upvote
>I wonder if we’ll have room temp superconductors and fusion this time next year.

No, because to advance in a real science you need to conduct real experiments. And it's a lot slower than multiplying matrices on a million GPUs.

reply
If multiplying matrices on a million GPUs reveals a unique lattice structure that in simulation enables room temp superconductors, the real experiments would immediately begin.

Look at LK-99, it felt like a global effort was inspired at the time to verify room temperature superconductivity.

It might not even require a specific lattice structure to be identified - there is still quite a bit we don't understand about regular superconductors, a large breakthrough could materialize RTS's within a year. Producing them at scale is a separate issue.

reply
Not my area of expertise but I do work in scientific computing. I know that we can do simulations like this for meta materials, where to some extent you just have a really large search space that is not bridged by logic but by brute force, possibly aided by feedback from physical experiments. Assuming we have similar for semiconductors, it’s not clear to me where AI clears the compute bottleneck.

We already have supercomputers to run simulations like this, and can already spend millions of dollars on compute to answer hard problems. AI may help improve experimental design but this is something humans can already do reasonably well. I’m less optimistic we can derive similar results as we have for math, but I’d love to be wrong.

reply
Math is easier because it's text and symbols only. In physics experiments you have non-textual input and this is not the LLM stuff for now.
reply
LLMs are already multimodal, though? Plenty support native image inputs for instance.

But also: human physicists mostly act on symbols, too. They take measurements using sensors, which produce a bunch of numbers, that they then tease theories out of. This seems in LLMs' wheelhouse.

reply
This is what people seem to be missing when they extrapolate all LLM mathematical brute force discoveries to any X or Y scientific problem. Most real-world stuff is not as cut and dry as formal logic and math. Computer science is applied mathematics and almost all logic-based, it follows that LLMs would be good at this. Lot harder to model proteins or materials science this way.

That said the stock price juicing counts on you seeing these discoveries and then attempting to extrapolate bespoke frontier models' success in math (which is dubious in terms of methodology and tailored training anyways) and immediately jumping to the conclusion that AGI and huge scientific advances are right around the corner and can have us all living in fully automated luxury Yudkowsky MIRI effective altruist utopia. It's a farce.

And yes, someone is going to comment on this by saying "moving the goalposts, in 2022 models couldn't even do this." Yes, it's incredible how much throwing money at certain logically achievable outcomes helps achieve those outcomes. It is not clear how it helps achieve random wish casting by odd cult-like CEOs, however.

reply
but what if AI notices a higher order pattern that we humans with our measly cognition doesn't?
reply
I’m sure LLMs can and will help with physics research. I’m much less sure you can frame most physics problems in a way that will succumb to 10,000 agents running in parallel.
reply
This optimistic forecast is based on an assumption that we already know everything necessary to find a room temperature superconductor, so an LLM just needs to connect the dots.
reply
We already have a wealth of collected data from superconducting and fusion experiments. If agent swarms can be 10x or 100x better at interpreting this data and using it to build simulations than humans are, maybe we don’t need that many more experiments after all. And the ones we do need can be much more precisely aimed at the remaining gaps than human-designed experiments.
reply
You can get only so far by analyzing existing data. The real power of AI has been always unlocked (chess, programming, maths) by letting AI propose solutions and get feedback. For natural science that means conducting experiments, whether physical or computational.
reply
We're limited on materials and other technology which will need a lot more time for fundamental maths and physics to diffuse to. But yeah, the future is bright.
reply
Not with a sufficiently advanced simulation.
reply
An LLM is not a physical simulation of anything.
reply
I never claimed as such. It would be acting /with/ and in the confines of a simulator.
reply
The sufficiently advanced simulation is QCD and I believe the state of the art is a few thousand particles running on a top 10 supercomputer.

Particles meaning electrons, protons, and neutrons.

reply
There is every reason to think that AIs will be able to make dramatic improvements the efficiency of numerical quantum simulation codes.

Also, for superconductors, QED is sufficient, QCD is used to simulate hadrons, not electrons.

reply
It's worse than that, it's like asking if we'll have perpetual-motion next year.

Just because you can search faster for something, it doesn't mean that that something ever existed in the first place.

I think a lot of LLM stuff is just speed-running alchemy. Instead of wasting a lifetime, computers can now waste thousands of lifetimes.

reply
If you think people are sour on datacenters now, just imagine how pissed they're gonna be about AI training facilities that occasionally explode or leak novel kinds of life into the neighborhood.
reply
> fusion this time next year.

