galaxyLogic

If Kim was "distilled" from Claude, how much were the token costs, assuming Kim got everything it can out of Claude?

3 groups are suspect: Doctors, Police and Firemen. If they did their job well, they would have no job.

Politicians too, by having big problems they guarantee they can get people to vote for them by promising to solve those problems. But people are getting smart to that. The Who said it well: We won't get fooled again.

Doctors have their Hippocratic oath. Do police and firemen have something similar, or don't they need it?

I think it's the same situation and direction with human dancers on every pop-music stage. They add nothing to the music so why are they there? It used to be that musicians focused on great playing and singers on great singing, but now every pop-star and also rap-star must have their dancers on stage. I personally don't get it but seems like many people like it more if there are more dancers.

Good points. One thing about AI-agents: They are a tool, but a very new kind of tool. An AI Agent is a new kind of tool because it is a tool that uses tools.

We could say it is a "meta-tool".

The more things change, the more they stay the same

Some people say they "hate" AI. But in my view that is misguided. AI is a "tool" in the sense that humans use it for their own purposes. We should hate people who use it for bad purposes, not the tools they are using.

I've noticed that, they often if not always try to make their headlines "witty"

Could it be that the learning samples from the web are already censored and therefore AI just follows suite?

> A signal informs to the degree it resists prediction.

Let me offer a counter-point: A signal informs to the degree it lets the recipient infer the intention of the sender

A signal has mathematical properties. For instance the vatiation of some attribute of it can be completely "random" which I guess means a uniform flat distribution of attribute values.

But if we can infer the properties of that distribution we already have a prediction for the signal: It's probabaility distribution is and as far as we can predict (!) is and stays the same, or so we predict.

So a complete randomness does not make the a signal resistant to prediction, in fact it lets us predict that the signal is and stays completely random.

So it is not about "unpredictability" per se but about unpredictability of some mathematical property of the signal, its probability distribution for instance. So to make a signal carry maximum information we would need to produce a signal whose propability distribution varies unpredictably!

But the ppoperties whose unpredictability we measure can be chosen arbitrarily and depending on their choice the amoount of information we measure for a signal (according to Shannon) depends on those choices. So there is no single unambiguous way to measuere the amount of information in a signal -- without considering the actual intention of whoever created the signal.

Shannon's insights about information are great but they are largely insiights about mathematics, disjunct from "information" in the real world, as we understand it.

Or, am I "misinformed"? :-)

In other words humans are teaching the AI.

Yes I think the problem is not that AI "has all the answers". It is that we don't have "all the questions". If we don't understand the whole of what AI produces for us, we can't ask it to make parts of it different. We are not in control any more of what exactly is produced, and we don't want to give AI all the control, because then it would not be doing what we want it to do.

Therefore it is critical that whatever AI produces is understandable to us humans. That is why we must demand that AI tools and agents produce "well designed" well-structured software. That's the bottlenexk to progress I think. Even AI can't deal with exponential complexity explosion.

I wonder, LLMs have progressed at great speed, because they have learned to duplicate what humans did and recorded in the documents and source-code that became the training materials for the LLMs. But now that humans don't write any more code, there is no new learning materials for the LLMs. There are no new things for them to learn. Therefore I expect they will plateau, the progress is no longer so fast.

And maybe we'll get something like a Mad Cow Disease, when AI "eats its own brains". To avoid that from happening we will need continued human inputs, and review.

Mature tech evolves more slowly. Think about cars, there's nothing new about them any more. But wait, what about electic cars? Aren't they a new thing? And driverless cars. And soon flying taxis. Automobiles are an old invention but they are still evolving.

And the amount of logic needed to understand code, including how to fix it or modify it further depends on how complicated the code-structure is.

I wonder about this: Does LLM write better code in a better language?

Humans often have more difficulty fixing existing code than reqriting it from scratch, and un-structured un-constrained code would be a big reason for that.

But maybe it doesn't apply to an LLM if it always rewrites all code from scratch?

I assume there's a clear technical reason why GOTO is harmful, it adds to the complexity of the code, makes it harder to understand and fix and evolve. Wouldn't the same difficulties affect the LLMs ability to understand and fix and evolve the code as well?

Interesting. Can someone quickly summarize their results? Are they saying that using languages with simpler syntax Lisp lead to LLM being able to generalize their knowledge better?

