As I understand it, Astra is a specialist model for math, physics and AI research.

Lets see ...

Well,
you can not make up your own definitions what words mean.
Dictator: a guy who rules alone by dictating what the others have to do. Historically usually put into place for a limited time: by VOTE! Sometimes a small group.

Tyrant: a ruler who uses force and violence to suppress others, and usually rules alone.
Tyranny doesn't need to be a dictatorship, it can be an oligarchy.
No it can't. You are mixing that up with dictatorship.
The prime definition of a Tyranny is, that it is a single person. Which could a monarch, or an usurpator, anyone who however came into power.

Two completely different things.

And China is not a dictatorship. The leader of the party does not have any extraordinary power, nor is it a Tyranny as no one is suppressed. From the point of view of the people of China, they live in a democracy, that works better than any other on the planet.

You are reading the wrong magazines.

Also you are mixing up countries like Chile and Argentina, and names of rulers. Pinochet was in Chile, and his "contrahent" was Salvadore Allende. Voted into office by a democratic election, and not a communist or an enemy of the US and not connected to Russia, but deposed by the CIA, who put Pinochet into power. Who later murdered Allende.

Yes, but that is intention!
As BigDecimal is the correct datatype, you can not simply add an s, BigDecimals is not a datatype.
So I use - like everyone else in computer science - either an apostrophe 's or put (s) in parathesis.
So that it is visual clear, and easy to double click and copy/paste.

It is unlikely that I ever come to the USA :D

But thanks for the invite.

My center of live will be Thailand and Japan soon.

Are you mixing up Mithril with Lithium?

It is always only a matter of scale and momentum.

* Nuclear fuel is not reprocessed, as it does not solve the waste problem, and same time: making new fuel fram raw materials is simply cheaper than reusing the old fuel.

* Lithium batteries are not recycled (that changed recently), for the same reason: despite the absurd amount of waste, it is cheaper to make new ones from fresh lithium than extracting it from dead ones. That changes now - right now - as the the total amount of wasted batteries is so huge, the matter of scale kicks in.

* Wind turbine blades: same problem. Because actually: we do not have much wasted turbine blades yet. A wind turbine stands 10 to 30 years if nothing goes wrong. Then, what are we supposed to do with them to "recycle" them? It is not so that a whole wind farm gets over night "blade trash" ... recycling old blades will be a business in 10 or 20 years.

* even recycling gold from IT hard ware is only a niche market, same for aluminium, if it is not forcefully separated from ordinary trash, as we do in Germany.

Given how much of modern manufacturing is fully automated, why couldn't US companies compete on a level playing field with Chinese ones?
Because they have to buy the robots from China or Germany.

Or restart the whole supply chain from mining raw materials to producing advanced parts (like electric engines with exotic magnets) to CPUs, parts for robots, and so on ...

The problem is that Trump and his cronies think that USA is the "only market".

Supposed the market is not saturated, like for example the RAM and CPU market, as in: there is lots of demand, but not enough production capacity.

Then the one who puts tariffs on an import loses. As the producer simply sells elsewhere.

Then comes the silly part: I have a factory in China. It does not bother me at all, if my product costs more in your country. It only bothers me, what my contractor in your country pays me. Which will be the same like before. And if he orders less because my contractor is losing customers: I simply sell to a different place.

Most likely I have not even extra paper work, as the importer of my goods has that burden.

Tariffs per se might have a point, if they actually shape a market or are on "dangerous" goods, like alcohol or tobacco.

For example in a few years Germany will probably have high taxes for the owner of a gasoline car, and high tariffs for the imports of such a car. Because "the political will" is to get rid of them completely.

The general principle of tariffs is: you tax/tariff something "bad" and use the money to subsidize something "good". Pushing the market into the direction you want to go.

this one sounds cute, too: https://www.smbc-comics.com/co...

AI systems have a half a dozen of their own languages.
You must seriously living under a rock.

Where you in a coma in a hospital last 10 years, and just woke up?

