algon33

Yeah, that would make this post a lot better.

If you could somehow track the frequency spectrum of usage, then you could mostly handle that problem. "Oh, the frequency spectrum strongly peaks at about once a week? I guess users regularly need to use this feature. Probably important somehow." vs "Huh, the frequency spectrum isn't strongly peaked anywhere, but is substantially lower at high frequencies. We can probably remove this."

Kaku is a media sensationalist; that class of person has a terrible track record. Futurists, that is, people like Aasimov, have an ok track record at predicting the future. Interestingly, early futurist predictions about digital tech were pretty good. For "physical" goods? Not so much. A plausible explanation of this discrepancy is that they didn't predict the breakdown in the historial trend of increasing energy usage. See the Henry Adams curve.

"What I would suggest is to start taking the extra time to optimize your resume for every job you apply for. There are tools online to help you find out how these softwares scan and pick out pieces of your resume."

Is there one tool to rule them all, or a host for each specialty?

Huh. How's that fit with the general push against remote work I've been hearing about?

place an odalisque with an inscription on it Besides being cruel, odalisques aren't known to exist for more than a hundred or so years. Let alone millennia.

It's been strange these last 23 days.

This is the most insane thing I've read about this year.

Problem sets are useful for improving your mathematical ability, as is looking at an expert's attempt at a problem after your attempt. Both are a core part of reading books well. So when someone asks for books to improve their mathematical ability, a likely interpretation is just "they want books with good problem sets, clear presentation and elegance". Another interpretation is "what's a book which, when after I intrepret all the individual sentences into a vague impression, will make me good at maths?"

Your answer is somewhat helpful in the latter sort of world where OP didn't know how to read a textbook (which is unfortunately common) but not in the former sort. You seem to be venting though, which is understandable. But venting with a side of helpful content would be even better.

For instance, advising OP on how to to read a maths book (generate content yourself, check dependancies, connect things to what you know etc.) or suggest books which contain advise like this alongside their main content (I think Tao's analysis texts might do this?)

There are a couple of people interejecting with answers to questions, or asking questions. I'm afraid I don't have a better estiamte than that. But in this case, I think lumping the students together as one speaker and the teacher as another would be fine.

I tried using this for a technical talk[1], and it got the amount of speakers wrong. Which is somewhat suprising to me, as I would have thought diarization tech would just worked by now.

[1]https://www.youtube.com/watch?v=5lFxURxbyEc&list=PLiayR7yJx8...

The FAQ contains re-states the content for point 14 in point 13. Point 14 is about why your code might be slower when using 2.0. 13 should be about how to keep up with PT 2.0 developments. Someone should change that.

By database do you mean a finite 2d array with column headings? Or something that can be represented by such an object? And its entries are elements of some countable sets? If so, what are you doing using ZFC there?

Personally, I'd guess that I wouldn't consider it "stealing" an identity if it were evenly trained off a couple of individuals. But it is a weird case ethically. Can I just train a model off of my favourite bloggers' works and tweets, and release a tweetbot replying to stuff I think would be topical? I've basically done that, but not hooked it up to Twitter, and I think that might be fine as long as the model is private. But this is certainly thorny ground.

Nice, a friend was looking for something like this.

Location: London Remote: Yes Willing to relocate: Yes Technologies: Python (Pytorch, Pandas, Numpy, Tkinter), Mathematica, Lisp, LaTeX Email: dbz_ali@hotmail.com CV/resume: https://github.com/Algon-33/Algon-33/blob/main/CV_Programmin... Github page:https://github.com/Algon-33/Algon-33/tree/main

I have some experience implementing deeprl models, statistical physics simulations, pre/post deep learning revolution AI algorithms and a couple of simple games. My degree was in theoretical physics. My interests are broad, so I'd be fine learning new stuff for a job, or working as a front end, or back end , dev, or doing ML research or designing high performance numerical simulations for SDEs or whatever.

Because it has Omicron's increased ability to avoid people's immunity. That would result in higher infectivity than just a copy of the wuhan strain, with near the same severity. Note that earlier Sars viruses had even higher severity than Sars Covid 2 (Wuhan strain), but weren't as infective and killed far fewer people. Which implies that a slight reduction in severity for a gain in infectivity is not a worthwhile tradeoff.

So will anyone be deleting their patreon accounts? Can we even be sure our data will be wiped from their servers?

I read a good book review[1] that moved me towards the view that figuring out how a living thing works is possible, though not necessarily easy. I highly recommend reading it, but here's a summary.

Evolution promotes fitness enhancing functionality, so we should expect biological processes to be useful for some purpose and hence not be distributed like a random graph (e.g. Erdos Reyni graphs). Indeed, if we look at biological structures, we can find that the causal networks they form are far from random. Furthermore, there are often repeated motifs present. These motifs are quite simple and seem to map neatly onto human understandable concepts (like XOR gates or autoregulators or feedforward networks etc.)

And often, the overall graphs seem like they're tree like rather than some complicated mess of feedback loops (barring autoregulation). This kind of structure is quite modular, and hence we can leverage our understanding of component parts to understand greater and greater pieces of the organism.

