jackson1372

I don't have sources, it's just a rumor.

It's an open secret that Mistral finetunes on GPT outputs

claude.ai

The article doesn't mention two great features on iOS: - Move cursor by tap-and-hold on soace bar - tap-tap for word selection and tap-tap-tap for sentence selection.

I use these two constantly

Claude Instant 1.1 is a better comparison for the price/performance of GPT-3.5

GRE writing: 5 for Claude vs 4 for GPT-4

Bar exam: 76.5 for Claude vs 75.7 for GPT-4

See the pricing PDF[^1] and API docs[^2], but TL;DR:

- Price per token doesn't change compared to regular models

- Existing api users have access now by setting the `model` param to "claude-v1-100k" or "claude-instant-v1-100k"

- New customers can join waitlist at anthropic.com/product

[1] https://cdn2.assets-servd.host/anthropic-website/production/... [2]: https://console.anthropic.com/docs/api/reference#parameters

ANTHROPIC ・ Frontend/UX Engineer ・ SF-based ・ Remote-friendly ・ https://www.anthropic.com

Rapidly build prototype interfaces for large language models. Align AI with human values.

Tech stack: Next.js, Svelte, Python

Apply: https://jobs.lever.co/Anthropic/be5f1be0-e0c8-4a43-934f-cea1...

Isn't the explanation for this that the world actually does work in some way or other and that it's not just infinite chaos and so if you keep throwing parameters at some problem, you will eventually stumble upon the "real" structure, but that's no guarantee of when that occurs, and with which parameters?

The reason you want to over-parameterize your model is that it protects you from "bad bounce" learning trajectories. You effectively spread out your overfitting risk until it's pretty close to 0.

Or at least that's the way I like to think of it.

The next step is to better compress the resulting model in a simpler, less computationally costly network.

If you read between the lines, it's clear that Spotify was able to make a normal watch app. But they wanted to make one that had special functionality not yet allowed by Apple, for any app, not just Spotify.

But what does 'corresponds more closely to the mathematics' mean? The relevant point is just that the mathematics, alone, doesn't settle the theoretical question.

In general, it's hard to use Occam's razor in a non-question-begging way.

All three interpretations are 'just the wave function'. What separates them is the metaphysical interpretation of what is 'behind' the wave function.

Precisely how to understand causation is controversial.

But I'll just say that most philosophers don't find anything incoherent in the idea of nonlocal causation. (Hume's classic attack on local causation might be worth a look.) See BoiledCabbages comment for a nice explanation of a model that involves nonlocal causation.

we have to assume some sort of isomorphism between the map (our theory) and the territory (the world)

But a mathematical model != a theoretical model. The very same mathematical model will be compatible with uncountably many theoretical models. (By 'theoretical model' I mean something like 'an interpretation of what the math is representing'.)So you can't read off theoretical structure from mathematical structure. And so you can't read off the structure of the world from mathematical structure.

if you already accept "superposition" at the level of quantum states, you don't have to invoke anything new to deal with so-called "collapse"

All three interpretations of QM 'accept' superposition, construed as a mathematical construct. What's at issue is how this mathematical construct maps onto reality.

If you mean something more by 'accepting superposition', you have to spell out what that is. And doing that just leads you right back to the three different interpretations.

The standard story about measurement/observation and quantum collapse is not the only interpretation of the available data.

There are, broadly, three kinds of interpretations of what's 'really' going on behind the scenes:

1. Collapse - The world exists in an indeterminate state until observation occurs, when the world collapses into one a single determinate state. The probabilities of quantum mechanics map onto the the different parts of the unobserved indeterminate state.

2. Many Worlds - Quantum phenomena cause the world the branch into multiple worlds. The probabilities of quantum mechanics represent the 'share' of reality that branches in each direction.

3. Hidden variables - The probabilities of quantum mechanics are artifacts of our inability to know all the relevant variables. Measurement necessarily involves causal contact, and causal contact will always disturb some of the relevant variables in unpredictable ways. It would be cool if we could observe what goes on when measurement occurs, but to do that would require measurement! So we're stuck with hidden variables.

The public tends to hear the collapse interpretation most often. Physicists tend to like the many worlds interpretations. Philosophers of science tend to like the hidden variables interpretation, because the other options require an incoherent metaphysics. (I'm a Philosophy PhD student.)