Aren't most of the remaining problems related to nuclear fusion engineering ones? Like building reactor walls that survive neutron damage.

reply
It's a problem of scale and engineering. ITER is already way too big and complex, to the point that it may never be completed. They might turn it on, find a leak and be stuck with a huge paperweight. To have viable power generation you have to go even bigger.
reply
Nah. The ITER project was prev-gen by the time they began to build it. The project underwent some upgrades, but it's not at the cutting edge. High-temp superconductors and plasma stabilization algorithms allow to make smaller reactors.
reply
High Tc superconductors can make tokamaks smaller, but they're still much larger than fission reactors of the same thermal power output. While ITER has 400x lower thermal power density than a PWR, ARC is just 40x worse.

Basically there's no hope even high Tc superconducting tokamaks can be competitive sources of energy.

You might say "but make the magnetic field even larger", but most of the mass of ARC is the steel structure to resist JxB forces, so much further increase in B is not practical. There are also minimum sizes for fusion reactors set by the need for sufficiently thick blankets to stop the neutrons; that's independent of magnetic field.

reply
Solar power plants have even larger footprint per kW, it doesn't stop anyone.
reply
That's apples-to-oranges, since PV omits other components, and because the high redundancy and modularity of PV makes construction much easier. DT fusion vs. fission, the balance of plant are similar to each other.

(If you meant concentrating solar thermal plants, those are also uneconomical.)

Decades of detailed engineering studies of DT fusion reactors (like ARIES) have had them coming in more expensive than fission, as one should expect.

reply
That is the narrative of the fusion start ups pursuising it. But, show me the plasma. Oh right, they don't have any. Might as well just be investment scams for now.
reply
Or have stronger magnets, see sparc and arc from commonwealth fusion. Or a different method entirely, see helion.
reply
Why can't AI solve engineering problems?
reply
It potentially can, but you still need to build the thing in the physical world and test it under realistic conditions. Maybe I should have focused my post more around "materials engineering".
reply
Same reason AI can't replace the engine in your car yet.

Engineering problems are solved by massive amount of logistics: finding and hiring the right people to be in the right place at the right time with the right tools and raw materials that need to be purchased at a physical location and delivered by cargo ships and trucks driven by truckers across the country. People get injured, sick, go on vacations. Trucks crash sometimes. Cargo ships get delayed. Tools break and need replacement. Materials turn out to be wrong quality or quantity.

This is an entirely different world from software development or math where you can try different solutions and know if they worked or not in a matter of seconds.

reply
Plasma control is still unsolved
reply
engineering problems are often just math problems.
reply
the next YC call for proposals can include LEAN for building reactor walls that survive neutron damage
reply
There is also the problem of how to stop bitcoin miners from draining the oceans dry
reply
I do wonder how well a lot of this translates into physical sciences. It's one thing to be able to RL on lean but physical sciences don't have such an obvious quick feedback loop. A lot of the scientific results have been closer to the theory side which isn't unexpected from the sheer mathematical capability.
reply
if you look at the current "high" temperature superconductors it becomes clear that a peculiar atomic arrangement is the key yet we don't have the ability to individually place atoms (yet) so we must rely on entropy to place them into said arrangement for us which restricts the space we can even test.

so the task for an LLM here is doubly difficult and the feedback loop is of very high latency. I do wonder if someone is trying though. simulate physical process and entropy/annealing -> simulate superconductivity. and very few working examples to work with.

reply
So a couple of purely informational question coming from a layperson:

1. Is there a good theory of high temperature super conducting that we can easily simulate? My furthest understanding of this was that BCS was insufficient.

2. Does such a theory impose limits on the temperature that are achievable?

3. Do we know if such a theory is complete in some sense? (e.g. if such a theory predicted no super conducting material at a temp K and pressure P, we would know that there could be no such material at temp K and pressure P based on reasonable assumptions)

reply
There is no such generally accepted theory.
reply
We will not. Physics experiments take time and money to complete.
reply
DARPA/DoD have been collaborating with the labs for years, with an infinite budget and access to compute, it's bound to happen.
reply
DARPA and DoD budget are far from infinite. Their combined R&D budgets are a fraction of what the AI labs are spending.
reply
They definitely don't have a strong lead, trading places with OpenAI and occasionally Google every few weeks

A coordinated pause needs to come at the national level

reply
International level. National would be a great start though.
reply
We're in a cold war
reply
> I wonder if we’ll have room temp superconductors and fusion this time next year.

Still waiting for the AI advancements which are not just taking away work from SWEs and mathematicians..

reply
> Still waiting for the AI advancements

At this point this line of "yeah but what about..." is getting silly. We've had 2 millennium prize problems fall this year, plus a lean proof for FLT (another effort worth 1m$) and people are still "yeah, but what'd be cool is if..."

Dunno, at this point I feel like people are trolling. This is absolutely insane pace, and just last year we were hearing how "there is no more data", "models are stalling" and so on. The negativity and goalpost moving is getting tiring.

reply
I think it is because we see these headlines, but as a layperson it has no impact on my life.

