful as hell, but the novelty will wear off in time. The limitations are obvious by this point.
I'm following LLMs, AI/ML for a few years now and not just on a high level.
There is not a single system out there today which can do what claude can do.
I stil see it for what it is: A technology i can communicate/use with natural language and get a very diverse of tasks done. From writing/generating code, to svgs, to emails, translation etc. etc. etc.
Its a paradigma shift for the whole world literaly.
We finally have a system which encodes not just basic things but high level concepts. And we humans are doing often enough something very similiar.
And what limitations are obvious? Tell me? We have not reached any real ceiling yet. We are limited by GPU capacity or how many architectural experiments a researcher can run. We have plenty of work to do to cleanup the data set we use and have. We need to build more infrastructure, better software support etc.
We have not even reached the phase were we all have local AI/ML chips build in.
We don't even know yet how a system will act if everyone of us has access to very fast inferencing like you already get with groq.
Yeah. It’s just fancier techniques than linear regression. Just like the latter takes a set of numbers and produces another set, LLMs takes words and produces another set of words.
The actual techniques are the breakthrough. The result are fun to play with and may be useful in some occasions, but we don’t have to put them on a pedestal.
You have the wrong idea of how an LLM works. Its more like an model that iteratively finds associating / relevant blocks. The reasoning are the iterative steps it takes.
> “I'm not sure why people on HN (of all places) are so divided regarding the perception of AI/ML.”
Everyone is a rational actor from their individual perspective. The people hyping AI, and the people dismissing the hype both have good reasons.
The is justification to see this new tech as ground breaking. There is justification to be weary about massive theft of data and dismissiveness of privacy.
First, acknowledge and respect that there are so many opinions on any issue. Take yourself out of the equation for a minute. Understand the other side. Really understand it.
>but alone for every researcher who needs to write code for their research, AI can make them already a lot more efficient.
scientists don't need to be efficient, they need to be correct. Software bugs were already a huge cause of scientific error, and responsible for lack of reproducibility, see for example cases like this (https://www.vice.com/en/article/a-code-glitch-may-have-cause...)
Programming in research environments is done with some notoriously questionably variation in quality, as is the case for the industry to be fair, but in research minor errors can ruin results of entire studies. People are fed up and come to much harsher judgements on AI because in an environment like a lab you cannot write software with the attitude of an impressionist painter or the AI equivalent, you need to actually know what you're typing.
AI can make you more efficient if you don't care if you're right, which is maybe cool if you're generating images for your summer beach volleyball event, but it's a disastrous idea if you're writing code in a scientific environment.
I basically agree, but want to point out two major differences to other "hype-y" topics that existed in the past that in my opinion make the whole AI discussions on HN a little bit more controversial than other older hype discussions:
1. The whole investment volume (and thus hope and expectations) into AI is much larger than into other hype topics.
2. Sam Altman, the CEO of OpenAI, was president of YCombinator, the company begind Hacker News, from 2014 to 2019.
The ultimate job of a programmer is to translate human language into computer language. Computers are extremely capable, but they speak a very cryptic overtly logical language.
LLMs are undeniably treading onto that territory. Who knows how far in they will make it, but the wall is breached. Which is unsettling to down right scary depending on your take. It is a real threat to a skill that many have honed for years and for which is very lucrative to have. Programmers don't even need to be replaced, having to settle for $100k/yr in a senior role is almost just a scary.
Google never gave a good reason for why they stopped making their cache public, but my theory is that it was because people were scraping it to train their LLMs.
> Just last week i asked for a script to do image segmentation with a basic UI and claude just generated that for me in under 1 Minute.
I agree that this is useful! It will even take natural language and augment the script, and maybe get it right! Nice!
The AI is combing through scraped data with an LLM, and conjuring forth some imagemagick snippets into a shell script. This is very useful, and if you’re like most people, who don’t know imagemagick intimately, it’s going to save you tons of time.
Where it gets incredibly frustrating is tech leadership seeing these trivial examples, and assuming it extrapolates to general software engineering at their companies. “Oh it writes code, or makes our engineers faster, or whatever. Get the managers mandating this, now! Also, we need to get started on the layoffs. Have them stack rank their reports by who uses AI the best, so that we are ready to pull the trigger.”
But every real engineer who uses these tools on real (as in huge, poorly written) codebases, if they are being honest (they may not be, given the stack ranking), will tell you “on a good day it multiplies my productivity by, let’s say, 1.1-2x? On a bad day, I end up scrapping 10k lines of LLM code, reading some documentation on my own, and solving the problem with 5 lines of intentional code.”
