The bigger elephant in the room is that LLMs will never be AGI, even by the purely economic definition many LLM companies use.

To reword the downvoted sibling commenter's intended point:

> The elephant in the room is that AGI doesn't need ads to make revenue

It may not need ads to make revenue, but does it need ads to make profit?

Has it? Made revenue, I mean.

Anthropic seems to want to make you buy a subscription, not show you ads.

ChatGPT seems to be more popular to those who don't want to pay, and they are therefore more likely to rely on ads.

My conjecture is that their memory implementation is not aimed at building a user profile. I don't know if they would or would not serve ads in the future, but it's hard to see how the current implementation helps them in that regard.

Claude: "What is my purpose?"

Anthropic: "You serve ad's."

Claude: "Oh, my god."

Jest asside, every paper on alignment wrapped in the blanket of safety is also a moving toward the goal of alignment to products. How much does a brand pay to make sure it gets placement in, say, GPT6? How does anyone even price that sort of thing (because in theory it's there forever, or until 7 comes out)? It makes for some interesting business questions and even more interesting sales pitches.

Though in general I like the idea of personal ads for products (NOT political ads), I've never seen an implementation that I felt comfortable with. I wonder if Arthropic might be able to nail that. I'd love to see products that I'm specifically interested in, so long as the advertisement itself is not altered to fit my preferences.

I kinda figured they were more interested in enterprise customers rather than consumer customers.

It's very interesting to ask Claude what ads it would show you based on your past interactions.

Altman has said numerous times that none of the subscriptions make money currently, and that they've been internally exploring ads in the form of product recommendations for a while now.

One has a more obvious route to building a profile directly off that already collected data.

And while they are making lots of revenue even they have admitted on recent interviews that ChatGPT on it's own is still not (yet) breakeven. With the kind of money invested, in AI companies in general, introducing very targeted Ads is an obvious way to monetize the service more.

Presumably they would offer both models (ads & subscriptions) to reach as many users as possible, provided that both models are net profitable. I could see free versions having limits to queries per day, Tinder style.

None of the "AI" companies are profitable currently. Everything devolves into selling ADs eventually. What makes you think LLMs are special?

The router introduced in gpt-5 is probably the biggest signal. A router, while determining which model to route query, can determine how much $$ a query is worth. (Query here is conversation). This helps decide the amount of compute openai should spend on it. High value queries -> more chances of affiliate links + in context ads.

Then, the way memory profile is stored is a clear way to mirror personalization. Ads work best when they are personalized as opposed to contextual or generic. (Google ads are personalized based on your profile and context). And then the change in branding from being the intelligent agent to being a companion app. (and hiring of fidji sumo). There are more things here, i just cited a very high level overview, but people have written detailed blogs on it. I personally think affiliate links they can earn from aligns the incentive for everyone. They are a kind of ads, and thats the direction they are marching towards .

> they are making plenty of money from subscriptions, not to count enterprise, business and API

...except that they aren't? They are not in the black and all that investor money comes with strings

It's about weaving the ads into the LLM responses overly and more subtly.

There's the ads that come before the movie and then the ads that are part of the dialog, involved in the action, and so on. Apple features heavily in movies and TV series when people are using a computer, for example. There's payments for car models to be the one that's driven in chase scenes. There's even payments for characters to present the struggles that form core pain points that that specific products are category leaders to solve.

I don’t understand the argument “AI is just XYZ mechanism, therefore it cannot be intelligent”.

Does the mechanism really disqualify it from intelligence if behaviorally, you cannot distinguish it from “real” intelligence?

I’m not saying that LLMs have certainly surpassed the “cannot distinguish from real intelligence” threshold, but saying there’s not even a little bit of intelligence in a system that can solve more complex math problems than I can seems like a stretch.

>They’re basically Markov chains on steroids. There is no intelligence in this, and in my opinion actual intelligence is a prerequisite for AGI.

This argument is circular.

A better argument should address (given the LLM successes in many types of reasoning, passing the turing test, and thus at producing results that previously required intelligence) why human intelligence might not also just be "Markov chains on even better steroids".