I don't see the value and I don't see the world changed.... yet

reply
It will come, don't worry about that.
reply
> I think it is because we see these headlines, but as a layperson it has no impact on my life. I don't see the value and I don't see the world changed.... yet

You could try to be happy for others who see the value.

reply
As a scientist the speed of developments has been breathtaking. As a layperson there has been 0, if not negative, change in my life because of AI.
reply
I remember reading about the counter example to the Jacobi conjecture in July. That was cool, but many said it was an exception, because AI is only good at finding counter examples. I believed that at the time. But four months later we have lean verified GRH.. I don't feel like it's possible to make any predictions anymore. This might be the peak and in four months AI is more or less the same, only cheaper and with more compute, or we might see more insane unexpected advances. But one thing is for sure, if AI can formalize and prove GRH there are huge swaths of math just waiting to be "cleaned up" and completely formalized. It's just a question of spending the tokens.
reply
Well, you know, I think it would be in fact actually cool if it solved, say fusion or climate change. Not just solve math problems without providing understanding or take all joy out of software jobs.

I'm not saying the technology isn't amazing. It's completely surreal how amazing it is. I just wish it did better things.

reply
Actually agree, it's dancing around the edges, until real world serious problems are being solved, I do think there is some level of theater going on, it's impressive for sure but even as a SWE, I still have my job for now and really if anything AI has just given me more work to do, sure that might just be limited, but I can only go off my real world experience.
reply
Surely faster than light travel is worth more than a million dollars.

(Yes, yes, I know it's Fermat's Last Theorem)

reply
[flagged]
reply
Why? Even as a SWE there's plenty of software that I want in my own life that I simply don't have the time/attention to make, or even to try to commission someone else to make (e.g. I've been making little educational games/demos for my kids). Having something that does what I want in a few seconds when I have the attention is fantastic.

"Taking away work from SWEs" means enabling everyone else (or even SWEs) to do things they couldn't do before. This is obviously an improvement.

reply
I have a todo list long enough I don't even have time to prompt for all of it...

If I could justify enough tokens, I'd point an agent at my harddrive and tell it to formalise all of my old notes and goals, and spin up agents to build all of the things I've made notes about that I would like to exist. It'd still likely take years...

I just kicked off Kimi building a game based on a concept I had kicking around. I wouldn't have time to build the game or even write the spec, but I had time to type "write a game based on [paste of the concept]" to see what falls out of it.

I can't imagine ever running out of things I want, because every thing I finish tends to result in my todo list growing with new things.

reply
They are already there. Sure you are not seeing positions disappear but hiring is starting to grind to a halt in a lot of places. Someone who might have needed an admin assistant no longer needs one and middle management in large orgs whose job was basically summarize and pass up will soon start to thin out through attrition.

You wont see whole sale replacement but you will see doing more with less.

reply
I truly don't understand how someone can be interested in technology and visit HN without understanding - even broadly - the history of technological progress and how these things take time.

Yeah, I want the cure for cancer yesterday as well, who doesn't? This is the most significant result in mathematics in over a century.

reply
They do understand. They’re afraid, and that fear leads to hatred of AI and the big labs, which leads to this kind of motivated reasoning that looks bewilderingly foolish from the outside.

People are scared.

reply
Thanks for the free psychoanalysis, but you're completely off.

To be clear, I'm just disappointed and sad that AI mainly takes our jobs (or at least the satisfying part of it) and solves mathematics without increasing our understanding. You know, the part that's the main point.

However, if in exchange it solved, say, climate change or fusion, that I would happily accept. No sign of that so far.

reply
You contradict yourself - you are sad AI will take your job but completely fail to understand why people are scared?

Maybe you are a well off rich kid or just cashed enough from faangs to be set for life, whatever. Is it really that challenging to muster some empathy to those who, upon seeing their profession disappear with 0 replacement in sight, are scared of the future since they don't know how they will even feed their kids, not even going into higher expectations of life?

At the end some math understanding is secondary at best when facing existential threats.

reply
I think you either completely misunderstood me, or replied to a wrong comment.
reply
> No sign of that so far.

Trend lines are a thing and progress does not happen overnight

reply
Sure, and I'll be happy if/when it happens. So far still waiting.
reply
We have a genie that grants unlimited wishes for intellectual output[1]. If you had witnessed someone multiplying bread and fish would your first reaction be to shed a tear for all the bakers and fishermen?

[1] Not to get too religious about it, limitations still exist

reply
Does it? Can I wish for it to solve climate change for us?

Also, your analogue is off. If you had a bread and fish machine that made bakers and fishermen unemployed, while only producing bread and fish for the top 1%, I would not be very happy about it.

reply
Climate change is not a challenging intellectual problem. It is challenging to convince others to do what is necessary to solve it. Could AI produce potential solutions? Sure, from memory. We already have them.