Please, PLEASE pay attention to this details that I added: Huge, poorly written codebases. This is just the reality at most software companies that have graduated from series A startup. What my colleagues and I are trying to tell you, leadership, is that these “it made a script” and “it made a html form with a backend” examples ARE NOT cleanly extrapolating to the flaming dumpster fire codebases we actually work with. Sometimes the tools help! Sometimes, they don’t.
It’s as if LLM is just another tool we use sometimes.
This is why I am annoyed. It’s incredibly frustrating to be told by your boss “use tool or get fired” when that tool doesn’t always fit the task at hand. It DOES NOT mean I see zero value in LLMs.
most work in software jobs is not making one-off scripts like in your example. a lot of the job is about modifying existing codebases which include in-house approachs to style and services along with various third party frameworks like Spring driven by annotations, and requirements around how to write tests and how many. AI is just not very helpful here, you spend more time spinning wheels trying to craft the absolute perfect script than just making code changes directly.
There is no single reason. Nobody will argue that LLMs are already quite useful at some tasks if used properly.
As for the opposing view, there are so many reasons.
* Founders and other people who bet their money on AI try to pump up the hype in spite of problems with delivery
* We know some of them are plainly lying, but the general public doesn't
* They repeat their assumptions as facts ("AI will replace most X and Y jobs by year Z")
* We clearly see that the enormous development of LLMs has plateaued but they try to convince the general public it's the contrary
* We see the difference on how a single individual (Aaron Swartz) is treated when making a small copyright infringement, and how the consequences for AI companies like OpenAI or Meta who copied the whole contents of Libgen are non-existent.
* Some people like me just hate AI slop - in writing and imaging. It just puts me off and I stop reading/watching etc.
The article initially appears to suggest that all AI in science (or at least the author’s field) is hype. But their gripe seems to be specific to an architecture named PINN that seems to be overhyped, as they mention in the end how they end up using other DL models to successfully compute PDEs faster than traditional numerical methods.
It's more widespread than PINNs. PINNs have been widely known to be rubbish a long time ago. But the general failure of using ML for physics problems is much more widespread.
Where ML generally shines is either when you have relatively lots of experimental data with respect to a fairly narrow domain. This is the case for machine learned interatomic potentials MLIPs which have been a thing since the '90s. Also potentially the case for weather modelling (but I do not want to comment about that). Or when you have absolute insane amounts of data, and you train a really huge model. This is what we refer to as AI. This is basically why Alphafold is successful, and Alphafold still fails to produce good results when you query it on inputs that are far from any data points in its training data.
But most ML for physics problems tend to be somewhere in between. Lacking experimental data and working with not enough simulation data because it is so expensive to produce. And also training models that are not large enough, because inference would be too slow, anyway, if they were too big. And then expecting these models to learn a very wide range of physics.
And then everyone jumps in on the hype train, because it is so easy to give it a shot. And everyone gets the same dud results. But then they publish anyway. And if the lab/PI is famous enough or if they formulate the problem in a way that is unique and looks sciency or mathy, they might even get their paper in a good journal/conference and get lots of citations. But in the end, they still only end up with the same results as everyone else: replicates the training data to some extent, somebody else should work on the generalizability problem.
The use of the term «AI» is, yet again, annoying by its vagueness.
I'm assuming that they do not refer to the general use of machines to solve differential equations (whether exactly or approximately), which is centuries old (Babbage's engine).
But then how restricted these «Physics-Informed Neural Networks» are ? Are there other methods using Neural Networks to solve differential equations ?
Replace PINN with any "AI" solution for anything and you'll still find it overhyped.
The only realistic evaluations of "AI" so far are those that admit it's only useful for experts to skip some boring work. And triple check the output after.
> After a few weeks of failure, I messaged a friend at a different university, who told me that he too had tried using PINNs, but hadn’t been able to get good results.
not really related to AI but this reflects a lesson I learned too late during some research in college: constant collaboration is important because it helps you avoid retreading over areas where others have already failed
Another reason why the idea of AI agents for science hasn't made much sense to me. Research is an extremely collaborative set of activities. How good would a researcher be who is very good at literature review, but never actually talks to anyone, goes to any conferences, etc?
Great analysis and spot on examples. Another issue with AI related research is that a lot of papers are new and not that many get published in “proper” places, yet being quoted right/left/center, just look at google scholar. It is hard to repro the results and check the validity of some statements, not to mention that research which was done 4 years ago used one set of models and now another set of models with different training data is used in tests. It is hard to establish what really affects the results and if the conclusions are applicable to some specific property of the outdated model or if it is even generalisable.
