lambdatronics

Fuel cells convert chemical -> electrical directly, batteries go both ways.

For nuclear fusion, the only practical reaction releases 80% of the energy as uncharged particles (neutrons), for which the only way to harness their energy is to convert it to heat. The 20% of the energy that is emitted as charged particles could in principle be converted with high efficiency, but the other 80% dominates.

For fission, more of the energy is emitted as charged particles, but it's not clear how one would harvest it directly in bulk. (It can be done on small scales, but not efficiently.)

It's been considered, for a variety of heat sources. I'm long-term bullish on a related technology (thermophotovoltaics), although presently the efficiencies are not great & the compatibility with nuclear radiation is probably poor.

Not the same kind of turbine. Internal combustion turbines are more cost-effective than steam turbines.

Depends on what kind of contribution you want to make & how much you want to learn, and what skills you already have.

If you wanted to be on the cutting edge as a fusion physicist, and let's imagine you are a bit rusty on calculus & physics, you'd probably be looking at >2yrs remedial math/physics classes at the undergrad level, plus about 6yrs of graduate work. If you're able to get into a big-name program, you could be doing something useful after the first couple years of the graduate program.

If you're a data analyst/engineering or programmer, you could probably get involved in a less direct way (supporting the science team) much faster.

Here's a pretty good list of (primarily US-based) jobs in the fusion sector, including both scientist & other roles. Have a look at the qualifications these roles require.

https://usfusionenergy.org/opportunities

What I liked least about Jira was how it turned a project into a maze full of pages of boilerplate, all alike. I was continually walking the issue/subissue tree via hyperlinks, and the pages were slow to load, and sometimes I went in circles looking for something that turned out to be over on a Confluence page instead. There's so much visual noise with all the optional tick boxes & fields (that were not used), I got banner blindness - which contributed to my difficulties in finding things.

My solution was to manually maintain a bulleted list of links directly to the relevant issues.

If we want the intelligence to be recognizably human-like, then yes. As to whether the AGI could be trained in a simulation vs needing to really be physically embodied: I used to believe not, but then I learned about sim2real. (Ex: [0]) Training blank-slate deep learning algorithms on real hardware is prohibitive. It requires millions of hours of experience. Those hours are really expensive on current-day hardware (also cannot be accelerated, stuck with wall clock time). But if pre-training in simulations can be effective, then I think we have a decent shot in the near-term.

Human-level and human-like are not necessarily the same though. I doubt human intelligence is as general as we like to think it is. There's probably a lot about what we consider intelligence that is domain-specific. Training AGI on unfamiliar domains could be super valuable because it would be easier to surpass human effectiveness there.

My pet conjecture, though, is this: the ability to experiment is key to developing AGI. Passively consuming data makes it much harder to establish cause and effect. We create and discard hypotheses about potential causal links based on observed coincidences (ie, small samples) all the time. Doing experiments to confirm/refute these hypotheses is much less computationally expensive than doing passive causal inference, which enables us to consider many more potential causal linkages. That allows us to lower the statistical threshold for considering a potential causal relationship, which makes it more likely for us to find the linkages that do exist. The benefit of developing accurate causal models is that they are much more compact & computationally-effecient than trying to model the entire universe with a single probability distribution.

[0] https://openai.com/blog/solving-rubiks-cube/

That's a really profound question that deserves serious research, IMO. Communication is easy enough to define using Shannon's theory of information. I would argue that computation is subset of communication: signals go in one end of a 'processor' & come out the other (so the processor acts as a 'channel' in communication theory), but the data comes out modified in some non-trivial way, which distinguishes computation/processing of information from mere replication of the signals at the output.

A trivial modification would be something like (X,Y) -> (Y,X), just re-ordering cables. A non-trivial modification would be something like (X,Y) -> (X XOR Y, X AND Y). Both of these 'channels' have 2 bits of capacity, but intuitively only the second one is actually doing any information processing/computing.

I haven't found a satisfactory way to formalize my intuition about what is 'non-trivial' and to quantify how much 'processing' is being done by a general system, in such a way that one could meaningfully compare a brain and a CPU, for example. I think it's a great research question.

