Agreed with the points about intelligence definition, but on a pragmatic note, I’ll list some concrete examples of fields in AI that are not LLMs (I’ll leave it up to your judgement if they’re “more intelligent” or not):

• Machine learning, most of the concrete examples other people gave here were deep learning models. They’re used a lot, but certainly don’t represent all of AI. ML is essentially fitting a function by tuning the function’s parameters using data. Many sub-fields like uncertainty quantification, time-series forecasting, meta-learning, representation learning, surrogate modelling and emulation, etc.
• Optimisation, containing both gradient-based and black-box methods. These methods are about finding parameter values that maximise or minimise a function. Machine learning is also an optimisation problem, and is usually performed using gradient-based methods.
• Reinforcement learning, which often involves a deep neural network to estimate state values, but is itself a framework for assigning values to states, and learning the optimal policy to maximise reward. When you hear about agents, often they will be using RL.
• Formal methods for solving NP-hard problems, popular examples include TSP and SAT. Basically trying to solve these problems efficiently and with theoretical guarantees of accuracy. All of the hardware you use will have had its validity checked through this type of method at some point.
• Causal inference and discovery. Trying to identify causal relationships from observational data when random controlled trials are not feasible, using theoretical proofs to establish when we can and cannot interpret a statistical association as a causal relationship.
• Bayesian inference and learning theory methods, not quite ML but highly related. Using Bayesian statistical methods and often MCMC methods to perform statistical inference of the posterior with normally intractable marginal likelihoods. It’s mostly statistics with AI helping out to enable us to actually compute things.
• Robotics, not a field I know much about, but it’s about physical agents interacting with the real world, which comes with many additional challenges.

This list is by no means exhaustive, and there is often overlap between fields as they use each other’s solutions to advance their own state of the art, but I hope this helped for people who always hear that “AI is much more than LLMs” but don’t know what else is there. A common theme is that we use computational methods to answer questions, particularly those we couldn’t easily answer ourselves.

To me, what sets AI apart from the rest of computer science is that we don’t do “P” problems: if there is a method available to directly or analytically compute the solution, I usually wouldn’t call it AI. As a basic example, I don’t consider computing y = ax+b coefficients analytically as AI, but do consider general approximations of linear models using ML AI.

I presume you don’t want an answer from an LLM, but maybe somebody can give their two cents about the options below:

1. CNN’s
2. RL
3. GNN’s
4. GBDT
5. Transformer Models
6. Symbolic AI/Logic Engines
7. VAE’s

They seem to be all specialized AI tools, so maybe doesn’t answer your question directly.

I think the answer might be a collection of specialized AI tools would be better than LLM’s

Thinking about it, a collection of federated specialized LLM’s would be great because you’re lot limited to a single tuning.

I mean, LLM are just next token in sequence predictors. Thats all they do. They take an input sequence, and based on that sequence, they predict the next thing to come. Its not magic.

Its just that embedded in language (apparently), there is information about meaning, at least as humans interpret it. But these methods can be applied to any kind of data, most of which doesn’t come prepackaged with some human interpretation of meaning.

Just to toot my own bugle, here is some output of a transformer model I’m building/ working on this very second (as in, its on my second monitor):

Here I’m using time series climate, terrain, and satellite data to predict the amount of carbon stored in soils over time. Its the same idea as an LLM, except that instead of trying to predict the next token embedding in sequence, I’m asking the model to predict the next “climate/ spectral/ terrain/ soil” state, in sequence.

Typical models in this domain maybe can predict at about a 0.35-4 R². Here you can see I’m getting a 0.54 R².

All ML as we currently know it is basically giant piles of linear algebra and an optimization function.

Edit:

Here is a bonus panel that I just made:

(pdsi: Palmer Drought Severity Index, basically wet season dry season.)

I mean, given that LLMs aren’t intelligent, I guess you could include video game NPCs. They are often referred to as AI but are also not anything remotely close to true AI the same way LLMs aren’t close to true AI.

I would propose calling such “AI” an “AIM;” Artificially Intelligent Mimickry. Because they mimic intelligence without actually possessing it.

To answer this sort of question, you first have to ask how one quantifies intelligence in an unbiased and rigorous way.

For instance, can an LLM be considered intelligent if it speaks like a person but can’t do certain math problems?

There are lots of niches for non-LLM AI models in tech that aren’t “intelligent” but can do quite a good job when it comes to narrow, well-defined tasks like image recognition, for example.

I think that it’s important to consider that language can evolve or mean different things over time. Regarding “artificial intelligence,” this is really just a renaming of machine learning algorithms. They definitely do seem “intelligent” but having intelligence and seemingly having it are two different things.

Currently, real AI, or what is being called “Artificial General Intelligence,” doesn’t exist yet.

How are you defining intelligence anyways?