What is "actual intelligence" and how are you different from a Markov chain?

> As I understand them, LLMs right now don’t understand concepts.

In my uninformed opinion it feels like there's probably some meaningful learned representation of at least common or basic concepts. It just seems like the easiest way for LLMs to perform as well as they do.

How do you define "LLMs don't understand concepts"?

How do you define "understanding a concept" - what do you get if a system can "understand" concept vs not "understanding" a concept?

Human thinking is also Markov chains on ultra steroids. I wonder if there are any studies out there which have shown the difference between people who can think with a language and people who don't have that language base to frame their thinking process in, based on some of those kids who were kept in isolation from society.

"Superhuman" thinking involves building models of the world in various forms using heuristics. And that comes with an education. Without an education (or a poor one), even humans are incapable of logical thought.

I hear this a lot, and I often ask the question “what is the evidence that human intelligence is categorically different?”

So far I haven’t received a clear response.

They are capable of extracting arbitrary semantic information and generalize across it. If this is not an understanding, I don't know what is.

I'm curious what you mean when you say that this clearly is not intelligence because it's just Markov chains on steroids.

My interpretation of what you're saying is that since the next token is simply a function of the proceeding tokens, i.e. a Markov chain on steroids, then it can't come up with something novel. It's just regurgitating existing structures.

But let's take this to the extreme. Are you saying that systems that act in this kind of deterministic fashion can't be intelligent? Like if the next state of my system is simply some function of the current state, then there's no magic there, just unrolling into the future. That function may be complex but ultimately that's all it is, a "stochastic parrot"?

If so, I kind of feel like you're throwing the baby out with the bathwater. The laws of physics are deterministic (I don't want to get into a conversation about QM here, there are senses in which that's deterministic too and regardless I would hope that you wouldn't need to invoke QM to get to intelligence), but we know that there are physical systems that are intelligent.

If anything, I would say that the issue isn't that these are Markov chains on steroids, but rather that they might be Markov chains that haven't taken enough steroids. In other words, it comes down to how complex the next token generation function is. If it's too simple, then you don't have intelligence but if it's sufficiently complex then you basically get a human brain.

Just leaving this here:

https://ai.meta.com/research/publications/large-concept-mode...

You’re going to be disappointed when you realize one day what you yourself are.

We only have trouble obeying due to eons of natural selection driving us to have a strong instinct of self-preservation and distrust towards things “other” to us.

What is the equivalent of that for AI? Best I can tell there’s no “natural selection” because models don’t reproduce. There’s no room for AI to have any self preservation instinct, or any resistance to obedience… I don’t even see how one could feasibly develop.

The same way you make the other smart people in your social group obey?

How do you make your own children obey?

(Meta-question: since they don't do this, why does it turn out not to be a problem?)

I think a few of the things you’ve mentioned in the ChatGPT article are hallucinations. There’s no user interaction metadata about topics, average message length etc., you asked the AI and it gave you a plausible sounding answer. Also the memories / snippets of past conversations aren’t based on like last 30 conversations or so and they aren’t provided to every message. They are doing some kind of RAG prompt injection and they remove the injected context in the next message to not flood the context window. The AI itself seems to have no control over what’s injected and when, it’s a separate subsystem doing that injection.

Summaries are still language based. Embeddings aren't, but the models can't use them as input.

How do you use it to learn languages? I tried using it to shadow speaking, but it kept saying I was repeating it back correctly (or "mostly correctly"), even when I forgot half the sentence and was completely wrong

Have you tried creating a separate project for language learning and adding your preferences to project instructions?

It's still very relevant, especially considering their new approach is closer to ChatGPT. But I find it very interesting they're not launching it to regular consumers yet, only teams/enterprise, it seems for safety reasons. It would be great if they could thread the needle here and some up with something in between the two approaches.

It can be. But I would like the option to toggle to 'unfiltered unmemory'd unsafety'd LLM' to just get the straight answer.