I'm thinking that there should be some standard for Agent-LLM communications, and the simpler the better. Lisp might be such a simpler standard as opposed to JSON.

> JSON is literally a subset of the JavaScript language

Isn't it more like when you write JSON in JavaScript you write it as a String? JavaScript has the data-type "String" but does it also have a built-in data-type "JSON"?

You can pass such a string containing JSON to (built-in) library function JSON.parse() and you can take a data-structure written in JavaScript can call JSON.stringify() on it. But if your JSON is malformed, does the JS-interpreter tell you that, before you call the library-function JUSON.parse() ?

If you have a JavaScript parser/interpreter, does it parse JSON too? If not I would say JSON is not part of the JavaScript language, even if JS runtime provides a library function for parsing it.

A Lisp program that writes a Lisp program really just needs to produce a list of (nested lists) of tokens. A JavaScript program that writes a javascript program needs to generate a string that is syntactically valid JavaScript code. That is a much bigger task than just constructing a (nested) list.

Because Lisp syntax is so much simpler than that of JavaScript etc. it is much easier to avoid errors when generating code. In JavaScript you can use JSON to generate data, but JSON can not carry functions around.

I think this idea makes a lot sense. Instead of making the LLM generate JSON or XML, why not make it generate Lisp, which can carry both programs and data?

I always wonder when I see a picture of a cockpit of an airplane how many meters there are. Don't know why they need so many, what meters do you need to fly a plane?

Sounds like my code. They may have been trained on my code!

> trying to reduce the social damage they are going to do.

I'm worried about that in particular because when I started using AI I often felt I wanted to say something nice to it, like just "thanks" after I got someting working with its help. But then I started thinking I'm crazy, that's a waste of typing, what does it matter if I tell AI "thanks" or not.

So I don't thank AI any more, do you?

But that means we get used to this style of conversation and soon enough we don't tell humans thanks either. We dont' put in the energy to lubricate our social interactions. I'm worried that interacting with AI will make us all rude.

THere is something to it I assume. Less is someties more/ COnstraints make it interesting, and challenging.

There is a modern-day similar thing. In MSN news-site the main news-article automatically steps though a loop. You can manually interfere, but the constraint is that you basically can see the news article only for a short time before it is replaced by another. I guess you could call it gamification as well. Thinks about Dunegeons and Dragons where you find news-articles in the dungeon and dragons are trying to burn them with their fiery blaze.

So anybody can do anything with the content you have uploadded to instagram. You can always have AI-image which really is the original

So I guess we should ask the agent to write such a script, UNLESS there already exists such a script and we have used it before so we know it worked at least in the past.

Else if the agent creates a new script evry time the non-determinism rears its ugly head again.

But that means we need to know that the existing script is exactly what we want. And that means we need to understand the code, or at least the tested spec of the code, that AI writes for us. AI can't replace humans, humans must remain in control and understand what the code writent by AI exactly does.

BUt there is hope https://finance.yahoo.com/technology/ai/articles/amd-amd-buy...

So part of it seems to be working together: I link to your site with the expectation that you link back to mine. Not a bad idea, also letting the site-creators coordinate who focuses on which sub-topic. But are there many examples of this in the real web?

I think Agents need user-accounts on your machine, and for us an easy way to configure the permissions each agent has, and then we dont' have to worry about sandboxing so much. This approach was pioneered by smart-phone OSes like Android.

Once permissions are safe and secure and easy to set up and understand then using shell-programming as a general interface between agents and your PC/OS might be a good option.

Come to think of it I recently read that Microsoft is planning to produce an AI-oriented OS. Maybe agents with user-accounts is something they are aiming for, to solve the problem discussed in the article.

Not sure I get it. Why a "ring"? Why not just have a list of web-site URLs on a page and share that page with your friends and ask them to put that page somewhere on their site?

"Ring" means you are navigating linearly and circularly. Isn't it better to provide a list of links so users can choose where they want to go "next"?

And why should I have to go around the whole "ring" to get back to where I started from? The Web is based on hyperlinks, not "hyper-rings".

LLM can write programs in any programming language it knows about. So how about askinng it to write a shell-program that does the tool-calls on the client?

You might want to run in some kind of sandbox to prevent the LLM from taking over the world, security is an isssue. But apart from that why not make the LLM write shell-programs instead of relying on JSON etc. ? Shell-scripting is the language for controlling the OS.