There are two problems with this arguement: one, that a lot of the data used for it is not nearly exhaustive. Maybe the people examining biological circuitry stumbled on the rare areas where there are repeated sub-components. Second, even if there are repeated sub components, why should we get modularity i.e. few connections, mostly local?

The former may not be an issue if there hasn't been a lot of dedicated effort towards finding human comprehensible structure in biological circuits, which there might not have been. These things are big and complicated, with many constituent parts, and teasing out the underlying structure may require loads of computation and statistical analysis, which was hard for most of the history of biology.

The latter is not adressed in the book review, or in the comments, but the review author's work makes me it plausible to me that modularity will be common in biological systems. I don't have a good summary of that, or can clearly articulate why I'm hopeful about this. But read the rest of the work of the writer of the article if you're interested in this kind of stuff (key words: natural abstractions, interfaces, selection theorems).

[1] https://www.lesswrong.com/posts/bNXdnRTpSXk9p4zmi/book-revie...

Perhaps principled fits better?

Gunpowder and galleys is a book on naval history that is highly rated and highly readable. It was written for historians, and changed the fields understanding of why galleys eventually lost out to sailing ships in naval warfare.

------------------------------------------ The Rise and Fall of the Third Reich is a massive tome written by an American reporter who lived through the eponymous era. It has a great deal of detail which I had never heard of before, and shaped my understanding of WW2. It is also quite moralising, maintaining that Nazi Germany was inevitable because of a servile German character, that the depravity of the SS was in part due to their homosexual deviancy, and a little too focused on the personalities of individuals. Admittedly, the last part is quite interesting. Why do I still recommend it? Because I think there is enough depth there, and enough of a distinction in the text between the author's theories and the facts, that you can learn a great deal from it.

For instance, the sheer stupid chance that led to him waging such terrible war. How so many men could have prevented his rise to power. He might have died a soldier in the first world war, or the French or English could have crushed his open violations of the peace treaties, or how his takeover of multiple nations was facilitated by the petty racism of British diplomats. There's also detail concering the poltical, economic and social factors that allowed this chance to exist in the first place. Like the disconnect between the German military and the Weimar Republic who they were sworn to protect, or the war-weariness of Western Europe which was partially responsible for their leaders to continually cede ground to a madman who publically outlined his plans to genocide entire peoples. ------------------------------------------ The 10,000 Year Explosion is not widely accepted by academics, but is not widely panned either. And whilst it does seem to over-play its thesis, the core idea is quite sound: evolution still applies to modern humans, and the rapid growth of the species over the past 10k years has allowed for a great deal more variation and selection to occur, and as such we should see far more changes in human biology now than throughout the average 10k period of history. The exact mutations described and the hypothesis for why they occured are, of course, more conentious. I don't have much of a general opinion, as the authors go through a wide variety of population differences like lactose intolerance, diease immunity, hearing adaptations due to language use and, yes, IQ. The depth and quality of research on these varies, so the reliability of conclusions in the book does too. ------------------------------------------ "The Dawn of Everything" covers some interesting anthropological data, but the reviews I've read paint a poor picture of the author's interpretation of what the data means. Whilst it points to an important idea, that the human transition to agriculture was not nearly as sharp or uniform as is popularly believed, I think you'd be better served by reading this page on Wikipedia: https://en.wikipedia.org/wiki/Behavioral_modernity and reading some of the cited papers and authors' works.

Didn't the author say that he's writing to push an agenda? Anyway, it is a fantastic blog.

If I remember rightly, the AlphaCode paper includes a list of benchmarks, including the results of a finetuned GPT-3 for coding. I think they did it because Codex wasn't available to them when were doing their tests, but I might be wrong there.

My statement was too strong. By "knew" I meant that it was pretty likely that strong cosmic censorship wouldn't work out [1]. In addition, I think QFT and GR are both probably broken in some way given that they're incompatible and that no one has coerced one of the theories into working with the other. That feels like evidence that the problem is not as simple as just rejecting QM or GR to me. But again, calling that "knowing GR is unphysical" was too strong a statement.

[1]https://en.wikipedia.org/wiki/Cosmic_censorship_hypothesis

https://physics.stackexchange.com/questions/403574/what-situ...

Non deterministic dynamics occurs in Newtonian physics as well. These things are worrying philosophically, but we can take some comfort in that the initial conditions for which these occur are "measure zero" i.e. if you were defining a probability measure over initial conditions, then the non-deterministic initial conditions would have measure zero. It doesn't sound like we have that option here though.

But in my view, this result is good news. We knew GR was unphysical anyhow. Now we've increased our understanding of how unphysical it is, potentially ruling out more replacement theories.

Any chance for a windows version? Or a mobile app? The usecase for a mobile app is more plausible, even given that this is already a great idea.

Approximation

Somehow I have never heard of Tsetlin machines before this. Are you talking about this https://ai.facebook.com/blog/the-first-high-performance-self... result by MetaAI?

For some reason I forgot to check metaculus for this. Thanks for the reminder.

How suprising did you guys find this? I'd have said there was a 20% chance of this performing at the median+level if I was asked to predict things beforehand.