Readit News
For me the key quote in the article is
"Most scientists aren’t trying to mislead anyone, but because they face strong incentives to present favorable results, there’s still a risk that you’ll be misled."
Understanding people's incentives is often very useful when you're looking at what they're saying.
Don't underestimate this. I'm peripherally in the startup world and funding has dried up for everyone unless you have some kind of AI story.
So people shoehorn in AI to their company stories.
Utter nonsense, these are some of the smartest people in the world who do incredibly valuable science.
If there's one thing working as a software dev has taught me, it's that all decisions are made from a careerist, selfish perspective. Nobody cares what's best, they care what's most impressive and what will personally get them ahead. After it's done, it's not their problem. And nobody can blame them either. This mindset is so pervasive that if you don't subscribe to it, you're a sucker. Because other people will, and they'll just out-lap you. So the result is the same except now you're worse off.
Some careers are vocations, and in vocations people work less for egoist reasons and more from the desire to help people. Fortunately in the UK, we still have a very strong example of a vocation - nursing. I know many nurses, none of them can be described as careerist or selfish. So to begin, we know that your statement doesn’t hold true. Nurses’ pay is appalling and career paths are limited, so I’m confident that these many datapoints generalise.
The obvious next question is why academia is not a vocation. You say it’s because it has become too like the private sector. Well, I can tell you that is also wrong, as I have spent many years in both sectors, and the private sector is much less selfish and careerist. This is surprising at first, but I think it’s about incentives.
In the private sector very few people are in direct competition with each other, and it is rarely a zero sum game. The extreme of this is startups, where founders will go to great lengths to help each other. Probably the only area their interests are not aligned is in recruitment, but it is so rare for them to be recruiting the same type of person at exactly the same time that this isn’t really an issue. There are direct competitors of course, but that situation is so exceptional as to be easily ignored.
In academia, however, the incentives encourage selfishness, competition, obstruction, and most of all vicious politics. Academics are not paid well, and mostly compete for egoist rewards such as professorships. I believe in the past this was always somewhat a problem, but it has been exacerbated by multiple factors: (a) very intelligent people mostly left, because more money could be made in finance and tech, and thus little progress can be made and there is no status resulting from genuine science, (b) governments have used research assessment exercises, nonsense bureaucracy invented by fools that encourages silly gaming of stats rather than doing real work, (c) a system of reinforcement where selfish egotists rise at the expense of real scientists, and then - consciously or not - reinforce the system they gamed, thinking it helped them up the ladder and thus must be a good system. The bad drive out the good.
Ultimately the problem is academia is now filled with politicians pretending to be scientists, and such institutional failure is I think a one way street. The only way to fix it is to create new institutions and protect them from infiltration by today’s “scientists”.
This is of course a generalisation, and there are some good eggs left, just not many. Most of them eventually realise they’re surrounded by egoist politicians and eventually leave.
At best, they're sortof ok for linear, small deformation problems. The kind of models where we could get an exact solution in ~5 minutes vs a fairly sloppy solution in ~30 seconds. Start throwing anything non-linear in and they just fall apart.
Maybe enough to do some very high-level concept selection but even that isn't great. I'm reasonably convinced some of them are just "curvature detectors" - make anything straight blue, anything with high curvature red, and interpolate everything else.
You'd have to map the results back onto the traditional model which has overhead; and using shaky results as a precondition is going to negate a lot of the benefits, especially if its (incorrectly) predicting the part is already in the non-linear stress range which I've seen before. Force balances are all over the place as well (if they even bother to predict them at all, which its not always clear) so it could even be starting from a very unstable point.
Its relatively trivial to just use the native solution from a linear solution as the starting point instead, which is basically what is done anyway with auto time stepping.
Anyway, winter is coming, innit?
Sure, it is often flag-planting, but when these papers come from big corps, you cannot "just ignore them and keep on" even when there are obvious flaws/issues.
It's a race over resources, as a (former) researcher on a low-budget university, we just cannot compete. We are coerced to believe whatever figure is passed on in the literature as "benchmark", without possibility of replication.
The central purpose of university research has basically always been that researchers work on hard, foundational topics that are more long-term so that industry is hardly willing to do them. On the other hand, these topics are very important, that is why the respective country is willing to finance this foundational research.
Thus, if you are at a university, once your research topic becomes an arms race with industry, you simply work either at the wrong place (university instead of industry) or on a "wrong" topic in the respective research area (look for some much more long-term, experimental topics that, if you are right, might change the whole research area in, say, 15 years, instead of some high resource-intensive, minor improvements to existing models).
