> We’re all used to two types of magnet. The common one, the fridge magnet, is ferromagnetic — its atomic magnets all point the same way (up or down), adding their magnetic effects. The less well known one, the antiferromagnet (AF), has neighbouring atomic magnets that point opposite ways and exactly cancel out magnetically.

This is a very bizarre introduction. People encounter diamagnets (e.g., copper) and paramagnets (e.g., aluminum) way more than they encounter antiferromagnets. I don't know why you'd ever cast magnetism as a false binary between ferromagnets and antiferromagnets, without acknowledging any other types of magnetic order.

(I did a PhD in magnetic materials)

Edit: I'll add that whether an antiferromagnet is useful, say, for exchange biasing a ferromagnetic thin film, depends on many factors. Just looking at antiferromagnetism alone you've got collinear vs non-collinear, G-type vs A-type vs C-type, commensurate vs incommensurate, and isotropic vs anisotropic; and all of that interacts with the interface structure, yada yada yada. It would be helpful if the authors elaborated on the expected properties of these materials. I personally don't know what people want room-temperature magnetic semiconductors for, but I'd be curious to learn what set of properties they think would be useful.

After the LK-99 debacle, I'm taking this with a truck load of salt.

Last night, my Fable 5.1 cluster of agents discovered cold fusion techniques. All you need is ordinary iron or stainless steel pot to contain the plasma and I am barely at 13% of the weekly limit of my 200 pro plan.

Amazing times.

The title has been editorialized.

Actual title:

Two Room-Temperature Antiferromagnetic Semiconductor Candidates

There's nothing unusual about finding room temperature semiconductors. I assume whoever posted it misread this as room temperature superconductors, but it has nothing to do with that.

What's interesting here is the antiferromagnetic part of the title, which was removed. I think this makes it relevant for e.g. RAM, but not superconducting. Someone can correct me if I'm wrong.

In a way, you can think of pretty much anything we express with language, especially things that are already modeled in scientific language, or logical language, or in equations or code; to be representable in a parametric/searchable space

Thus, you can build ai/ml models+agents to explore those spaces, at a speed and scope much larger than what any human can do

I can imagine findings like these are going to keep increasing in frequency to a point in which the bar for novelty goes a lot higher

I am not sure how this process looks like. When they "discover" these, what are they actually doing?

    The agents ran quantum-mechanical simulations of each crystal with the standard method for this, density functional theory, at two levels of approximation: a faster one (PBE+U) and a slower, usually more accurate one (HSE06). The band gaps and spin windows below come from the more accurate one.

Making very basic high school physics mistakes while discovering new superconductors.

Except they didn't discover new superconductors. Their AI came up with a novel idea to discover new superconductors, and they didn't even bother to check whether the results it hallucinated checked out.

The distinction between "having an idea" and publishing a paper where you demonstrate that the idea has merit, where you validate it, is one you shouldn't have to really explain to a high school student, much less a grad student publishing their first paper.

Okay? Aren't the semiconductors we use today room temperature? I certainly don't use helium to cool my phone.

I don't see any claims that this is better than the current silicon and gallium arsenide semiconductors that we use. And the use of "room temperature" seems a deliberate attempt to misconstrue this with superconductors

A lot of these ‘an agent invented’ or ‘an agent solved’ are actually the agent wading through a lot of info and finding something a human did that no one noticed or saw the relevance of at the time.

If ai becomes so prolific that we humans all stop doing those things then will they still work?

Who is vals.ai and why they keep submitting eye-catching claims. A few weeks ago they said fable 5.1 solved some obscure cipher and now opus 5.5 found room temperature semiconductor candidates. Meanwhile they seem to be in the business of making benchmarks.

Are they a promoter / influencer for Anthropic?

To be fair, this existed in a 1999 paper. They just simulated that it worked as predicted.

Many more things will be like this. The massive amounts of 'genius' buried under corporate management and obscurity in the past 500 years will be a treasure trove.

The people who wrote this seem to be lacking in expertise, and its just a model benchmarking company..Whos every article is just hyperbole about llms.

Not sure why we're calling it a discovery, when they've literally been made before, by a human.

Interesting; but until actually made and tested, not worth getting excited over.

I don't care unless it actually works. The computer saying it SHOULD work is not interesting.

Sounds like a good reason to hire a lab to make some, and then make a big deal about it if the results pan out.

I can think of worse uses of VC AI funding.

When I did my PhD we had theorists come up with new candidates for high- & low-temperature superconductors all the time, that wasn't so difficult, fabricating the stuff is the hard part! You need to layer atom by atom using chemical vapor deposition or another technique, that's akin to alchemy, think of a hugely complex machine mounted in a temperature controlled room with a 30 ton concrete dampener below. That machine was hell to operate and ruined so many PhDs lives, sometimes they went years without ever producing a single working sample. I remember we used some high electron mobility amplifiers in the lab and at the time there was only a single lab in the entire world that could fabricate these because no one else could figure it out.