People say that the hidden variables interpretation is ruled out on experimental grounds, but this is demonstrably false. Experimental data shows that hidden variables, if they exist, violate locality:

Locality - Causal interaction is a local phenomena. No action at a distance.

So long as you're willing to abandon locality, hidden variables can work. Given that the other two interpretations posit equally weird things, abandoning locality won't seem so weird.

For more on this, see Tim Maudlin's excellent paper, "Three Measurement Problems"

https://www.academia.edu/32885328/Three_measurement_problems

Philosophy PhD Student

Nissan has been working on this:

https://www.wired.com/2017/01/nissans-self-driving-teleopera...

I'm no expert, but I think this is a "known bug". Russian oligarchs buying property in London, for instance, is a big reason for that market's insane prices.

One 'solution' is property taxes, which disincentivizes buying property that you leave unoccupied. Planet Money had a good episode on the property tax and how it's a surprisingly effective tax.

1. I wouldn't put it in terms of time being "more fundamental than physics". We can mean two things by 'physics': the thing that is the object of study in the Physics department or the theory that gets generated by that study. Time, I take it, is one aspect of the object of study, one aspect of the natural world. So the question is: does the natural world, at the most fundamental level, have asymmetric temporal order? Or, rather, is asymmetric temporal order mearly an 'illusion' that gets explained away once you have the most fundamental picture of the world?

That's likely still unclear, but that's to be expected. We typically can't ask questions about fundamental structure without invoking that fundamental structure itself. That's why I find the analogy to left/right so helpful. We have a pretty good idea of what it means for left/right orientation not to be built into the fundamental structure of nature. And so that can provide a way of testing various claims about the fundamentality of asymmetrical temporal ordering.

2. I'm not sure I have a grip on what this additional thing is supposed to add. But I'll note that you reference 'history' and one might reasonably think that that is a temporal notion. If you meant 'history' to mean something like "the universe's extension in the temporal direction", that doesn't quite seem to allow for the distinction I took you to be trying to make.

(And, yes, McTaggert's framing is highly unhelpful.)

I should disclose: I'm an A-theory-supporting philosopher. (Sigh. Why are philosophers and physicists always disagreeing?! We should be friends!)

My above analogy to the left/right issue is the example I use to introduce students to the A- vs. B-theory.

The A- vs. B-theory debate is, of course, complicated, and you point to one possible complication: can we separate the issue of temporal asymmetry from the issue of temporal change?

As I see it, the answer to this question is "No".

We can talk about how some physical object 'changes' its spatial properties as we move from one spatial slice to another. For instance, as we move from the pointy-end of a cone to its base, each circular slice of the cone 'changes': they keep getting bigger until you reach the base.

This use of 'change' is symmetrical. There's no preferred direction of 'change' here. It makes equal sense to speak of changes "from bigger circle-slices to smaller-circle slices" as it does the other way around.

But in the context of the temporal dimension, talk of 'change' is not symmetrical. It's fine to say "The President changed from Obama to Trump", but you can't say "The president changed from Trump to Obama".

The point of all this just is: the issue of whether (temporal) change is symmetrical is tied up with the issue of whether time itself is symmetrical. You can't have your cake and eat it too. There's either an objective temporal ordering of physical events from past to future (as A-theory posits) or there isn't (as B-theory posits). If you're a B-theorist, you have to say that "Trump came after Obama" is like saying "The cup is on the left".

This is not a refutation of B-theory. But it, I hope, makes clearer what the stakes are. And when those stakes are made clearer, B-theory, in my view, looks less attractive.

Consider an analogy to the 'reality' of right/left orientation. We stand on opposite sides of a table, and I say "There's a cup on the right" and you say "There's a cup on the left". Who's right?

Well, there's not really a deep disagreement because facts about left/right are observer-relative. Furthermore, we might reasonably conclude that left/right orientation isn't part of the fundamental structure of the world.

Now, if I looked at a timeline of events and said "Trump became president after Obama" and you (or any other observer from any other point in time) said "Trump became president before Obama", who is right?

Barbour is saying that this disagreement is like the left/right dispute. Maudlin is saying that it isn't.

I think it's fair to characterize this disagreement as being about whether time is really part of the fundamental structure of the world. And it's clear that there's a meaningful disagreement here.