I agree it's a problem across all of science, but AI seems to attract more than it's fair share of researchers seeking fame and fortune. Exaggerated claims and cherry-picking data seem even more extreme in my limited experience, and even responsible researchers end up exaggerating a bit to try and compete.
I have not seen anything like it before. We literaly had not system or way of even doing things like code generation based on text input.
Just last week i asked for a script to do image segmentation with a basic UI and claude just generated that for me in under 1 Minute.
I could list tons of examples which are groundbreaking. The whole Image generation stack is completly new.
That blog article is fair enough, there is hype around this topic for sure, but alone for every researcher who needs to write code for their research, AI can make them already a lot more efficient.
But i do believe, that we have entered a new ara: An ara were we take data again very serious. A few years back, you said 'the internet doesn't forget' then we realized that yes the internet starts to forget. Google deleted pages, removed the cache feature and it felt like we stoped caring for data because we didn't knew what to do with it.
Then ai came along. And not only is now data king again but we are now in the mids of reinforcment ara: We now give feedback and the systems incorporate that feedback into their training/learning.
And the ai/ml topic is getting worked on on every single aspect of it: Hardware, Algorithm, use cases, data, tools, protocols, etc. We are in the middle of incorporating and building for and on it. This takes a little bit of time. Still the progress is crazy exhausting.
We will only see in a few years if there is a real ceiling. We do need more GPUs, bigger Datacenters to do a lot more experiments on AI architecture and algorithm. We have a clear bottleneck. Big companies train one big model for weeks and month.
Thing is we just see that it's copy pasting stack overflow, but now in a fancy way so this is sounding like "I asked Google for a nearby restaurant and it found it in like 500ms, my C64 couldn't do that". It sounds impressive (and it is) because it sounds like "it learned about navigating in the real world and it can now solve everything related to that" but what it actually solved is "fancy lookup in a GIS database". It's useful, damn sure it is, but once the novelty wears off you start seeing it for what it is instead of what you imagine it is.
Edit: to drive the point home.
> claude just generated that
What you think happened is AI is "thinking" and building a ontology over which it reasoned and came to the logical conclusion that this script was the right output. What actually happened is your input correlates to this output according to the trillion examples it saw. There is no ontology. There is no reasoning. There is nothing. Of course this is still impressive and useful as hell, but the novelty will wear off in time. The limitations are obvious by this point.
There is not a single system out there today which can do what claude can do.
I stil see it for what it is: A technology i can communicate/use with natural language and get a very diverse of tasks done. From writing/generating code, to svgs, to emails, translation etc. etc. etc.
Its a paradigma shift for the whole world literaly.
We finally have a system which encodes not just basic things but high level concepts. And we humans are doing often enough something very similiar.
And what limitations are obvious? Tell me? We have not reached any real ceiling yet. We are limited by GPU capacity or how many architectural experiments a researcher can run. We have plenty of work to do to cleanup the data set we use and have. We need to build more infrastructure, better software support etc.
We have not even reached the phase were we all have local AI/ML chips build in.
We don't even know yet how a system will act if everyone of us has access to very fast inferencing like you already get with groq.
The actual techniques are the breakthrough. The result are fun to play with and may be useful in some occasions, but we don’t have to put them on a pedestal.
Everyone is a rational actor from their individual perspective. The people hyping AI, and the people dismissing the hype both have good reasons.
The is justification to see this new tech as ground breaking. There is justification to be weary about massive theft of data and dismissiveness of privacy.
First, acknowledge and respect that there are so many opinions on any issue. Take yourself out of the equation for a minute. Understand the other side. Really understand it.
Take a long walk in other people’s shoes.
scientists don't need to be efficient, they need to be correct. Software bugs were already a huge cause of scientific error, and responsible for lack of reproducibility, see for example cases like this (https://www.vice.com/en/article/a-code-glitch-may-have-cause...)
Programming in research environments is done with some notoriously questionably variation in quality, as is the case for the industry to be fair, but in research minor errors can ruin results of entire studies. People are fed up and come to much harsher judgements on AI because in an environment like a lab you cannot write software with the attitude of an impressionist painter or the AI equivalent, you need to actually know what you're typing.
AI can make you more efficient if you don't care if you're right, which is maybe cool if you're generating images for your summer beach volleyball event, but it's a disastrous idea if you're writing code in a scientific environment.
Still a lot of researchers can benefit from code tools for their daily work to make them a lot faster.
And plenty of strategies exist to saveguard this. Tool use for example, unit tests etc.
I've seen this over and over again and been on different sides of the question on different technologies at different times.