So, while this is great and I think LLMs will accelerate materials science the main bottleneck isn't having enough promising candidate materials, I think we have a backlog of at least a few hundred candidates that are worth pursuing. Maybe some money that goes to data centers would better go into CVD machines.

I wouldn't describe them both as being newly-discovered. The second one, KV[Cr(CN)₆], had already been discovered.

It read a paper and then a person who barely understands what a spin is working with a berkeley ai metric nonprofit did a blogpost.

It didn't discover anything. This is how cooked people are.

While we should be skeptical until made in a lab or verified by others, this is a much better use of LLMs than solving math theorems/conjectures

Finding some new combination or iteration in the literature and running DFT is the kind of thing a senior undergraduate or first year grad student typically does (and typically with Claude anyway these days). (And yes, they'd probably use Quantum Espresso to start, like this writeup and its agent does). They'd probably show it at a weekly lab meeting where it would get ripped apart. And they would not be blasting a preliminary calculation around the world as if they'd made a new discovery.. but hey, we're in a brave new world; maybe they should!

Sounds interesting. Excited to see physical versions of this cooked up. Also, very excited for a world a few years from now where we can talk about accomplishments like this from the frame of the driver of the AI, rather than hype that AI helped.

Ok, I have no idea about Navier-Stokes or those famous math problems which LLMs helped solve or even solved solo, but I do have a rudimentary knowledge about semiconductors and their history in particular. Even without opening the link, I can bet 1$ to anyone that they did not in fact discover anything which will result in room temperature semiconductors.

PS: edit - I was thinking about room-temperature superconductors. My mistake.

Astounded at how many people didn't check this and just splitted about it online to help with their opinions on the matter.

What percentage of these comments did any form of validation of the findings before diminishing the author for AI use and discarding the findings as though they had invalidated the results?

Your normal heuristics like word choice cannot help you validate or invalidate a superconductor.

This should probably read: "Researchers discover two room-temperature magnetic semiconductor candidates. They used Opus 5.5 agents to perform some checks."

> In the spirit of transparency, I invite the reader to go through all the computations that produced the above predictions, including the calculations behind the candidate designs

"In the spirit of transparency I invite you to ask your own LLM to verify what my LLM did"

I really wish headlines would stop using words like "discover" and "found" when they really should use words like "says" and "reported" because an LLM was involved. IMHO anything produced by an LLM should be treated like something said by a cable news host.

This reads like neither the prompter or the AI know enough about magnetism to claim they found a new candidate for a magnetic semi.

I've got a friend who has been doing this research since the 90s. There is real money involved in this. This isn't like a math proof with a 1mm dollar payout. I seriously doubt this discovery. Until they show it working, I call bullshit. A room-temperature semiconductor is worth WAY more than an AI company.

"candidate" is doing a lot of work here.

Miss leading title; “Opus 5.5 CLAIMED to have discovered two room-temperature magnetic semiconductor candidates”

If they are confident in their discovery they should pay some scientists to start making this and testing it out.

I just pasted the url in 6.1 sol xhigh and it said it works.

correction: decades of human research is stolen and by happenstance sampled by users of LLM Opus 5.5

The operative word is candidates.

I’ve been burned before on this topic

vals.ai has great marketing! Unclear if the product is great yet. Or what it is exactly.

if you read it fast, it almost made my heart beat real fast

I'm a little late to the show, but wanted to leave a note here explaining why this article about spintronics in semiconductors may potentially relate to excitonics and twistronics in superconductors.

Traditional superconductivity is limited by the masses of atoms vibrating in the lattice, which sets an upper temperature threshold because thermal noise acts like a phase change that returns the superconductor to being an ordinary conductor or insulator above its critical temperature.

Electrons are fermions with half-integer spin, meaning that their total quantum wave function must be antisymmetric. Swapping any two electrons flips the sign of their wave function, meaning that they can't overlap with the same spin, by the Pauli exclusion principle. But two electrons with opposite spin form a spin-singlet state where their spatial wave functions can overlap. Which is why two electrons with opposite spin can be in the same atomic orbital forever.

For larger quantum wave functions in molecules, electrons occupy larger molecular orbitals forming covalent bonds, and in conductors form what's known as a Fermi sea. If many electrons spontaneously align their spins via the exchange force which arises from exclusion and electrostatic repulsion (the same effect which causes Fermi holes, conceptually analogous to electron vacancies in semiconductors), their magnetic moments become large enough to be observed at the macro-scale as ferromagnetism.

At low temperatures, electrons have restricted momentum states at the sharp edge of the Fermi sea, so by the Heisenberg uncertainty principle, their spatial wave function uncertainty (coherence length) expands, causing them to overlap. Below the critical superconducting temperature, they pair up to become composite bosons. Bosons have integer spin and symmetric wave functions, so they aren't bound by the exclusion principle and undergo Bose-Einstein condensation.

In Bardeen-Cooper-Schrieffer (BCS) theory, these composite bosons are known as Cooper pairs. They share a wide coherence length and act as a unified macroscopic quantum state that flows around atomic nuclei without scattering, even in the presence of atomic vibrations (phonons) or imperfections in the lattice. In effect, this makes the superconductor a macro-scale orbital, similarly to how a permanent magnet is macro-scale spin.

