Resolution criteria

Each option in this independent multiple-choice market will resolve to YES, NO, or N/A based on whether the specific criteria are met by December 31, 2030, at 11:59 UTC.

Definitions

• “Well-known researcher” means an academic or industrial research scientist with a track record of active publication in their field lasting at least three years who is well-respected in their community OR has an H-index of at least 15. This is a broad subjective definition but is meant to exclude eg. student projects, independent lone wolves who only publish to arXiv, etv. This researcher must not be a collaborator of the original paper authors.

• “Global workspace” means representations describable using J-lens-like techniques OR representations that act functionally similar to the representations described by Anthropic that carry broad domain-general applicability

Elaboration

• Competing frontier lab shows a global workspace on one of their own closed models using similar experimental methods: Resolves to YES if a major closed-source frontier AI lab (such as OpenAI, Google DeepMind, Meta, or xAI) publishes a paper, technical report, or official blog post demonstrating they have mapped a global workspace or J-space equivalent in their frontier closed models using similar Jacobian Lens or related mathematical/experimental interpretability techniques.

• Well-known researcher uses Anthropic’s methods on open-weight models to find novel evidence of a global workspace: Resolves to YES if a well-known researcher publishes novel evidence or extensions of the global workspace (J-space) in open-weight models using the Jacobian Lens methods, and shows its existence with at least similar experiments

• Well-known researcher attempts to reproduce global workspace results on open-weight models: Resolves to YES if a well-known researcher independent from Anthropic publishes a dedicated, independent attempt to replicate the main findings of the paper (either successfully or unsuccessfully) on open-weight models (e.g., Llama, Qwen, Gemma) along with their code and results. (Note: Neel Nanda's preliminary reproduction on Qwen 3.6 27B released alongside the paper on July 6, 2026, does not count because it was produced in private collaboration with the original authors; I feel they could have influenced his cherrypicking decisions in this case).

• Well-known researcher casts public doubt on Anthropic’s claims, methods, or results: Resolves to YES if a well-known publishes a peer-reviewed paper, detailed blog post (e.g., on LessWrong or Hugging Face), or formal commentary expressing substantial methodological skepticism, failure to replicate, or casting doubt on the validity of Anthropic's "J-space" as an analog to a global workspace.

• Paper providing independent evidence of global workspace passes peer review in any major conference: Resolves to YES if an independent (non-Anthropic) research paper presenting evidence of an emergent global workspace in LLMs is accepted for publication at a major AI conference (e.g., NeurIPS, ICML, ICLR, CVPR) by EOY 2030.

If you add your own prop answer, please provide an elaboration of what you mean in the comments.

Background

See Anthropic’s July 6, 2026 paper titled "Verbalizable Representations Form a Global Workspace in Language Models".