Dr_Birdbrain

Are they all Latin-derived languages? Because that’s the Latin root :)

Watches can be used to find fresh water sources. (Joking, of course you are completely correct)

My father has a mechanical watch that he bought in the 70s and he has never serviced it.

A Jaeger LeCoultre Memovox, a thing of beauty.

An SVM is a quadratic program, which is convex. This means that they should always converge and they should always converge to the same global optimum, regardless of initialization, as long as they are feasible, I.e. as long as the two classes can be separated by an SVM.

Just want to be pedantic and point out there is no such thing as a Canadian that is a dual citizen with China. China does not allow dual citizenship.

Source: I lived in Mexico for a few years.

I don’t think it’s accurate to say Mexican governments are underfunded—they have oil money, from the nationalized oil industry, to the extent that they are extremely chilled out about tax collection.

It’s not that they are underfunded, but the criminal gangs (drugs, extortion, etc) are absurdly overfunded. Strong organized crime weakens and delegitimizes the government, which means that weaker, petty crime can also thrive. Like if you are sick with something big, small infections are more likely to opportunistically take advantage of your weak immune system.

Aside from organized crime, there is something else that is hard to articulate. It’s hard to explain exactly what it is, but to a large extent it feels like the population at large can’t agree who are the good guys and who are the bad guys, I.e. in many cases they will not unambiguously say that the drug cartels are bad and the government is good. They’re probably right about that, in many cases the two are inextricably enmeshed. It’s a vicious spiral that is hard to break, expectations of crime and corruption lead to more crime and corruption, but in many cases people won’t even agree that the gangsters are unambiguously bad. Sometimes the baddies get lionized to an unhelpful extent, and there are even ballads “narcocorridos” written about their exploits

The thing is, I see a huge difference between high-end health plans (for example, at places like Meta, which include IVF coverage) and middle-of-the-road plans (like my employer provides) and bare-minimum plans (like MediCal).

Which plan would we be talking about for government coverage? I assume barebones plans.

In that case, it still matters who you work for, because some employers still provide much better coverage.

I think this is more targeted at new graduate students.

I remember when I first started my PhD, literally all of the talks were completely over my head, and I did not understand enough to engage as you describe here :)

Gathering three key takeaways would already require a good deal of effort :) and, I appreciate that the author of this post absolves you of having to have the speaker’s message as a key takeaway, since as a beginner grad student that is frequently beyond reach.

I think in this context, grammatical correctness is irrelevant, and what you actually want is just raw text as written by real humans from all walks of life.

Ah yes, we’re about to get photographic and evidence of aliens right when widespread generative AI explodes, cannot possibly be a coincidence

Came here to say this. The code it generates for me is usually not perfect, but for me it provides conceptual insights that have gotten me unstuck from gnarly situations.

The conventional narrative is that chatGPT can code as well as a junior engineer, but I feel like it’s more like a senior engineer scribbling on a whiteboard—no, it’s not perfect code, but it’s the right idea.

I think parent comment was talking about Lemoine, who never claimed the AI was dangerous, but rather that it was sentient and not being given the rights it deserved.

I started my ML journey in TensorFlow, and now I am a happy PyTorch user, for many years.

TensorFlow, yuck

I liked the way Matt Levine put it: if you’re the sort of company that sometimes randomly finds large amounts of assets you had lost track of, you should be worried you might later find large amounts of liabilities you had lost track of.

One of the people interviewed made a flippant comment that the focus on STEM may have had a negative impact on abstract reasoning. Haha what? Sounds like that person is one of the people impacted by the reverse Flynn effect.

I have been a regular Anki user for many years.

My approach to making cards is actually to take long-form notes on the topic at hand, then apply close deletions to key parts, like equations, key ideas, and definitions.

This makes it so that the piece I need to actively recall is small(ish), but I still have the context

Interesting idea. If so, it was a risky gambit, and also they only get to do it once, because the next bank is bigger than SVB. Do they really care enough about lowering interest rates to play such a risky gambit, which can only be played once?

Also we are coming on the debt ceiling limit at some point in the next few months, which might trigger a crisis of its own. If I were a VC I wouldn’t play my one-time hyper-risky card before seeing how the debt ceiling shenanigans play out.

That’s funny, because I wouldn’t buy a Quest, but I was just thinking I would LOVE FaceTime in AR, so I could take calls and meetings while walking in the park. I already do that but I have to use a hand to hold my phone.

Great perspective. In many ways the current tech scene has become entrenched around a handful of big players. I am looking forward to hopefully a fresh batch of startups creating new dynamism.

Yes, but I think the folks who work on technical alignment would like to have stronger guarantees—I think they would like to end up with formal guarantees that the AI will only be helpful, and not harmful.

I spent a few years of my career in formal verification. As a result of that, I suspect that the desiderata of “technical alignment” research is most likely undecidable, and the existing RLHF and prompt engineering approaches are probably the best we can do.

A neural network is a large, nonlinear function. In many applications, such as those that release their source code and allow fine tuning, it is adaptive. Internally the weights evolve according to solutions of differential equations. All of these things smell of undecidability when trying to prove safety/alignment properties of them:

- The theory of differential equations (which governs the evolution of the weights during training and retraining) is undecidable - Nonlinear functions have decidable cases as long as we stick to polynomials, but even then the decision algorithms are doubly exponential—-much higher complexity than anything in NP-complete, so practically undecidable for very large functions - The neural network is a very large function

In summary, they are large, nonlinear, and adaptive, these three characteristics make the problem smell intractable and likely undecidable to me.

Well a lot of the compensation is not cash, but bonuses and stock grants. I know, stock isn’t cash and is subject to risk, but it’s a lot more than 190k even after you factor in the risk.

I am both an index investor and an AI enthusiast, and I think it’s an excellent analogy.

If you are an investor who does not have interest in maximizing your returns and you just want a place to park your savings, an index fund is great, and will return the market average return.

Similarly, if there is a skill that you need that is not your “special” skill, it makes sense to subscribe to the “market average” of that skill, as implemented by a statistical language model, which is in a sense “averaging” over the collective language skill of the entire internet.

In both cases, the ingredients of the “average” are formed by the people who try to do better. Active investors, including hedge funds and short sellers, think they can do better, and they provide signal that in turn feeds back into the index.

Similarly, anybody who thinks they are better at writing than the language model is free to try to outperform it, and eventually will feed back into the training data.

Was it confidential info? The public earnings call was three days ago, investors didn’t like what they saw and the stock went into free fall, then the run started yesterday.

I think insiders might not even have realized how bad their balance sheet looked. On the earnings call the CEO was talking about how much he likes cycling to de stress, etc

Matt Levine reports that the reason they had to buy those low-yield investments that plummeted is because it’s tech-sector customers had too much money in the boom times. Too much deposits means they need to buy a lot of something, and in the boom times that was low-yield stuff.

If you play moves that are common across low-rating games, the average completions will be low-level moves. If you play moves that are common in high-rated moves, it should play high rated moves.

In practice, it mostly plays gibberish that doesn’t follow the rules.

You might be interested to know that the status of the sovereignty of Ukraine was not in question at any point right up until the very second that Putin said the Kyiv government was a Nazi government, and invaded. Russia itself was a signatory to the accord that guaranteed the territorial integrity of Ukraine, yet here we are.

I think this functionality is not available on the app, only on the browser version. The browser version runs great on mobile, though.

I used to believe this, but then I saw the demo of ChatGPT attempting to play chess.

The rules of chess are much simpler than any structure of the real world we might hope for it to understand—and yet it failed to learn the rules of chess.

Based on this, I am pretty sure it has not learned any meaningful structure.