Wait until you find out about enstrophy.

> Isn't "order" a metaphysical concept

Yes. The most scientific way of talking about "order" in that sense is the Kolmogorov complexity, which is still extremely poorly-defined. The best way to put it is that "ordered" states are ones that have low Kolmogorov complexity.

Yeah, you also have to transform the "reference" function, and then the entropy stays the same. I prefer to think of it as the "density of states" -- it's necessary to make the argument of the logarithm dimensionless, after all.

>how physicists get around the discrete and finite restriction

By turning a sum into an integral. The probability 'density' is p(x), and the 'density of states' is n(x), so then entropy is then integral of p(x)log(p(x)/n(x)) over dx.

>It seems to say that "given a system we know everything about, there is no way to go to a system that has some unknown things to us".

That's backwards: information is the negative of entropy. The 2nd law says that entropy never decreases, so information never increases (it can only be preserved or lost).

>I don't know what Entropy is supposed to mean on the level of individual states/configurations.

The entropy is a property of a probability distribution, not of a state. Entropy is defined as H = -sum(p_i log(p_i)). A 'state' implicitly defines a probability distribution: uniform probability over all the microstates compatible with the state description.[0] In the case of a microstate, the entropy of the probability distribution over microstates consistent with that state is zero - there's only one compatible state, so p_i = 0 for all other states, and log(p_i) = 0 for the compatible state. In the case of a macrostate, the entropy of the probability distribution over microstates consistent with the macrostate works out to -sum((1/N) log(1/N)) = log(N), where N is the number of consistent microstates. That's the Boltzmann entropy.

Sometimes people will write about the entropy of a 'state' in such a way that it sounds like they're talking about the entropy of a microstate -- but what they're probably talking about is "the entropy of the macrostate that this microstate belongs to." It's sloppy to talk like that, because "the" corresponding macrostate isn't unique. There are many sets of macrostates that could contain a microstate, depending on what properties of the microstates one considers 'macro.'

(Ex: 10100101 is a member of both "symmetric bit strings of length 8" and "bit strings of length 8 that average to 1/2". The entropy of "symmetric bit strings of length 8" is 4 bits, whereas the entropy of "bit strings of length 8 that average to 1/2" is ~6.1 bits. And of course, the entropy of "the bit string of length 8 that is exactly 10100101" is zero.)

Is that just a coincidence for N=2?

There are a broad range of motivations. In my case:

- I wanted to work on nuclear fusion energy, for the future of humanity.

- I was told that if I had a PhD in physics, I could do just about anything - it's a way of keeping options open.

- I liked tinkering in the lab & learning new mind-expanding concepts in textbooks.

- I viewed it as a test of my intelligence. (Turns out, it's more about perseverance.)

I know at least one string theorist, and they seem to be motivated primarily by liking to mess around with abstruse mathematics. Some others seem to enjoy the 'nerd cred.' I think the ones whose sole motivation is 'getting to the bottom of the universe' probably burn out early, b/c there's so little of that to be had right now. (I've heard from several people who got their PhDs in particle physics who went on to do data science/programming, saying that the field is depressing and that's why they didn't pursue it further.) As for me, I found the 'shut-up-and-calculate' attitude a major turn-off to studying quantum physics. (Plasma physics uses little or no QM, so that worked out for me.)

Fusion researcher here. Yes, this piece is sensationalized, as are most pieces about fusion, and about science/technology in general, unfortunately. It's depressing that tokamak performance hasn't improved much in all that time. The only two ways to scale up the gain are to build the device larger (like ITER) or with higher magnetic field (like SPARC). The existing devices have been around for many years and have been pushed basically as far as they can go, the only way forward is to build new devices. There's really no reason to expect a breakthrough on tokamak performance that will suddenly change the game. The only really positive news for tokamaks in the last ~30 years is the advent of REBCO superconductors & the success Commonwealth Fusion has had so far in capitalizing on this technology.

CFS claims that it's manageable, although I don't know enough to evaluate that claim. It's also not clear from this quote if the once/twice per year is referring to full replacement of the vacuum vessel, or maybe just to inspection or replacement of a subset of components.

