Nesco

It feels more like a ad for getting people to invest in AI that anything else. Plus Eric Schmidt hasn’t got a lot right lately in my opinion

> “There real use-cases”

> Proceed to link to a non-constructive proof

Why are mathematicians like this?

LLMs are trained on the entire internet so loads of copyrighted data, which Meta can’t distribute, and is afraid to even reference

Especially because genAI is a copyright laundering system. You can train it on copyrighted material and none of the content generated with it are copyright-able, which is perfect for social apps

The training data contains most likely insane amounts of copyrighted material. That’s why virtually none of the “open models” come with their training data

Those lines generate insane losses just to operate and maintain them

America is way less dense than France and Europe at large. Only in BoWash such an high density network would make sense

Could you provide some resources/links please?

Nobody can predict the future, but what is currently researched: - coupling deep learning with some kind of memory

- non generative hierarchical architecture (see LeCun JEPA)

- mixing deep learning with symbolic methods (usually search over token or latent space, it can also be a coupling with symbolic engines. Coupling LLMs with wolfram was a very early example)

- ability to perform unbounded computation per token inside a deep network. Something like Universal Transformers

Thanks!

Hey, I would be interested in doing the same. How hard is it?

If it comes through rna/dna editing it will be a good evolution

I have been thinking about this topic for some time. It might be done using the energy of the token. If it's still higher than an energy limit, then process it again, and increase the energy limit. The energy could be computed using log-sum-exp: https://openreview.net/pdf?id=Hkxzx0NtDB

Could be very large, even close to 1, depending on how you react to it

I cannot upvote this enough. Having a bad trip is plainly one of the most horrible experiences of my life

I find it a huge red flag that the article tries to combine together several “woke” talking points. It often indicates that each points is a product of ideology more than anything else: - the strange, I am quite sure straight up false, self-referring point about African and Asian metallurgists of the past, which basically add nothing to the core message

- A strong “degrowth” orientation based on a reduction of the demand, because steel production emits CO2. It seems to be the core message. It doesn’t address the issue the author himself raise that most steel is used for machinery or buildings

- A digression which tries to find an additional reason to reduce steel production: reducing potential war output. If someone needs two reasons for something it usually smells fishy, plus I would love to hear what the Ukrainians and NATO members have to say about reducing the already failing western arm production.

Why shouldn’t Europeans do the same with US social networks then?

Here is a colab notebook that replicates some of their examples without needing a togetherAI API: https://colab.research.google.com/drive/1pzz1KHQjylrpfZt2xS7...

Choosing the A100 environment is required though

Isn’t it similar to the majorana anyons Microsoft hopes to use to build a topological quantum computer?

I should have meant, have you already found customers? I am also working on a project on genomics, which is very orthogonal to yours, and finding customers in this field the difficult part

Really interesting, what would be your real-world use cases?

I disagree here : there is a noticeable marginal cost to having additional human tutors. It’s quite different with chatbots.

Also I personally think you are under a misconception. Commoditised high quality AI tutors would increase variance/inequality of outcomes, not decrease it. As it would increase the learning rate of gifted individuals even more drastically than the one of the general population.

The more edtech improves and becomes cheaper, the more the limits of individuals will be set by their biological constraints. Which are mostly genetics

Title doesn’t reflect it’s the result of a newly hypothesised model

Strange, I am reading everywhere that India is developing rapidly and will become soon a major economy

Not an issue per se. It shows the limit of the analogy used by defrost to justify such a world would be as functional as 1975’s was.

The basic assumption that the two situations are similar is just… false

Your reasoning is very flawed for a single reason. If by let’s say 2075 the population was reduced to 1975 levels, the demographics would still differ widely.

1. Age distribution would look like an inverted pyramid instead of a pyramid

2. The geographical, racial/ethnical and cultural demographics would also look like the totally different. The share of global population of the global south, which is currently highly dysfunctional, would be drastically more important, especially in the younger age groups.

Note that the increase of wealth of the “third world” is mainly due to China / India, followed a bit by Indonesia / Bangladesh / Vietnam. Nearly all of them have already a TFR < 2.1

Other countries’ economies even lost complexity and the share of their economies that are basically commodity exports increased

The article offers no solution to developed nations demographic collapse other than ideological brainwashing.

Could you please detail why do you think Machine Learning will obviate jobs that are already useless?

Misleading title: the study doesn’t concern the « human brain » but the neural system of a worm

That’s assuming there are no externalities whatsoever to immigration and that the immigrants will be able to satisfy the demand in labor and in entrepreneurs as well as the natives.

From my own experience, in France, both of those assumptions are heavily violated. It can vary depending of the type of immigration though, I bet the story is different for California