To me, this is same as it ever was!
1. The whole investment volume (and thus hope and expectations) into AI is much larger than into other hype topics.
2. Sam Altman, the CEO of OpenAI, was president of YCombinator, the company begind Hacker News, from 2014 to 2019.
LLMs are undeniably treading onto that territory. Who knows how far in they will make it, but the wall is breached. Which is unsettling to down right scary depending on your take. It is a real threat to a skill that many have honed for years and for which is very lucrative to have. Programmers don't even need to be replaced, having to settle for $100k/yr in a senior role is almost just a scary.
I agree that this is useful! It will even take natural language and augment the script, and maybe get it right! Nice!
The AI is combing through scraped data with an LLM, and conjuring forth some imagemagick snippets into a shell script. This is very useful, and if you’re like most people, who don’t know imagemagick intimately, it’s going to save you tons of time.
Where it gets incredibly frustrating is tech leadership seeing these trivial examples, and assuming it extrapolates to general software engineering at their companies. “Oh it writes code, or makes our engineers faster, or whatever. Get the managers mandating this, now! Also, we need to get started on the layoffs. Have them stack rank their reports by who uses AI the best, so that we are ready to pull the trigger.”
But every real engineer who uses these tools on real (as in huge, poorly written) codebases, if they are being honest (they may not be, given the stack ranking), will tell you “on a good day it multiplies my productivity by, let’s say, 1.1-2x? On a bad day, I end up scrapping 10k lines of LLM code, reading some documentation on my own, and solving the problem with 5 lines of intentional code.”
Please, PLEASE pay attention to this details that I added: Huge, poorly written codebases. This is just the reality at most software companies that have graduated from series A startup. What my colleagues and I are trying to tell you, leadership, is that these “it made a script” and “it made a html form with a backend” examples ARE NOT cleanly extrapolating to the flaming dumpster fire codebases we actually work with. Sometimes the tools help! Sometimes, they don’t.
It’s as if LLM is just another tool we use sometimes.
This is why I am annoyed. It’s incredibly frustrating to be told by your boss “use tool or get fired” when that tool doesn’t always fit the task at hand. It DOES NOT mean I see zero value in LLMs.
As for the opposing view, there are so many reasons.
* Founders and other people who bet their money on AI try to pump up the hype in spite of problems with delivery
* We know some of them are plainly lying, but the general public doesn't
* They repeat their assumptions as facts ("AI will replace most X and Y jobs by year Z")
* We clearly see that the enormous development of LLMs has plateaued but they try to convince the general public it's the contrary
* We see the difference on how a single individual (Aaron Swartz) is treated when making a small copyright infringement, and how the consequences for AI companies like OpenAI or Meta who copied the whole contents of Libgen are non-existent.
* Some people like me just hate AI slop - in writing and imaging. It just puts me off and I stop reading/watching etc.
There are many more points like this.
Where ML generally shines is either when you have relatively lots of experimental data with respect to a fairly narrow domain. This is the case for machine learned interatomic potentials MLIPs which have been a thing since the '90s. Also potentially the case for weather modelling (but I do not want to comment about that). Or when you have absolute insane amounts of data, and you train a really huge model. This is what we refer to as AI. This is basically why Alphafold is successful, and Alphafold still fails to produce good results when you query it on inputs that are far from any data points in its training data.
But most ML for physics problems tend to be somewhere in between. Lacking experimental data and working with not enough simulation data because it is so expensive to produce. And also training models that are not large enough, because inference would be too slow, anyway, if they were too big. And then expecting these models to learn a very wide range of physics.
And then everyone jumps in on the hype train, because it is so easy to give it a shot. And everyone gets the same dud results. But then they publish anyway. And if the lab/PI is famous enough or if they formulate the problem in a way that is unique and looks sciency or mathy, they might even get their paper in a good journal/conference and get lots of citations. But in the end, they still only end up with the same results as everyone else: replicates the training data to some extent, somebody else should work on the generalizability problem.
I'm assuming that they do not refer to the general use of machines to solve differential equations (whether exactly or approximately), which is centuries old (Babbage's engine).
But then how restricted these «Physics-Informed Neural Networks» are ? Are there other methods using Neural Networks to solve differential equations ?
The only realistic evaluations of "AI" so far are those that admit it's only useful for experts to skip some boring work. And triple check the output after.
not really related to AI but this reflects a lesson I learned too late during some research in college: constant collaboration is important because it helps you avoid retreading over areas where others have already failed
There are practices like pre-registration, open data, etc. that can make results much more transparent and replicable.