soVeryTired

Look to your own tortured words. Referred to a system as ‘performant’ recently?

It's a machine learning method for turning a sample of text into a fixed-length vector. We usually try to find embeddings with the property that samples of text with similar meanings are mapped to vectors that are close to one another.

Completely agree, and I think this is where a lot of the confusion in the thread is coming from. The prize claims to compress “human knowledge” but it’s actually a fairly rigid natural language text compression challenge.

I’d conjecture that the algorithms that do well (as measured by compression ratio) on this particular corpus will also do well on corpuses that contain minimal ‘knowledge’.

Compression algorithms that do well might encode a lot of data about syntax, word triple frequencies, and Wikipedia editorial style, but I really doubt they’ll encode much “knowledge”.

There’s no way I could reproduce a calculus textbook verbatim, but I can probably prove all the important theorems in it.

Even then, given half of any sentence in the book, I don’t rate my chances of reproducing the next half. That’s more a question of knowing the author’s style than knowing calculus itself.

Does arithmetic coding capture all of that?

I have it and strongly dislike it. “Undefined PTO” is a much better term than “Unlimited PTO”.

There is a limit, the company just delegates the definition of that limit to team leads. It makes planning holidays uncomfortable since it’s never quite clear whether you’re pushing a boundary.

Generally I just stick to the UK norm of 25-27 days plus public holidays.

A generous but fixed allowance that you’re actually expected to use in full would be a much better system.

Cubic splines are universal function approximators. As is piecewise linear interpolation. The universal function approximator thing is way overblown.

Is "eliminates" the right word here? My reading of the article abstract is that the treatment kills virus particles and in higher doses, the macaques showed increased white blood counts. But that's not to say that the monkeys are cured. Maybe "reduces viral load" might be a better phrasing?

I'm not sure I get what you're saying here. ChatGPT and GPT-4 absolutely can write code and poetry. GPT-4 arguably does both better than I do.

What's taught in undergrad economics courses is incorrect. Fractional reserve banking doesn't really exist anymore. Banks don't lend out the deposits they recieve, they create new ones 'out of thin air'.

Arguably the most important constraint on a bank's ability to create new loans (and hence new deposits) is its core tier one capital ratio: that is, its total liabilities divided by its shareholders' equity.

Can you explain what's going on there? Sounds like he's doing something to prevent patenting of drugs but I don't follow what it is.

> but this whole story is more of a parlor trick than anything else

I don't think it is really. You're right in saying that 2/3 and 3/2 are the right numbers to ensure that the population wealth stays the same. But the point of the exercise is to show that the population wealth can increase while the wealth of almost any individual in the population decreases.

This is the author's jumping-off point to argue that the standard arguments in economics about average welfare can be far removed from the experience of any single worker in the economy.

In the first regime, where we calculate at the population level, we stop and average the wealth of the whole population after every coin toss. So if E[X_t] is the expected sum of wins and losses from the coin toss,

W_t = exp(k E[X_t])

Whereas the wealth of any individual grows like

W_t = E[k exp(X_t)]

There's a result called Jensen's inequality that says that

f(E[X]) > E[f(X)]

for any random variable X and any convex function f (exp is one such function). In a sense, I think this all just falls out of Jensen's inequality.

I think it's the videos of levitating samples that people are getting excited about. At this point it's either outright fraud (unlikely), an interesting materials science discovery that isn't superconductivity (most likely IMO), or superconductivity itself (still unlikely, but the balance of evidence is shifting).

If it happens to be the latter, it's a huge scientific leap forward.

Subtle nitpick, but inflation favours fixed-rate debt holders. Floating rate debt holders can get burned pretty badly in an inflationary period.

I’m not going to argue that asset managers and trading desks have plenty of resources and that they can transact very quickly and cheaply. But having been on the inside of small and large asset managers for almost ten years, I can say there’s a lot of groupthink and rather brain dead behaviour to be seen on a trading floor.

Call me jaded but I’ve worked with both systematic and discretionary traders. The algos I’ve seen tend to be heavily overfit, and stop working as soon as they hit production. The discretionary traders usually have a tonne of gambler’s tics and have a bad habit of assigning narratives to market noise.

Most institutional traders aren’t the best in the universe. They just do dumb things faster and at bigger scale than day traders.

Have you tried asking GPT-4? I got a pretty good explanation of attantion from it.

The 'key' and 'value' description in the attention mechanism probably doesn't make much sense to you because at best it's a strained metaphor, rather than an actual principle by which attention operates. If you take a look at the maths, you can interchange the role of 'keys' and 'values' and get something mathematically identical.

The Drake equation? We can see the stars, we can see the planets (and there are lots of them). We can infer the number of earth-like planets. So where's all the intelligent life?

A manufacturer and fitter of wooden legs: “Ken’s knee-log in”

Just a guess: is this answer GPT output?

> The map is not the territory.

Turing would argue that if the map resembles the territory so faithfully that you can't distinguish between them, the difference is moot. We're not there yet, but we're a hell of a lot closer than we were ten years ago.

ChatGPT is most certainly AI: all of machine learning is a subset of the field of AI, as is much of symbolic logic and Bayesian inference. It's a massive and wide-ranging academic discipline.

Maybe you're thinking of "AGI", which is a term that appeared in the late 90's to make the distinction between the expert systems of the time and machines that could think like a human?

The model is mildly interesting but surely there’s actual data available out there that shows the variability?

Over the past decade more and more devs have been out-earning surgeons and white-shoe lawyers. The tech salaries of today feel like the i-banking salaries of the early 2000s. It’s not just the c-suite that’s been earning crazy money.

> I realize there's a 99.9% chance this will cause some emotions, so I won't post any further.

Or maybe go the whole hog and just don’t post? I’m not religious but questions like this can only be asked in very specific circumstances without causing offence.

This a thread trying to wish a significant segment of the hn community a happy festival. Is now really the time to question their beliefs?

Honestly is Demis that big a deal? I always figured it was the researchers a layer or two down (e.g. David Silver) who were doing the real work.

It wouldn’t look too dissimilar to a standard sinusoid, just shifted upwards a bit. In the range [-1, 1], e^x is reasonably similar to 1 + x.

Maybe you could provide us with a reference then?

As an externality, the price of emitting 1kg of carbon should be a little greater than the cost of capturing and storing 1kg of carbon.

That's too expensive a target to be realistic right now, so the real question is what's a level of taxation that won't cause social unrest, but that will change behaviour?

It's civil service. There's basically no negotiating on pay - they certainly won't pay an internal candidate more.

> They do not even reason, just make up words that fit some context - thus they "hallucinate".

But they can explain their 'reasoning' in a way that makes sense to humans a lot of the time. Serious question: how do you know if something does or doesn't reason?