EvgeniyZh

> In the Copenhagen interpretation, when I look at schrodingers cat, the wavefunction will collapse. After I have looked at it, no way to go back.

In the Copenhagen interpretation, there is a difference between measuring a system and entangling with it. There exists a unitary that entangles you with the cat, just like there exists a unitary that entangles the cat with the quantum device. By definition, unitary evolution is not a measurement in the Copenhagen interpretation.

> Showing collapse happened is really tricky because of decoherence. Similar to unmixing coffee and milk.

Unmixing would show the lack of collapse (in the specific case), though? To show collapse you'd need somehow experimentally prove that it is impossible to unmix. And to falsify collapse, you'd need to somehow experimentally prove that it is always possible to unmix. And if we go with your analogy, would you say the coffee with milk hasn't forgotten its initial state because it is technically possible to unmix them? Being able to manipulate the whole universe's wavefunction to unentangle particular subsystems is akin to having access to Maxwell's daemon, since decoherence increases entropy.

> This means that if you entangle a state with a large system, in theory you could ‘disentangle’ it.

This can be done in any other interpretation? Any unitary evolution on an isolated system can be reversed. This is a mathematical statement.

> Many worlds is falsifiable, if you can show wave function collapse actually happens

So, what experiments do you think would falsify it? How can you show that collapse actually happened?

> You don’t like what the physics is saying so for emotional reasons you’ve come up with a completely new and novel framework to discount what the simplest set of equations is saying.

Nah, physics is not saying anything. I do physics because I enjoy discovering and modeling new phenomena. If someone wants to do philosophy and ask questions like "but what is __really__ happening when we measure an atom", it's his business, but I don't believe we'd be doing the same thing, because these questions are impossible to answer with empirical evidence and don't help us model reality in any way. I don't see any difference between QM interpretations; one of them might be "true"

Many-worlds is an interpretation and not a theory. It makes no additional predictions beyond what QM does. Believing in any interpretation (or shutting up and calculating) is purely emotional; it has no scientific content.

h4 push was AlphaZero/Leela thing wasn't it?

He claims 80% ytd profits, and there are speculations that Citadel got their whole public portfolio, so the remaining 10-15b are all private (and marked in unknown way) including ~5b in Anthropic, those who invested early in the year probably still in green (assuming there is liquidity for other private investments, which probably there is given it's AI stuff)

You don't have to sample uniformly. You could take the lowest index of all maxima. But yeah, the main source of randomness is non-deterministic matmul, and temperature does nothing with it

> in the end, everyone needs housing, every single house would be occupied just as now, but instead of having renters being squeezed by supply and demand you would have owners.

Not every house is occupied now. Not everyone will sell at loss if they can't rent out. So more homeless people?

What about people who don't have any savings living paycheck to paycheck? There is clearly some floor on housing price and not everyone can afford housing even at zero margin. Giving out free mortgages like it's 2007?

What about new housing? With lower prices supply will further decrease, and in places with high prices it's usually already insufficient.

I don't see how they can keep their margins with all the pressure from Chinise models. It got to be race to the bottom on margins.

They (together with OpenAI and maybe Google) can have better margins on frontier models, but the demand on those got to be much lower

Yep, the scoring function is just piece value difference, so it can only detect checkmate in 0 (i.e., when king capture is available).

What I'm saying is that it can't be qualitatively different.

You can't guarantee every IC they will become a manager (which a professor essentially is) in any area and academia is no different. Larger number of students is what allows both the scientific progress and better filtering of potential professors.

I did none of my degrees in US, and my physics degree was 95% math and physics. Physics degree is quite sequential anyway. You can't do QM in your first year or QFT in your second year.

I've checked random people I know from oxford and none started after 3-year undergrad and those how did after 4 all did 4 year dphil (small sample size warning). 4+4 is reasonable.

Maybe this is CS-specific? Finishing physics PhD from high school in 6 years sounds just not enough time. Even exceptional people I know in my field needed at least 7-8 (3+4 or 3+2+3). 3 years into theoretical physics grad school is around the time people start doing decent research

There can't be enough roles unless we either grow academia indefinitely or reduce grad school spots to like 1 per tenured faculty

With 2 years master's before for the same total?