The rest of your post is funny, though.

Just to fix some of your wordings:
- the agent is not the problem, that is only a few dozen kilo bytes
- the LLM is, but a focused LLM on some special topic can be scaled down to 2 or 3 GB
- and then: if you want optimize that, you use a "fixed point" math library that encodes every weight in 4 bits and scale it down to an 200MB model

Of course the last point, is costly in computation, as you have to extract 16 weights out of a 64bit word to calculate with them.

Point is: no one is using 64bit floating point numbers for weights ... and hence we have tensor processing units or ANN units for "specialized" calculations.

The 10 proofs this article is about: where constructed by math models, not by general purpose models.
It is plain in the story that the models where special purpose trained to find new proofs.

Hehe, interesting.
The instructions would be basically the same for me.
I hope you had a nice dive!

This is the research. What I have learnt so far is even the suppliers are not sure how to apply their AI, let alone the customers.
That is wrong.

First: a customer of Scale.ai for example, simply only wants one who helps him to train an LLM. That is what I did the last 5 years. They do not want to buy the hardware and hire the expertise. And do not want to be exposed.

Secondly: Customers using LLMs/AI, perfectly well know what to do with it. Why would they not?

Instead of wring an incredible complex SQL query, that only works on data that actually is in a database, or asking a programmer and IT to set up a GraphML system: they ask in natural language.

Example:
Give me the ten biggest customers, where my department is responsible for, that used to order from us, but did not order recent 6 month.

Simple. What is not to grasp about that? I don't get it.

I suggest to check Youtube. There are thousands of videos about high tech companies that use AI for scientific problems, like "very special" rocket engines, protein or vaccine "invention" or metallurgic advancement or Perovskite materials, production processes, or solid state batteries, where the break throughs were made with AI tools.

AI, as we call it (wrongfully in my opinion) is used in any kind of big data analysis to find: patterns.
You have a hunch about a better (as in: make more money) product, but can not correctly put it into math formulars: you write a natural language query, and let the AI lose on your data.

Example:
Supposed I had a hypothetical customer, which needs electric power following this load curve over a day (upload photo! picture!). Give me 100 (potential) customers which did not sign a contract with us the recent 24 months.

Point is: the examples are in natural language. There is no programmer translating it into SQL or an IT department providing special pre prepared data that is curated to be in "machine readable" form.

You seriously should simply start using the "AI mode" on google and "ask questions". Normal questions. It is as simple as that: "what are the 20 biggest natural parks on the planet, give me a table, with name, location and size, and a short summary about their specifics".

That might be completely useless for you but I am pretty sure, you find quickly something where AI gives you answers where "normal Google Fu" would not be enough.

Now imagine, you work for a company like BioNtech ... "I have an idea about an RNA approach tackling the BGN45 receptor of a COVID2019/beta-tetha-4 virus. Make me a list what we have tried in our company, and what other companies have published, summarize it to maximum 25 pages".

I seriously have problems to grasp what YOUR problem with AI is.

Of course: feel free to translate everything above into SQL/GraphQL and explain me: who did when set up the databases for those queries.

P.S. depending on the LLM, it will search the web for the stuff I proposed regarding BioNtech, for example: finding and analyzing human readable PDFs for example from arXiv.org

Now something completely different:
""" I want to set up a restaurant. In Thailand. I would like to call it "Nonglak's temple of the moon". Do you think, calling a restaurant with the Thai word for temple (Wat) would be affronting in Thailand? What would be alternatives? Like Shrine or Grove? Or Park or Garden? Would it make more sense to call it "Nonglak's Temple to/of the Goddess of the Moon"? Give me a variation of names, in English, Thai letters and Thai pronunciation in English letters"""

Just put it in any "AI search engine" and look what happens ...

Yeah, I forgot the "whole joke".

It was like: some guy storms into Kennedy's office and yells "The Russians want to land on the moon and paint it red!"
And Kennedy simply replies: "who cares? We write Coca Cola on it!"