AI Hype is completely out of control – especially since ChatGPT-4o [video]

The problem is that capital isn't that patient. People are sinking billions into LLM integrations, at discount rates of 5%+. If it takes 20 years for a tech to pay off, at a 5% discount rate, and you sunk a billion dollars into it, it needs to earn back $2.65B. Moreover, at some point during that 20-year period, somebody's going to ask "Where's my billion dollars?" and pull the plug on the project.

I think the tech behind LLMs may eventually be game-changing, but that tech is going to change ownership and get reinvented several times, and we won't actually see profitable, sustainable benefits until industries get refactored into cost structures that make sense for what LLMs can actually do. The web needed its Netscape and Yahoo and Geocities and Apache, but it also needed Google and Rails and Django and GitHub and Stripe and Facebook and Webkit and nginx and MySQL to really become what it did.

It also does concern me how basically nobody is building a real product around the current state of "AI", but are rather hinging their success on what they believe it will be in the near future.

edit: to be clear, I'm saying that we are running into scaling "walls" that make hard extrapolation based on increased investment senseless.

It will likely be disruptive in some areas short term, but will take much longer in other areas.

It's one of those short term is overhyped, long term is underestimated things.

Not too many applications can justify spending huge amounts of energy generating answers that may just be nice sounding BS.

* User interfaces. It does provide a genuinely new way to interface seamlessly with existing software.

* Translation.

* Generating bullshit. Yes, there's demand for this even if perhaps there shouldnt be.

* Not much else. That includes using it as a specialized autocomplete. I think it falls down pretty badly at that.

For coding I have long given up on it. I only see it useful for people that want to do small scripts in languages they are not familiar with or to write toy examples of web/mobile apps.

For art I can sort of see a workflow in adding details to textures and using existing images and 3d assets to guide the generation of images. This is all just my speculation so I could be completely wrong just like the person that knowns very little to no programming thinking they will be able to code whatever they want with it.

The only skill I brought to the table was app architecture.

It's not just hype.

“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” ― Edsger W. Dijkstra

Moreover, ML has historically been very brittle compared to the generalization humans provide. A machine that reads addresses might be 10x better than a person, but it may not be worth it if you can’t do things like say “we don’t even sort the blue envelopes—there’s a specific rule for those” like you could with a human. (The specific case here is irrelevant—humans can handle arbitrary variation on the rules infinitely better than a bare CNN.)

It’s this ability to cope with human-level ambiguity that has so much potential in LLMs. But even they don’t work like people—giving it a new rule could make it worse at all other tasks with no explanation as to why.

Even ignoring the people who insist it comes with baggage of consciousness or qualia.

AI can train on millions to billions of examples in a matter of months as transistors outpace synapses by the degree to which marathon runners outpace continental drift… but AI also need to do that well just to get up to the current level.

If AI could learn from as few examples as we require, Tesla would have achieved level 5 driving years ago — but as is, they've got millions of vehicles with years of real world experience and it's still merely "ok", not even "expert".

If we never learn how to make machines learn faster, then we'll have an economy of people whose jobs are essentially "give the machines examples to learn from"… and for some of those jobs, they could spend years providing those examples and still no machine will do as well.

Human-level intelligence should adapt to many uses.

It may require teachers, but not prompt engineers.

The article author is very valid in the stance that this is far from intelligence.

Yes yes, you wouldn’t criticize a fish’s intelligence for not being able to climb a chessboard.

But these things are stupid as hell. They are not even close.

I used the opportunity to send some queries which GPT-3.5 was bad at, and while GPT4 was better and closer to the correct answers, it still failed to produce them. When I pointed out the problems, it got even closer, but still ended up insisting that its (wrong) answer was correct in the end.

So, yeah, GPT4 is much better than GPT3.5, but it's got a long way to go before it becomes real AI.

Currently, I still use it for 3 cases:

1. When exploring something completely new. It saves time reading and learning because you can ask questions with "layman words" and get to learn the specific terms used in the domain quickly.

2. When google search is swamped by SEO spam or google insists on ignoring some of the keywords

3. When I need some boilerplate code that has been written a thousand times. It saves time which would otherwise get spent on concatenating different code segments from 5 different Stackoverflow threads and reading the docs afterwards to ensure it's correct. Now I just copy/paste from ChatGPT and read the docs afterwards to fix any places where it messed up.

https://www.theonion.com/guy-who-sucks-at-being-a-person-see...

So, will they do better? I personally think that knowledge is about implications. Notably, things that are similar do not necessarily have identical implications. And I can't see LLM tackling this issue anytime soon. (I even doubt that LLMs provide a suitable approach to this, at all.) So we'll probably end up with "we managed to get rid of reliability", which may be just the opposite of the role, the computer plays in Star Trek scripts.

*) To provide a bit of context to reliability and repeatability: These are really externalization of the major generalizations of Husserl's life-world (Lebenswelt), as in "and so on" and "I can do this again". We are generally fine with machines, because they embody these principles in confined parameters. However, there's also a certain danger in this, as we tend to generalize and idealize along these lines, as this is essential to how we construct and navigate the world, we live in. We'll also apply this to things that are just "mostly", and are apt to be fooled by this, as may be the case with LLMs. (And in the past, machines were a great way out of this trap, as long – and as soon – as we were able to verify the parameters and boundaries.)

And "intelligence" can be defined. People just think of different things when you use the word unless you define it. Here is how I like to define intelligence: the intersection between knowledge and reason. An example of a system that is highly knowledgeable but lacks intelligence due to its inability to reason is Wikipedia. A system that is highly reasonable but lacks intelligence due to its lack of knowledge is a calculator.

It's obvious that language models encode more knowledge than humans do. Even a few GB model that I can download and run on my computer can spit out facts about random things. Way more than any human could. Sometimes they are wrong about facts, but so is any knowledge system, including humans.

An LLMs reasoning capability is a bit less clear cut. When you have any understanding of how LLMs work, it seems like they should have no ability to do novel reasoning. Yet when you try to push them, they are often surprisingly good. I think that people who say the best LLMs have worse than human reasoning capabilities probably overestimate the average adults reasoning capability. Plus it is not hard at all to hook up a language model as a component in a greater intelligence system, by adding other components that specialize in reason. For example an environment to run code in.

https://morbotron.com/caption/S02E14/267783