Statistically speaking "murderer is black" is a sensible assumption in US [1], but I'd prefer it wouldn't be made

[1] https://ucr.fbi.gov/crime-in-the-u.s/2019/crime-in-the-u.s.-...

Tensor core performance is inversely proportional to precision across all generations (i.e., reducing precision by a factor of 2 increases OPS by a factor of 2). 8-bit precision will give you the same improvement ratio. A100/H100 didn't support 4-bit if I remember correctly.

So FP4/INT4 will likely improve the same 30% OPS/W. You could get a separate improvement by reducing precision, but going 1-bit for 4x improvement feels unlikely for now.

There are two generations and 4.5 years between A100 and B200.

A100 has 312 TFLOPS of FP16 for 250W, i.e., 1.25 TFLOPS/W.

B200 has 2250 TFLOPS of FP16 compute for 1000W, i.e., 2.25 TFLOPS/W.

This is ~34% growth per generation and ~14% per year. It's hard to believe it will be 400% per generation this time

> Do we have an example of a real quantum computer doing some kind of a computation that is not easily accessible by the regular computer?

Simulations of condensed matter simulations performed on QCs (google's OTOCs, quantinuum's HUbbard model) are not easily accessible by the regular computer. There are people working hard on simulating these results classically so it's quite likely they'll be simulated eventually. We're at point where classical computers are still in the race thanks to immense scale and algorithmic progress, but I think it won't be the case soon.

> something useful in real life?

usefulness is subjective. There are results that are potentially interesting to some people on Earth (as opposed to RCS).

Thanks for the answer (and what seems to be unflagging the comment). Having some experience moderating (of course, much smaller) communities I understand it's impossible to keep everyone satisfied.

I, of course, can't judge the intent or the effort. What I can say is that I read all captions of 150+ votes submission, rarely skipping any, and I saw 20+ pro-Palestine ones and zero pro-Israeli ones. I think this is quite objective measure.

At some point I thought it might be intentional but now I think it is just bias amplification: these submission are flagged too fast and upvoted too slow to get anywhere.

Yes they are, just like the comment you answered to will.

Can you remember any pro-Israeli posts you turned flags off for since the October 7 attack?

Yes it can [1].

https://docs.kernel.org/kbuild/llvm.html

It's worth noting that this is "compute-bound optimal", i.e., given fixed compute, the optimal choice is 20:1.

Under Chinchilla model the larger model always performs better than the small one if trained on the same amount of data. I'm not sure if it is true empirically, and probably 1-10B is a good guess for how large the model trained on 80B tokens should be.

Similarly, the small models continue to improve beyond 20:1 ratio, and current models are trained on much more data. You could train a better performing model using the same compute, but it would be larger which is not always desirable.

> your car, TV

yeah I hope I won't ever be shown ads on TV for which I already paid

Was I supposed to? My comments makes it quite clear that I don't have information to do that.

The author of the question appears to have some information I don't (unless he made it up of course), so I asked him to share it

How many? What are the top 3 countries scammers flee to?

Asic for matmul is systolic array more or less

> fidelity [across all qubits involved in the circuit]

I don't see a scenario in which the fidelity of 2QG between two far away qubits matter. Stress tests should be somehow related to the real tasks the system is intended to solve.

In case of quantum computers, the tasks are either NISQ circuits or fault-tolerant computation, and in both cases you can run them just fine without applying 2QG between far-away qubits that translate in large amount of swaps.

If you're interested in applying Haar-random unitaries, then surely QV is an amazing metric, and then systems with all-to-all connectivity is your best shot (coincidentally, Quantiniuum keeps publishing their quantum volume results). It's just not that interesting of a task.

Quantum volume is a good metric but that's kind of one-dimensional take. Almost any interesting circuit doesn't requires all-to-all connectivity and superconducting QC are bad at all-to-all connected circuit so we can have interesting NISQ experiments without particularly large QV