"Bob Mumgaard, CEO of Commomwealth Fusion Systems, regards neutron flux as part of a fusion power plant’s wear-and-tear—a routine aspect of its servicing cycle, happening perhaps once or twice a year. “We can simplify the internal components, develop maintenance scenarios,” he says. “We have such a scheme substantially in place.”" https://nautil.us/issue/86/energy/the-road-less-traveled-to-...

Interesting. I wonder if personification is analogous to a "memory palace": it's a way to hijack our innate brain capacity for working on something more abstract. I've always felt like physics derivations were like mystery stories - the best ones had a surprise twist of reasoning that leads the detective to the solution from the scattered clues.

I like having static analysis for undefined names in my IDE for exactly this reason: I can write the program in the natural order (high level to low level), and then have a little reminder about what I haven't finished yet.

Cool idea! I'm pretty busy a.t.m. but just in case, I'll toss my hat out there.

About me:

-PhD in physics, specialty in optical spectroscopy for fusion reactor plasmas

-Familiar with basic optics, analog circuit design & fabrication, signal processing, sensor fusion, data analysis, visualization, physics-based modeling & parameter inference (including tomography), Bayesian methods, & MCMC

-Very comfortable with the scientific python stack (Numpy/Scipy/Matplotlib/xarray), some experience with Julia (esp. SciML/ModelingToolkit)

Projects I'd be interested in (mostly stuff I've been meaning to try my hand at but haven't gotten around to):

-Unsupervised/self-supervised learning on real-world data (incl. algorithmic trading)

-Machine vision (and/or analysis of other sensory modes) for robotics

-Brain-inspired (ie, systems-neuroscience-based) AGI approaches (ex. hierarchical temporal memory, predictive coding, sparse distributed representations)

-Information-theoretic approaches to AGI (ex: total correlation explanation, free energy methods)

-Causal/generative/physics-based approaches to AGI

-Renewable energy & clean tech: battery systems modeling, demand/weather forecasting, or hardware projects

-Off-the-wall physics ideas: I'd be willing to listen to them & provide feedback. I'm personally interested in trying to reformulate quantum mechanics starting from geometric/Clifford algebra and Bayesian probability theory, as a way to make it more intuitive.

Email is in my profile.

The famous Alan Kay appearance: https://news.ycombinator.com/item?id=11939851

This one? https://news.ycombinator.com/item?id=24537865

What's your take on the idea of on-orbit propellant gathering?[0] That is, the concept of a solar-electric propulsion sat in low orbit 80-120km, scooping up atmosphere, ejecting the nitrogen to compensate drag while storing the oxygen for a propellant depo. I tried to run some numbers on it and it seems like it might work, but maybe doesn't pay for itself fast enough to be worthwhile.

[0] http://hpepl.ae.gatech.edu/papers/ProgAerospace_Singh_V75_20...

Cool stuff! I'm curious if you considered Julia when choosing the language to work in, & if so, what were the pro's/con's vs Go.

Great idea, I was wondering if someone had started working on something like this. My thought was to use the identity for distributed social media (blogging platform where you can easily restrict viewership of your posts by granting revocable keys to your contacts).

Thanks! What kind of problems outside academia are you interested in?

Wow, the "Authoritarianism" section is the essay I wish I had written, but better than I would have written it! Thank you!

I'd like to think I've gotten better about focusing on a relevant goal. I'm working in industry now, although still doing physics R&D.

This seems like a pretty low-level detail that doesn't warrant being heralded as a "breakthrough." They apparently have a new, more efficient power electronics to supply electric current to the superconducting coils. The electronics are situated inside the cryostat, so they are also cooled to very low temperatures apparently (few degrees Kelvin). The article is really vague, though.

Great point. I clashed a bit with my advisor early on - they were trying to shepherd me away from a rabbithole, but I was a bit pigheaded and determined to fix something that I felt was sloppy. I ended up with a publication on it, but it probably wasn't a great use of my time (aside from providing me with a chance to flex my wings).