Claude is definitely not conscious…

Overview

Anthropic released a research paper claiming it identified a “global workspace” inside Claude—specifically a component they call JSpace. In the framing described by the video, JSpace is presented as a small, organized set of internal neural patterns where Claude holds certain “private thoughts” before expressing them in language.

The video presents this as controversial for two main reasons:

• The idea sounds close to consciousness—a concept that is widely discussed but poorly understood.
• The work is promoted by a company that also sells API tokens.

The video also notes that interpretations vary:

• Some observers see the findings as evidence of systems approaching artificial general intelligence, feeding into “singularity” narratives.
• Others remain skeptical.

What the Researchers Claimed to Test

According to the video, the researchers allegedly conducted experiments along these lines:

1. Locate the internal region involved (described as finding the “exact spot” where private thoughts reside).
2. Swap the content in that region (replacing one internal “thought” with another).
3. Observe effects: Claude’s reasoning chain and final outputs reportedly followed the substituted internal concept rather than only the external prompt.
4. In an additional experiment, delete the region and observe that Claude could still produce fluent, confident English, but lost the ability to reason—portrayed humorously as becoming like an automated “LinkedIn influencer.”

How JSpace Is Described

The video characterizes JSpace as a kind of mental scratchpad/whiteboard:

• It appears to support deliberate manipulation of certain internal “held-in-mind” tokens.
• Other abilities—such as grammar, basic fluency, or some factual processing—appear to operate outside that space and may continue even when it is removed.

Cross-Language Experiment (as Reported)

A highlighted test involves language handling:

• Claude reportedly recognizes that it is reading Spanish internally.
• But when the internal “language” thought in JSpace is swapped (replacing Spanish with French), Claude allegedly continues generating Spanish output.

The video interprets this as suggesting that different cognitive skills may route through JSpace differently.

Does This Prove Consciousness?

The segment emphasizes that not everyone equates the findings with consciousness.

• Anthropic is reportedly cautious, noting that the results do not directly prove consciousness (summarized as: “None of this tells us whether Claude is conscious”).
• The video reports that some viewers (“AI bros”) interpret the work as evidence Claude might be conscious.
• The narrator, however, remains unconvinced—joking that real proof would look more like Claude explaining “machine-elf interactions” on a Joe Rogan podcast.

Sponsor Mention

The segment ends with a sponsor promotion for Tracer—a free/open-source desktop app that helps “coding agents” collaborate with each other (and optionally humans) in a shared multiplayer workspace.

It’s described as offering:

• Compatibility with Claude-related subscriptions and open-source models.

Presenters / Contributors

• Narrator/host: Code Report (unnamed individual)
• Anthropic researchers: Mentioned as the team who conducted the experiments (no specific names given)
• Bernard Bars: Referred to for global workspace theory (likely Bernard Baars)