rafaelero

That's why we have Lmsys for. And spoiler alert, it performs better than 3.5 no matter the complexity of the question.

Does this guy really believe that GPT-4's answers are of the same quality as GPT-3.5? That's ridiculous.

VQVAE is trained to reconstruct the image, so in theory it should contain all the information (both content and location) inside its embeddings.

They are very likely using VQVAE to create a dictionary of tokens and then just converting images into them with an encoder.

Still, if you use GPT-4 it gives you 60% of accuracy in predicting if it's going to go up and down, which is considerably better than median human forecasters. Stop being so dismissive and start reading the numbers.

Are you telling me you think it's not realistic that the sum of all US institutions will own >50% of the tokens? I think that's quite likely.

> How would "the ecosystem" prevent this?

Yeah, not sure it was possible given the incentives.

> How do you expect this to happen?

Institutions will stake their tokens and have disproportionate voice in the consensus. Whenever the US government tell them to do something fishy, they will oblige and no one can do anything. That's not surprising, though: this deficiency in PoS has been well known for a long time.

Such a dumb move for the eth ecosystem. Contrary to BTC, concentration of tokens inside one institution is a direct threat to their project. Why the hell would they want an ETF?

> Like no one, even the anti-AI writers like OP even seem to want to interrogate this point.

People have tried that a lot in 2020 soon after GPT-3/Dall-E 2 was released. It's a losing game. The value is very clear and gigantic.

What a pathetic move. Feel bad for the state of the US.

Isn't S&P 500 basically inflation + GDP growth at this point? I think BTC will follow similar pattern given enough adoption.

This is called self-consistency: sample multiple times and then select the answer that appears the most. You can read Alphacode's paper, which uses a similar method. I think we can do even better, though. Over 1000 runs, it's likely that the correct answer appears at least once. Instead of selecting based on the majority, we could use the LLM to judge the answer individually. Since it's easier to verify than to generate, I think this method would work very well. I don't know if any lab has tried this, though.

I can't wait for the 400b to be released. GPT-4 is too expensive and the fact that we can distribute the workload between different companies (one company trains it, another creates a performant API) means we will get a much cheaper product.

It will certainly decrease. Also, there are multiple ways to deal with hallucinations. You can sample GPT-4 not once, but 10, 100, 1000 times. The chances of it hallucinating the same things asymptotically reaches 0. It all depends on how much money you are willing to invest in getting the right opinion, which in the field of medicine, can be quite a lot.

I am just extrapolating on a trend that started in 2018. I think I am well justified in my opinion, but feel free to ignore the trend and listen to substack fools.

Unless he has access to GPT-5 his opinion is worthless.

I am very bullish for GPT-5. It's going to surpass even the best diagnosticians we have right now. And we will all have access for $20 a month. Crazy and exciting times.

> I find to be absurd

It's not intuitive, but it's what decades of behavioral genetics studies say. Adoptees have much stronger correlations with their biological parents than with the parents that adopted them (in all socially meaningful metrics: intelligence, income, etc.). Monozygotic twins correlate much stronger than dizygotic twins, etc.

See for example the classic association between childhood maltreatment and future antisocial behavior [1]. As intuitive as it may seem that a child that is maltreated may develop negative externalizing behavior because of that, it looks like the true route of transmission is genetics, not environmental.

[1] https://www.cambridge.org/core/journals/psychological-medici...

What's missing here -- and in most of social sciences -- is the realization that adverse events is itself a product of genetics, and bad social outcomes are only weakly mediated through those events. Genetics is most of the story here and, although it's a depressing narrative, I'm sick of seeing people push a narrative that is not based on facts.

I agree. But what does it have to do with your rant about women particularly getting a worse deal?

Computers got 99% cheaper, so by your analogy of only considering an arbitrary product, we have actually experienced deflation!

That's what equality means and what they fought for.

What about gold and Bitcoin?

Nothing you said has anything to do with the article and its proposal to incorporate interest rates into inflation. Maybe it's your type of post that should be considered polution.

> Transformers are just pattern matching.

That's trivially true. The question is: are we any different?

A lot of organizations still think they should have their own [finetuned] model to provide a custom experience to their users, so that may come as a surprise for them.

Improbable since reducing added fats is still as good advice today as it was 5 decades ago.

Ok, then let's start by getting away with all the wasteful animal farming.

What a ridiculous complaint. Energy efficiency won't remain static, and even if it were, it's not up to you to decide how to best leverage the available electricity.