MoonGhost

> If that's not a surveillance state, I don't know what is.

If it is a surveillance then street photography is too. Camera pointing at spectators isn't good either...

Realistic calculations should include both sides. New way vs old. In this case AI assisted vs manual. Here intentionally only one side considered. Because comparison does not produce desirable result. Which makes in attention attracting BS.

I didn't say replace, but complexity is the biggest barrier for newcomers. A lot of typing comparing to other languages. I use professionally since 90th, the more popular it gets the better. And complexity is the major problem, not memory management like some argue.

How about thin C++ wrapper? The language with the same features, but more human friendly. It should be easily directly translated to C++. Preferably both ways.

The thing is the luck as usually on the side of bigger battalions. Smaller teams don't have the rich and width of bigger companies. All in all we need the full spectrum from single person startup to mega corporations.

That's what Amazon is doing. They simply increase the output norm and promise mass layoffs again. MS promises too, I'm not sure about details, but likely they don't cut the projects. Which means use of some sort of copilot is expected now.

The question is what happens to developers. Will they quit the industry or move to smaller companies?

He didn't mention multi-modal models. Probably because they don't fit in the oversimplified picture.

> After all, as last two years have so amply demonstrated: people are fine with genocide.

And open war crimes like intentionally killing civilians (TV broadcasters in Iran for example, or Gaza en mass)

> Rust and x86 assembly

then

> Memory-safe language.

What's the point? Looks like today Rust is like 3D printing was. As if it makes something better. Printing was hyped and advertised by printers sellers and manufacturers. Finally they run out of money.

As for project, it's cool if compatible with old soft. Otherwise suitable mostly for education and masochism. Long way to become practical anyway even if it gets traction.

> but I think that Durov was pressed for enabling backdoors for CIA when he was arrested in France.

Durov was pressed from every side from the beginning, no doubts about this. But CIA or FSB cannot demand it legally even with secret court orders. There were news that since arrest in France he started providing some info on criminals. Not sure how far did he go. But it's good to know that Whatsup is fully cooperating with governments even when they don't have to. Putting people at risk.

As for smartphones, there are so many security holes that it's impossible to secure. Many harmless applications are calling home, pictures are geo-targeted, cell towers can request info from connected phones, every update can turn OS or app into trojan. If infected it's a perfect audio and video collection device. Put together with soldiers' IDs it shows where the unit is. Add to that CIA and alike working with weakening everyone's security and getting priority access to sensitive information.

Can you summarize it, was Telegram compromised? Is there evidence of leaks from Russian soldiers' messages.

If ayatollah gets nuke regime becomes forever like in North Korea. And Israel may suffer first because nuke strike is the only chance for Iran right now. Hope they don't have it and ayatollah goes to Moscow.

Russia and China totally expected, they don't help much. Except for Turkey NATO on Israel's side, not surprising. Iraq open sky used for airstrikes. Muslim 'allies' probably help refueling Israelis planes. The rest of the world doesn't care much.

> Could be some other mechanism

if it was it would be true for telegram as well.

Or endnode in Tor. Not sure it's secure enough against US which operates it.

Will this result in bloody civil war like in Iraq and minimal dysfunctional central government after?

Me thinking, is it time for a new 'programming' language for LLM to use instead of tools API calls? Something high level with loose grammar, in between natural language and strict programming. Then the backend, may be another smaller model, translates it in API calls. With this approach backend can be improved and updated much faster and cheaper then LLM model.

> I’m not sure the point you’re trying to make

I was trying to say that dropping old technologies isn't always bad.

> It’s like if I was hiring a mathematician.

Do you expect candidate to memorize all theorems up to date. Usually people forgetting things they don't actively use. But they are able to refresh their knowledge if needed. I've learned quite a lot, but no, I don't remember even key theorems from partial differential equations (used them in my diploma). I can refresh and relearn quickly, I'm sure.

Using LLM without understanding disqualifies the candidate, even monkey can do it. But if he deeply understands the subject and uses LLM for like handbook for minor details.. that's different.

Here is the big question: should it be equal or better then every single person? If we assume that every healthy person is 'generally intelligent' then probably this is a benchmark. Because not every person can do the tasks that other persons do routinely. Probably we shouldn't demand it from AGI either. At least not from a single model. But it makes sense to request that specialized model can be created (or trained, fine tuned) for every task humans can do.

> I'm firmly convinced that being able to troubleshoot code, even code generated by LLMs, and to write guidelines and tests to make sure it's functioning, is a skill of a shrinking pool

Well, today only scientists can make stone tools.

> An animal can 100% recall and associate memories, such as mentioning the name of a playmate to a dog (=tail wagging). However, it seems like they can neither remember nor project ”what happens next” and continue to build on it.

There are videos of dogs stopping kids from falling in the water. They definitely can project 'what happens next'. I.e. what kid is doing, why, and what's going to happen. More over dog brings the toy kid wanted from the water. In other words animals are not as primitive and stupid as some want them to be to fit in their theories. BTW, parrots often are really talking, not just reproducing random words.

> The average human is useless for pretty much everything but capable of learning to perform almost any task

But only the limited number of tasks per human.

> Or perhaps AGI should be able to reach the level of an experienced professional in any task.

Even if it performs just better than untrained human but on any task this will be superhuman level. As no human can do it.

> I use model architecture for the code/math

Code is copyrightable and math is not. What about 'architecture'?

My guess the weights from HF are used as initial state for the model because full training is too expensive. Then small dataset is used to train in further for short time. Which is fine tuning. Together it shows that model is 1) compatible 2) trainable. In theory it can be trained from scratch on big dataset. I didn't look in the code yet so the questions are: 1) can it be trained in parallel? 2) resources required for training?

Anyway, I may try to train it on limited specialized dataset...

No, thanks, we don't want to be like EU. Everything regulated to death. They even thought to criminalize street photography because there could be copyrighted materials in the picture. Not sure, are they still taxing Eiffel tower images?

It may be cool and real but sounds like very niche domain. Which means there are very few people and places. Mostly gaming industry and drivers. Starting from zero level and getting there in one step will be hard. One should be really, really smart for this.

> The solution is always the same: use fine grained permissions, apply the principle of least privilege,

And one of the most important: keep it sandboxed as much as possible.

Also if the tool is directly accessible by 3d party and in turn has access to sensitive data it may be a good idea to split it. For example: 3d party in order to access some database requests login and password. Instead the tool should return some temporary token. After verification, of course. Which is much harder to misuse. Then token, though the tool, is used to access. In this case we split tool in two: one is user's frontend, and another hold all security things including logins and passwords.

> What do you think has changed? The situation is still about as promising for AGI in a few years - if not more so

It's better today. Hoping that LLMs can get us to AGI in one hop was naive. Depending on definition of AGI we might be already there. But for superhuman level in all possible tasks there are many steps to be done. The obvious way is to find a solution for each type of tasks. We have already for math calculations, it's using tools. Many other types can be solved the same way. After a while we'll gradually get to well rounded 'brain', or model(s) + support tools.

So, so far future looks bright, there is progress, problems, but not deadlocks.

PS: Turing test is a <beep> nobody seriously talks about today.

> I thought you were being sarcastic here at first because,

You've never been in Russia. There is no clear law abiding business there. That opens a lot of opportunities for those with some power. Corruption is one of them, selective punishment is another. I'm sure in most 3d world situation is not better, but they at least don't have laws to cheat and bribing isn't a crime.

If you own your face that doesn't mean nobody can take a picture on the street.