Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI in my plasma physics research didn’t go the way I expected

understandingai.org

Post reply on HN
Print view View original

291–300 of 307 posts

Re: AI in my plasma physics research didn’t go the way I expected

#291

Post by vrighter »

sailingparrot wrote: post #225
vrighter wrote: post #195

Earlier quoted context omitted.

If there is a single objective right answer, the model should output a probability of 1 for it, and 0 for everything else. Ex. If I ask "Is a sphere a curved object?" The one and only answer is "100% yes" not "I am 99% sure it is" (and once in a while actually say it isn't) This is pretty much impossible to achieve with current architectures (which aren't all that different to those of old, just bigger). If they did,…

> Anyone who understands the tech does know this Yes, and this is does not mean the technology can never be useful. I work everyday with that have false beliefs about a tech, I have a friend that until recently thought there were rivers on the moon, and some believe climate change is a hoax, I often forget things people told me and they have to tell me again. Are humans not useful at anything ?

If someone responds with "no" to that question even once, then yes, I don't consider them trustworthy for anything involving complex thought.

Re: AI in my plasma physics research didn’t go the way I expected

#292

Post by boxed »

flobosg wrote: post #251
boxed wrote: post #163

Earlier quoted context omitted.

I'm personally waiting for the other shoe to drop here. I suspect that, since nature begins with an existing protein and modifies it slightly, AlphaFold is crazy overfitted to the training data. Furthermore, the enormous success of AlphaFold means that the number of people doing protein structure solving has likely crashed. So not only are we using an overfitting model that probably can't handle truly novel proteins,…

> that probably can't handle truly novel proteins AlphaFold is able to predict novel folds, see https://www.nature.com/articles/s42003-022-03357-1

Could once is not the same as can predictably first of all.

Second, how can they possibly know this fold isn't actually in the ENTIRE PDB? I doubt very much that the can. The PDB is enormous.

Re: AI in my plasma physics research didn’t go the way I expected

#293

Post by boxed »

pixl97 wrote: post #223
boxed wrote: post #163

Earlier quoted context omitted.

I'm personally waiting for the other shoe to drop here. I suspect that, since nature begins with an existing protein and modifies it slightly, AlphaFold is crazy overfitted to the training data. Furthermore, the enormous success of AlphaFold means that the number of people doing protein structure solving has likely crashed. So not only are we using an overfitting model that probably can't handle truly novel proteins,…

Why do you expect this, or is this just a "I need to find a reason to hate AI" thing?

It's like every time AI "can code" and then falls on its face when presented with super basic problems that are outside its training data. Why would protein folding be different?

Re: AI in my plasma physics research didn’t go the way I expected

#294

Post by Closi »

strogonoff wrote: post #272
Closi wrote: post #169

Earlier quoted context omitted.

Totally agree. Also the term 'LLM' is more about the mechanics of the thing than what the user gets. LLM is the technology, but some sort of automated artificial intelligence is what people are generally buying. As an example, when people use ChatGPT and get an image back, most don't think "oh, so the LLM called out to a diffusion API?" - they just think "oh chat GPT can give me an image if I give it a prompt". Altho…

> As an example, when people use ChatGPT and get an image back, most don't think "oh, so the LLM called out to a diffusion API?" - they just think "oh chat GPT can give me an image if I give it a prompt". Note: your first part skipped entirely the process of obtaining the data for and training of both of the above, which is a crucial part at least on par with what called which API. I don’t think it’s unreasonable to…

I like your analogy, but I don't think people even go "the cow gave this milk" - I think they tend to just go "mmmm yummy yummy milk"

People usually see the product rather than the process.

Re: AI in my plasma physics research didn’t go the way I expected

#295

Post by epistasis »

rhubarbtree wrote: post #5

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.

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?

Re: AI in my plasma physics research didn’t go the way I expected

#296

Post by gthompson512 »

literalAardvark wrote: post #246
gthompson512 wrote: post #180

Earlier quoted context omitted.

> "the start of the banana zone" What does this mean? Is it some slang for exponential growth, or is it a reference to something like the "paperclip maximizer"?

It's slang for the J part of the exponential curve. Didn't expect that to be a problem here, sorry.

There is a bunch of new phrases people have been using around AI topics, so it can be hard to tell what exactly is being talked about, like "Roko's Basilisk", the "lottery card hypothesis?"(not quite remembering the phrasing on this one), "the bitter lesson", "paperclip maximizing", "stochastic parrot", etc.. Thank you for clarifying. I was kind of hoping there was a fun blog or story with a "banana zone".

Re: AI in my plasma physics research didn’t go the way I expected

#297

Post by strogonoff »

Closi wrote: post #294
strogonoff wrote: post #272

Earlier quoted context omitted.

> As an example, when people use ChatGPT and get an image back, most don't think "oh, so the LLM called out to a diffusion API?" - they just think "oh chat GPT can give me an image if I give it a prompt". Note: your first part skipped entirely the process of obtaining the data for and training of both of the above, which is a crucial part at least on par with what called which API. I don’t think it’s unreasonable to…

I like your analogy, but I don't think people even go "the cow gave this milk" - I think they tend to just go "mmmm yummy yummy milk" People usually see the product rather than the process.

