Abstraction Research

Jasper Gilley
San Francisco, California
September 2026

TLDR: I'm making my research monorepo public at github.com/jagilley/abstraction. It's a live lab notebook for a single research program: understanding how a learner can create its own abstractions, which current models can't. It contains the experiments, the idea docs I use to reason about their results, and the belief trees that record what I currently think. It will be updated weekly.

This is a live research repo built to investigate the following observation:

here is a very specific sense in which current LLMs are not AGI. If you define “general” as “capable of representing a general assortment of inputs”, then they are general. But if you define “general” as “capable of learning a general assortment of data-limited domains”, then they are not general, because they cannot endogenously explore and learn complex domains.

Supervised learning learns robustly from external data, but can't mint its own knowledge. RL learns everything for itself, but learns with specific regard to an external verifier, and thus can't navigate great complexity. The process of creating new abstractions requires a synthesis of both.

Other forms this problem statement may take:

• Sample efficiency ("how do you learn more from limited data?")
• Continual learning ("how can the things you know today influence the way in which you learn new things tomorrow?")
• Interpretability ("how do you judge the quality of your own thoughts before you have supervision on them?")
• Taste/style/originality ("how do you create abstractions which are both endogenous and coherent?")

Ideas which are related but subtly different:

• Meta-learning (learning over the class of abstractions rather than the class of data is the right idea, but traditional meta-learning still creates abstractions passively with respect to the data it sees)

Repo structure

This repo consists of two main components: experiments used to test predictions, and various forms of "idea docs" used to reason about the findings generated by the experiments. Experiments live in the experiments/ directory, and ideas live in the ideas/ and beliefs/ directories, as well as in the READMEs written post hoc for each experiment.

Here is a selection of relevant experimental lines:

• experiments/rhm is a controlled setup for generating language-like data with a known data generating process.
• experiments/a2a_forward is our main experiment for training forward self-models for decomposing activations
• experiments/mjc is a place we run MuJoCo experiments for motor learning
• experiments/canvas is a compositional image substrate: a tile grammar rendered to swatches

All experiments are implemented and written up by Claudes; human authorship lives at the epistemic and conceptual level.

Getting started

Due to the nature of this repo as an auto-organized live lab notebook, the best way to understand its history is likely to start from live threads which have been added recently. New updates to this repo will be pushed in a weekly PR, which will briefly explain the intellectual lineage of each item added.

More broadly, the ideas/ directory contains human-readable documents which are intended to serve as epistemic glue between experiments. Not all of these are still "live" (that is, currently thought of as being broadly representative of reality), but a few which are reasonably still-in-play are:

• ideas/heterogeneous_graders.md
• ideas/practice_manufactures_its_own_credit.md
• ideas/intelligence-as-rotation.md
• ideas/two_climbings.md
• ideas/two_timescale_value_loop.md
• beliefs/trees/operators_not_footprints.md

Apologies in advance for the Claudish prose!

FAQs

Why is abstraction creation an important problem?

Consider several known open questions in the field of AI. My view is that each of these phenomena is an instance of what it looks like to form model abstractions passively, based on the data and verifiers the model sees during the training process.

• Q: why do the models seem to generalize much worse than humans?
  • A: under the current model training paradigm, generalization, for example, is a property of the particular distribution of data a model sees during training. If the model only forms abstractions with respect to the data it sees during training, it must rely upon transfer learning over the geometry of the data to generalize to inputs it has never seen.
• Q: why can the models easily achieve superhuman capabilities on tasks like competitive programming but not those like open-ended research?
  • A: because competitive programming requires skillful deployment of existing abstractions rather than the invention of new ones
• Q: why does finetuning the models on domain-specific data not produce gains outside that domain, when for humans it is possible to learn things that generalize from seemingly specific domains?
  • A: because models adapt during fine-tuning by treating the fine-tuning domain as the new ground truth against which abstractions must earn their worth
• Q: why are superhuman Go engines trained by self-play vulnerable to adversarial policies?
  • A: because self-play only creates abstractions passively, in regions it has visited
• Q: why do models collapse when recursively trained on their own outputs?
  • A: because they have no need to form new abstractions passively to represent data already generated by their policy

Why do you think frontier research can be done on non-LLMs?

I think there are two questions bundled together here:

1. Can non-LLM research currently be frontier-level useful?
2. Do LLMs work the same way as other types of AI models, in a learning-theory sense?

To the first question, I would agree with the scaling-maximalist consensus that the answer is "no", and I find the fact that the answer is "no" to be thought-provoking. To my mind, this fact would seem to imply something about the data on which we train LLMs: what is it about human-generated textual data that makes it uniquely useful? I suspect the answer has something to do with the directness with which our language encodes the abstraction byproducts of our cognitive processes.

To the second question, I would say "yes", and I don't think many would disagree. Whatever it is that makes training a transformer on exabytes of pure language more useful than training a transformer on exabytes of pure images seems like it ought to be downstream of the language data itself, and the model representations required to be a good next token predictor over it.

Biological learners seem to be capable of building abstractions over non-linguistic data, and the language data which makes LLMs so powerful comes from biological learners. So it seems like the default position ought to be that non-language data can be fertile ground for frontier research, until proven otherwise.

What's the research strategy?

This work is generally guided by investigating how the human/mammalian brain works, and building models that functionally approximate its biology and how it handles various realistic tasks.

What's the relationship between abstractions and superintelligence?

I define "superintelligence" as capabilities which are nontrivially in excess of those of current AI models. My belief is that abstraction creation is an acute deficit in current model capabilities. Therefore, "superintelligence" is one word you could use to refer to this research program.

That being said, it's pretty robustly established that biological learners - even some which are surprisingly phylogenetically distant to humans - do not have this same deficit. So I would argue that intelligence (super- or otherwise) is likely better thought of as a vector than a scalar.

Contact

If you want to fork this or work on superintelligence with me, please reach out! I can supply you with additional tools for working effectively in this repo, in the house style.

You can DM me on Twitter or email me at [my name] @ gmail.com.