BCS theory explains superconductivity from the bottom up like an engineering approximation. More foundational frameworks like Ginzburg-Landau (GL) theory use spontaneous symmetry breaking to explain it from the top down.

Using GL theory and going back to first principles, instead of avoiding phonons/vibrations to raise the critical temperature, we can manipulate lightweight electron clouds. To do this, we use electron-hole pairs (excitons) in semiconductors to dynamically polarize the electron clouds. This electronic framework leads to orbital fluctuation theory and excitonic superconductivity.

It was recently discovered that graphene layers twisted at a 1.1-degree magic angle create superconductivity. The overlapping atomic lattices form a Moiré superlattice that slows electrons down in places to restrict their momentum and expand their spatial uncertainty. This mimics the effects of low temperatures spatially instead of thermally, and is known as twistronic superconductivity.

Note that I wrote this summary by hand after doing a deep dive with AI, which also helped me edit it. While it's notable that AI can explore large problem spaces and find solutions independently, I'm at least as fascinated that it can now help us tackle complex topics in an approachable manner, which should lead to new insights and hopefully inspire future researchers.

This is nothing new. This is just Claude reinventing things.

Suuuurrreeeee

Quoting README.md:

> A designed web version with the same text and diagrams is in docs/index.html; turn on GitHub Pages for the /docs folder to serve it.

Yeah, the author did not even reads the slop Claude produces.

From the article:

> Candidate 1: Designed a Luttinger Compensated Magnet, YBaMnFeO₅

> Candidate 2: Identified a Luttinger Compensated Magnet in KV[Cr(CN)₆] from 1999

> KV[Cr(CN)₆] belongs to the same family as Prussian blue, the 300-year-old pigment.

An actual validation would be complex; so let's try and see what parts of this argument are grounded and feasible?

/? Prussian blue antiferromagnetic: https://scholar.google.com/scholar?q=Prussian+blue+antiferro... :

- a number of articles confirming antiferromagnetic effects

/? Luttinger-compensated Prussian blue:

- "Luttinger-compensated bipolarized magnetic semiconductor" (2025) https://journals.aps.org/prb/abstract/10.1103/9syc-71w8 .. "[2502.18136] Luttinger compensated bipolarized magnetic semiconductor" https://arxiv.org/abs/2502.18136 :

> The Luttinger compensated magnetism not only has the zero total magnetic moment as the antiferromagnetism, but also has the -wave spin splitting as the ferromagnetism, thus our work not only provides theoretical guidance for searching Luttinger compensated magnetic materials with distinctive properties, but also provides a material basis for the application in spintronic devices.

/? Luttinger-compensated : https://www.google.com/search?q=Luttinger-compensated

We could model this as a logical proof that's checkable also in lieu of doing actual work to confirm or reject the (AI) hypothesis, but first let's reason about the feasibility:

Are the described effects real?

Are there reported, reputable similar findings in similar materials?

So, at least the blue one could really work. Like it's 1999.

---

Without even validating the argument, what else do we think we know about this problem?

Did the authors know that there is a laser way to laser program the normally random domains of a magnet or antiferromagnet?

But is this going to be lower-cost and more sustainable than carbon-based room-temperature semiconductor computing, and does anyone know whether that will work yet (with ABC stacking in trilayer and pentalayer rhombohedral graphene) either?

I guess we can follow up later by searching for citations that reference this article that does not have DOI (which are free from Zenodo and FigShare).

Have we sufficiently reasoned or inferred whether the study is repeatable and reproducible?

At least we didn't inappropriately reject the hypothesis without experimentation or evidence

Discovered in whose data?

I could have gotten this in one prompt lmao

Here we go again

Lmao, anything goes

[flagged]

[flagged]

[flagged]

[flagged]

The most encouraging part here is that they published full calculations, code, and caveats rather than just a press release. DFT (especially PBE+U, and even HSE06) is notorious for getting band gaps and magnetic ordering energies wrong, and the YBaMnFeO5 candidate requiring perfect Mn/Fe checkerboard ordering is a huge synthesis ask — disorder could kill the compensated state entirely. Same with the 420K -> 490K calibrated Neel temp: calibration against a known magnet helps but doesn't remove systematic error. That said, as a screening workflow this is exactly how agents should be used: fast search over composition/structure space with a cheap objective function, then human-readable artifacts others can reproduce and falsify. The real test isn't the simulation, it's whether a lab tries to synthesize these and reports back, negative result included.

[dead]

[dead]

[dead]

[flagged]

Anyone remember LK99 lol

Ugh. Unless this has been actually experimentally verified to be a room-temperature and room-pressure superconductor, it's about as ground breaking as "Yet another promising nuclear fusion candidate theoretically described."

Current frontier LLMs empower effectively anyone with limitless knowledge. Historically, if I wanted to hire an engineer to, say, create something like this I would have needed a multi-million dollar budget. Now, anyone with $200 (or less) can achieve it.