I disagree; people are not mindless consumers, people are interested in what comes from where, and given the knowledge most of them would make a choice that would be ethically good, when they can afford it.

What breaks this (and prevents the free market from working as intended) is lack of said knowledge, i.e., information asymmetry. In case of the milk example, I think it mostly comes to two factors:

1) Lack of this awareness is financially beneficial to respective industries. (No need for conspiracy theories, but they sure as hell not going to engage in educational campaigns on these topics, or facilitating any such efforts in any way, including not suing them. Further complicated by the fact that many these industries are integral parts of many local economies, which would make it in the interest of respective governments to follow suit.)

2) The facts can be so harsh that it can be difficult to internalise and accept reality.

Even still, many people do learn and internalise this—ever noticed the popularity of oat and almond milk in coffeeshops, even despite higher prices?—so I think it is not unreasonable to expect this in certain ML-based industries, either.

Re: AI in my plasma physics research didn’t go the way I expected

#298

Post by Flamentono2 »

lossolo wrote: post #162
Flamentono2 wrote: post #140

Earlier quoted context omitted.

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 h…

> Its a paradigma shift for the whole world literaly. That's hyperbolic. I use LLMs daily. They speed up tasks you'd normally use Google for and can extrapolate existing code into other languages. They boost productivity for professionals, but it's not like the discovery of the steam engine or electricity. > And what limitations are obvious? Tell me? We have not reached any real ceiling yet. Scaling parameters is the…

> ... but it's not like the discovery of the steam engine or electricity.

completly disagree. People might have googled before but the humancomputer interface was never in any way as accessable as it is now for a normal human being. Can i use Photoshop? yes but i learned it. My sisters played around with Dall-E and are now able to do simiiliar things.

It might feel boring to you that technology accessability drips down like this, but this changes a lot for a lot of people. The entry barrier to everything got a lot lower. It makes a huge difference to you as a human being if you have rich parents and good teachers or not. You had never the chance to just get help like this. Millions of kids struggle because they don't have parents they can ask certain questions required for understanding topics in school.

Steam Engine = fundamental for our scaling economy electricity = fundamental for liberating all of us from day time internet = interconnecting all of us LLM/ML/AI = liberating knowledge through accessability

> 'There hasn’t been a real breakthrough in over two years.' DeepSeek alone was a real breakthrough.

But let me ask an LLM about this:

- Mixture of Experts (MoE) scaling

- Long-context handling

- Multimodal capabilities

- Tool use & agentic reasoning

Funny enough your comment comes before claude 4.0 release (again increase in performance, etc.) and the Google IO.

We don't know if we found all 'low hanging fruits'. The meta paper about thinking in latent space came out in February. I would definitly call this a low hanging fruit.

We are limited, very hard, on infrastructure. Every experiement you want to try consumes a lot of it. If you look at the top x GPU AI clusters, we don't have that many on the planet. We have Google, Microsoft, Azure, Nvidia, Baidu, Tesla and xAI, Cerebras. Not that many researcher are able to just work on this.

Google has now its first Diffusion based Model active. 2025! We are so far away from testing out more and more approaches, architectures etc. And we are optimizing on every front. Cost, speed, precision etc.

Re: AI in my plasma physics research didn’t go the way I expected

#299

Post by Flamentono2 »

nancyminusone wrote: post #193
Flamentono2 wrote: post #140

Earlier quoted context omitted.

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 h…

LLMs are great at tasks that involve written language. If your task does not involve written language, they suck. That's the main limitation. No matter how hard you push, AI is not a 'do everything machine' which is how it's being hyped.

Written language is very powerful apparently. After all LLM can generate SVG, python code to use Blender etc.

One demo i saw with LLM and code use: "Generate a small snake game" and because the author still had the Blender MCP tool connection, the LLM decided to generate 3D assets through Blender for that game.

Re: AI in my plasma physics research didn’t go the way I expected

#300

Post by flobosg »

boxed wrote: post #292
flobosg wrote: post #251

Earlier quoted context omitted.

> that probably can't handle truly novel proteins AlphaFold is able to predict novel folds, see https://www.nature.com/articles/s42003-022-03357-1

Could once is not the same as can predictably first of all. Second, how can they possibly know this fold isn't actually in the ENTIRE PDB? I doubt very much that the can. The PDB is enormous.

> Could once is not the same as can predictably first of all

I was merely addressing your claim in the previous post.

> Second, how can they possibly know this fold isn't actually in the ENTIRE PDB? I doubt very much that the can. The PDB is enormous.

There are well-established fold classification databases (such as SCOP and CATH) where you can query newly solved structures using several structural comparison algorithms (DALI, TM-align, etc).

The Protein Data Bank might be enormous, but there is an enormous amount of structural redundancy as well, from which reduced datasets can be (and are) derived. Protein structure is, after all, much more conserved than sequence, mainly due to the physicochemical principles that govern folding and stability.

Post reply on HN