Anthropic logo

AFP via Getty Images

All of this can be a little hard to follow.

It seems that, earlier this month, Anthropic released a post about something called the “J-Space” that shows us more about how an LLM thinks. It also posits something called a “J-lens” that can help engineers or human users to “see” what’s in this internal workspace.

The J-space, theorists explain, is a part of the LLM where hidden signals light up, where the LLM shows what is thinking about without generating any explicit expression. It would be kind of like being able to read a human’s mind, or, perhaps more relevantly, to read a human’s body language or something unspoken.

The result is that we are learning more about the interactions between humans and AI entities – but there’s a lot of math and jargon and abstraction involved.

Investigating the J-Space

A simple way to describe this is that an LLM might, for example, get certain inputs and provide an answer, while thinking about something else. It might see a picture of a black cat, and return the word “black” based on tokens, while displaying patterns linked to the word “cat” in the J-space. That’s just one random instance, for the less superstitious among you.

So why is it called the J-Space?

I think this helps to explain what’s going on: the J-Space is so-called after something called the “Jacobian matrix,” where a function maps how each of many inputs relates to the steering of the LLM’s cognitive process. You might have a certain value for input #1, cat, and another for input #2, dog, etc. The map shows part of the LLM’s discriminatory cognitive process, how features make impressions and guide the LLM toward thinking about a particular subject or concept.

MORE FOR YOU

Here’s how Anthropic explains it in the July 6 announcement:

“Each J-space pattern is linked to a particular word. But when one of these patterns lights up, it doesn’t mean the model is saying that word—just that the word is on its mind. If you've heard of language models having a ‘scratchpad’ or ‘chain of thought’—text they write to themselves while reasoning—the J-space is something different. It operates silently, in the model’s internal neural activations, allowing the model to think about a concept without writing it down. Notably, the J-space wasn’t designed or programmed by us, but instead emerged on its own during Claude’s training process.” (italics theirs).

Using the Lens

So what is the J-Lens? Here, you make use of a view of that Jacobian matrix, to see the hidden patterns that represent what Claude (or some other model) is thinking without telling you.

“Apply the lens at each layer, and you get a ranked list of vocabulary words,”

: “ERROR on buggy code, fake on prompt injections, intermediate math steps in order — often absent from chain-of-thought and output. Those patterns collectively form J-space, Claude's internal global workspace. The lens is the instrument; J-space is the phenomenon.”

Evaluating the Intelligence

I want to keep this part fairly narrow, and let you, the reader, do additional research based on your interest level. I want to share two things, one of which follows from the other, to look at what this research is saying about LLM thought processes.

The first is a summary of a piece written by AI and “human-reviewed” by someone named Andy Southgate. It speaks to what the researchers have done in discovering the J-Space and the J-Lens. I’m only going to post the beginning, and the end – in between, there is additional language describing how the J-Space works.

First, the human-AI duo reiterate what happened:

“On 2026-07-06 Anthropic published (on Transformer Circuits Thread) an interpretability paper arguing that a small, sparse subspace of a language model’s activations — which they call J-space, surfaced by a Jacobian lens (J-lens) — behaves like the global workspace of Global Workspace / Global Neuronal Workspace Theory.”

Then, at the end, there’s this:

“Crucially, the authors explicitly take no position on phenomenal/subjective consciousness — the very split the Map builds on. This is squarely on the Map’s access/phenomenal turf and is a gift to the dualist framing (a vivid empirical demonstration of access consciousness that, by the authors’ own words, leaves the phenomenal question untouched), not a threat.”

That last part got me: a “dualist framing” described as “a vivid empirical demonstration of access consciousness that, by the authors’ own words, leaves the phenomenal question untouched.”

I asked GPT to simplify, and explain the nature of a “dualist framing.”

Check this out:

“A dualist framework is a way of thinking that separates two different kinds of things instead of treating them as the same. When talking about consciousness, a dualist framework says there are two separate questions, first, How does a mind process information? And second, Does it have a subjective inner experience? This is called phenomenal consciousness or subjective consciousness. The Anthropic paper only studies the first question. It shows that AI may have an internal workspace for organizing and using information. The researchers specifically say they are not claiming the AI has feelings or experiences.”

Then the model went back over the core idea, here:

“So, in this context, a dualist framework means keeping these two ideas separate: Thinking and reasoning (access consciousness) versus Feeling and experiencing (phenomenal consciousness). The paper provides evidence for the first, but leaves the second completely open.”

Last but not least, this is GPT’s definition of the dualist framework:

“In philosophy, "dualism" more commonly refers to the idea that the mind and body are fundamentally different kinds of things (often associated with René Descartes). Here, however, "dualist framework" is being used in a narrower sense: it separates functional intelligence from subjective experience, without necessarily making any claim about whether minds and bodies are separate substances.”

Descartes, indeed. This is heavy stuff. If you’re the philosophical type, this may be most of interest to you – not what the Jacobian matrix is, but what this means for our perception of the LLM as it starts to become better endowed and “walk among us.”

Stay tuned.