I'm a philosophy Philosophy PhD student. There can be a number of different reasons to read the original text of some older work of philosophy. But two central reasons that seem to be ignored here are:

1. Philosophy is closer to logic/math than science. And the reason to learn 'what Aristotle said about X' is similar to the reason to learn 'what Pythagoras said about triangles'. The pythagorean theorem doesn't become outdated in the way that scientific knowledge becomes outdated. (The difference here is just between a priori and a posteriori domains of knowledge.)

2. Explaining philosophical reasoning is, itself, doing philosophy--it requires careful reasoning and argumentation. The central thing we do (in UC Berkeley Philosophy classes) is evaluate philosophical reasoning. And so you can't simply rely on some interpreter's take on the original text--you must grapple with it yourself. Sometimes interpreters do get things right, but sometime's they don't. And telling the difference requires doing some philosophy.

Understood in the right way, there's nothing wring with that comparison.

Certainly, people misuse/misunderstand the information-processing metaphor/claim. (Though the people who do this tend not to be cognitjve scientists.) But that's not to say that the metaphor is complete garbage.

My argument is not: there should never be clubs that are, in fact, single gender.

My argument is: gender itself shouldn't be used as a criterion for group membership.

I'm a PhD student who studies this stuff. This article is, quite simply, poorly argued.

Most importantly, there's nothing in here that directly argues against the information processing metaphor. The author gives a bad argument that he thinks lies behind the metaphor and shows why that argument is bad. But a) I don't recognize this argument as what motivates the metaphor and b) to show that one argument for a conclusion fails is not to show that the conclusion is false.

The author often points to differences between us and actual computers. No one who employs the information processing metaphor is claiming that we are identical to human-made computers. Rather, the claim is that there's some abstract sense of 'information processing' under which both human brains and computers are implementations. In order for both of those things to be implementations, they need not do so with the exact same kinds of mechanisms, nor must they have the exact same kinds of abilities. (This is why I find the discussion of the dollar drawing case so strange - no one is claiming that we store visual representations of objects in the way that an actual human-made computer does.)

When the author writes:

"To catch the ball, the player simply needs to keep moving in a way that keeps the ball in a constant visual relationship with respect to home plate and the surrounding scenery (technically, in a ‘linear optical trajectory’). This might sound complicated, but it is actually incredibly simple, and completely free of computations, representations and algorithms."

What he describes is not a non-algorithm. Insofar as it's a procedure that takes in certain inputs and follows a series of steps in order to achieve some goal - it just is an algorithm, albeit a somewhat simple one.

More importantly, the author seems to misunderstand what is meant when researchers say that the brain 'runs' an algorithm. Of course you don't consciously crunch numbers when your visual system is chugging away, trying to analyze and sort through the...data...it's getting from the retina. But the lack of conscious computation is not evidence for the complete lack of computation.

There are all sorts of legitimate criticisms of cognitive science - and the author discusses some of these at the end of the piece without really understanding/explaining what's going on behind them. But the author says nothing convincing that it's the failure of the information processing metaphor that explains the limitations/failures of cognitive science.

At Harvard, officially-recognized groups cannot use race/gender/sexual-orietation/religion as a qualification for membership. So, yes, a white person can join the Black Student Association. And yes, a male can join Women in Business.

Boy Scouts and Girl Scouts are allowed to exist. But I, myself, think their gender segregation is gross.

There are male and female sports teams for the same reason heavyweight and lightweight boxers compete separately: we recognize that there's a meaningful difference in type of bodies and so we segment them into different groups for competition. But not that the reason for separation, here, is not gender; it's types of bodies.

Author here.

Think about how all this would sound if you were talking about race instead of gender and see if you'd still agree. If you don't still agree, try to figure out why.

We find it gross to say things like "Whites and blacks have different interests and enjoy different kinds of things, so there's nothing wrong about separating them into different social institutions". I think it's still gross if you replace 'whites and blacks' with 'men and women'.

I drink Soylent 2.0 everyday. Definitely doesn't taste like glue. I has a very mild, but quite pleasant, vanilla-y taste. It kind of tastes like liquid cheerios.

It tastes good enough that I have cravings for it. But it's not so strong a flavor that I get sick of it.