In anticipation of a followup, "can I wish for it to solve climate change advocacy for us?" It can provide research direction on nearly any topic but as soon as we are talking about actions in the real-world it requires more than purely intellectual output.

Anthropic and OpenAI make their models available to anyone for use. I call up cornucopias of bread and fish daily and you have the power to do this too.

reply
Oh but climate change has tons and tons of very much intellectual sub-problems. Better battery technology is a huge one. How to optimally generate or build renewables. Even wild ideas like carbon capture, say through genetically modified trees that grow faster. That's what I'd like to see from AI, something that humans can come up with and solves a real-world problem.
reply
AI is working on problems in all of these fields every day. Yes it can help. No it is no panacea.
reply
Yes, and that's why I originally wrote:

> Still waiting for the AI advancements which are not just taking away work from SWEs and mathematicians..

reply
Climate change is not a challenging intellectual problem.

100% wrong, if a 95% efficient solar panel could be produced really cheaply, it would be solved. If fusion power could be created, it would be solved.

reply
>I wish for it to solve climate change

[laughing_greta.gif] they still do that? :DD

reply
Don't worry. It will take work from other career paths, too. Let's just celebrate the advancement in human civilization altogether on new year eve.
reply
> I wonder if we’ll have room temp superconductors and fusion this time next year.

No, we don't get useful things.

reply
That's a depressing world view.

I don't know about room temp superconductors, but fusion has gotten a bunch of investment recently that make it seem like it'll happen before I die.

Specifically, REBCO superconductors are in production by Commonwealth Fusion Systems to make SPARC, targeting net fusion gain for 2027 with a tokamak design.

The NIF has been able to achieve ignition, repeatedly.

Whatever you want to call it, AI/ML/GPUs; point is, supercomputers today are insanely powerful for running plasma control simulations and software.

Commercial contracts for fusion are in place.

More interestingly, Marathon Fusion has figured out a way to generate money instead of energy which needs to get sold off at energy rates. Fusion moves subatomic particles around. Instead of generating power with it, they're going to use the technology to transmute elements. Like alchemists of yore. Just, like, with science and it actually working. They estimate they can make money from converting mercury into gold, but more lucrative is radium into actinium, and zinc into copper. Specific isotopes of those are used in the medical industry to help fight cancer and are very expensive.

That's more of an economic technology than a technical one, but money is always the issue, and if they're able to make money with their scheme, that money can be used to fund further fusion research.

Maybe not next year, but some time before I die, fusion power seems entirely possible!

reply
You really shouldn't have gone out of your way to reply to that snide remark of mine, but thank you. Most of what you said went way over my head, but I'm glad people smarter than me are seeing hope.
reply
> I wonder if we’ll have room temp superconductors and fusion this time next year.

That’s child’s play. I wonder if we’ll have star trek replicator by next year.

reply
> star trek replicator

That's a walk in the park. I wonder if we'll be sipping life juice in comfy matrix cells by next year.

reply
I’m using my replicator to conjure life juice to sip out of thin air. Thanks anthropic.
reply
> I wonder if we’ll have room temp superconductors and fusion this time next year.

Tell me you've not worked in hardware engineering without telling me you've not worked in hardware engineering

reply
I’m a robotics engineer with a mechE degree. Agents are massively speeding up my hardware engineering work, yes the latency is still much higher than software, but each iteration is far more effective when you have agents to help. You can simulate whatever you’re building in far more detail than was possible before.

Yesterday I built a whole new system for agent-human PCB co-design (skidl+kicad). Maybe you don’t feel it yet but massive hardware speedups of the same OOM as swe and math are very clearly on the horizon now. I mean right now on the front page there’s a new agent swarm designed CPU claiming massive performance improvements.

reply
mean right now on the front page there’s a new agent swarm designed CPU claiming massive performance improvements.

Key word is "claiming" and there are plenty of comments on that thread pointing out that the supposed new design is actually just a very old design that was disgarded for practical reasons that are still not addressed and appear to have been hand-waved away.

reply
I work in hardware engineering and just today after setting up the oscilloscope probes for Claude, it was able to trigger my fault scenario and read the captures to debug a flaky comms link, and write a small reproducer, all done autonomously while I worked on something else.

So you might want to check yourself since it's already helping massively in hardware.

reply
Imagine flipping the switch on a fusion reaction that nobody knows how it works, but Clause says it will, and has 1 billion lines of code that supposedly proves it's not an H-bomb.
reply
So dramatic
reply
>I wonder if we’ll have room temp superconductors

>we

Luckily, this matter does not pertain to the permanent underclass at all.

reply