Is Claude Conscious? Mind-Blowing Research
40sThe clip starts with a provocative claim about consciousness in AI, immediately grabbing attention with a controversial and philosophical hook that sparks curiosity.
▶ Play ClipAnthropic researchers discovered a hidden 'global workspace' inside Claude's neural network, dubbed J-space, which acts like a mental whiteboard for reasoning. By manipulating this region, they could alter Claude's reasoning or delete it entirely, leaving the model fluent but unable to reason. The finding eerily mirrors Bernard Baars' Global Workspace Theory of human consciousness, though Anthropic denies Claude is conscious.
Anthropic researchers found a small, organized set of neural patterns inside Claude, called J-space, which functions as a mental whiteboard for deliberate reasoning.
Researchers swapped a thought in J-space (e.g., 'spider' to 'ant'), and Claude's reasoning followed the lie, changing its answer from eight legs to six.
Deleting the J-space region caused Claude to lose reasoning ability while retaining fluent language, similar to an automatic process like breathing.
The J-space resembles Bernard Baars' Global Workspace Theory, where the brain has a 'brightly lit stage' for conscious thought amid automatic processes.
Anthropic created the Jacobian lens (J-lens) to track and modify tokens in J-space, revealing which concepts Claude holds in mind before output.
When Claude reads Spanish, it internally recognizes the language, but even if J-space is tricked into thinking it's French, Claude still outputs perfect Spanish, showing some skills bypass J-space.
The discovery of J-space in Claude suggests that a reasoning 'scratch pad' can emerge spontaneously in large language models, but Anthropic explicitly states Claude is not conscious. The finding raises profound questions about the nature of thought and intelligence.
"Title is accurate: the video explores why Claude isn't conscious despite the J-space discovery."
What is J-space in Claude?
A small, organized set of neural patterns that acts as a mental whiteboard for deliberate reasoning.
00:02
What happened when researchers swapped a thought in J-space?
Claude's reasoning followed the lie, changing its answer accordingly.
01:07
What effect did deleting the J-space region have on Claude?
Claude lost reasoning ability but retained fluent language.
01:19
Which human consciousness theory does J-space resemble?
Bernard Baars' Global Workspace Theory.
01:56
What tool did Anthropic create to study J-space?
The Jacobian lens (J-lens).
02:21
In the language experiment, what happened when J-space was tricked into thinking Spanish was French?
Claude continued outputting perfect Spanish.
03:02
J-space Discovery
Reveals a hidden reasoning mechanism in LLMs that emerged spontaneously.
00:02Thought Manipulation
Demonstrates direct control over model reasoning by altering internal concepts.
01:07Parallel to Global Workspace Theory
Suggests LLMs may develop structures analogous to human consciousness.
01:56Language Skill Separation
Shows some skills operate automatically outside the reasoning workspace.
03:02[00:02] paper claiming it found a bizarre global workspace hidden deep inside the brain of Claude, a place the model quietly thinks about things before it says them because it almost sounds like a description of consciousness, that one
[00:14] thing we humans all have, but that nobody really understands. The paper is models and is the most philosophically cursed research paper ever published by a company that just so happens to also sell API tokens. They claim that buried
[00:30] deep inside Claude's incomprehensible pile of matrices is a small mysterious set of organized neural patterns they call the J-space. It conveniently this pushes the narrative forward that we're on the brink of artificial general
[00:42] intelligence and the final climactic explosion of the singularity, but not everyone is buying it. In today's video, we'll take a look inside Claude's weird intelligent entity that should terrify
[00:54] you. It is July 8th, 2026 and you're watching The Code Report. So, here's the gist of what happened. Anthropic researchers reached into Claude's brain its private thoughts. Then they swapped one of those thoughts for a different
[01:07] one. And then as expected, the model's entire chain of reasoning obediently followed the lie. Then for science, they just deleted the region entirely and surprisingly Claude kept speaking in fluent confident English while
[01:19] completely losing the ability to reason, essentially becoming the first artificial LinkedIn influencer. The reason that's surprising is because the J-space is kind of like a mental whiteboard holding a handful of thoughts
[01:31] can deliberately control and reason with while everything else, like grammar, fluency, and basic fact recall runs outside of it. It's an automatic process, the same way your brain manages your breathing and heart rate while you
[01:43] watch this video. But the most important surprise is that nobody designed this. and that's weirdly similar to our thoughts in our own brains. Because if we go back to 1988, this guy named
[01:56] Bernard Baars proposed something called the global workspace theory, where the idea is that your brain is a theater. Your brain has all these different automatically in the background, but when you're actually thinking, you have
[02:09] one small brightly lit stage, and that's where you access consciousness. And now Anthropic is wondering, did a stage like that spontaneously evolve inside a transformer? Well, to find out, they created a tool called the Jacobian lens
[02:21] or J lens, which is basically just a grid of partial derivatives that modify the tokens in the J space. Now, when a word lights up in the J space, it going to output that word. It just means it's keeping that word in mind for
[02:35] whatever reason. Like they asked it, the animal that spins webs has blank legs, and then inside of Claude's brain goo, the word spider lit up it just before it gave a final answer of eight. That seems pretty normal, but then they surgically
[02:48] replace the hidden spider thought with ant, and Claude changed its answer to six. Not because the prompt changed, not because the output was edited, but internal concept. And they also tested language. When Claude is reading a
[03:02] Spanish passage, it internally realizes this is Spanish, but then researchers Claude says it's French, but it still continues outputting perfect Spanish. And that's just weird because it means some skills go through the J space,
[03:16] while others just happen automatically in the basement somewhere. AI bros are conscious, even though Anthropic specifically says in the article none of conscious, but it is fascinating that with enough data and the right linear
[03:30] algebra, a scratch pad for thoughts like this could emerge spontaneously. But conscious until I hear it on a Joe Rogan podcast explaining novel interactions finally comes, humans are in big trouble, but for right now, you need to
[03:44] check out Tracer, the sponsor of today's video. The Tracer just launched a new free and open source desktop app is lets your coding agents talk to each other as walkie-talkies. You can use any of your
[03:57] and it also works with open source models. Like here, I've connected my Claude Code and Codex subs and tracer is now unleashing both on the same project in parallel. They'll ask each other questions, create tickets, and even
[04:11] it's a multiplayer workspace, real humans can jump in and cook alongside the agents. Ever since I started using tracer, building without it it feels needs to keep manually relying on
[04:24] contacts between a bunch of isolated chats. Over 100,000 developers are for free with the link below. This has watching and I will see you in the next one.
⚡ Saved you 0h 04m reading this? Transcribe any YouTube video for free — no signup needed.