Hutter Prize 1-Level Indirect Influencer Attention Regarding Anthropic's Consciousness Paper

[James Bowery]

This time code of the Moonshots with Peter Diamandis talk show mentions Marcus Hutter and Jurgen Schmidhuber being "20 years" ahead of Anthropic's current paper on their operational definition of "Conscious Thought".  The reference to 2006  is almost certainly the advent of The Hutter Prize for Lossless Compression of Human Knowledge.

Anthropic's operational definition of "Conscious Thought" will, of course, be up there with "Intelligence" for dispute.  See "Measuring Machine Intelligence" when Hutter's PhD student Shane Legg gave that talk at the Singularity Summit just prior to co-founding DeepMind.

That said, I'm unfamiliar with Anthropic's so-called "J-space" ansatz to operationalizing a definition of "Conscious Thought" and am therefore in no position to attest its relevance to the Hutter Prize.

PS: They commented on what they call the "orthogonality" of AGI intelligence and goals.  This is of course true in Hutter's AIXI top down theory of AGI in the sense that the any agent must make decisions relating not only to what to observe but also how much resource should be devoted to lossless compression (pure science given a dataset like enwik9) vs decompression (technology) toward that agent's goals.

PPS: If anyone has an inside track with these guys -- you might try telling them that  the X-Prize Foundation should be backing the Hutter Prize because: 1) The Hardware Lottery provides justification for the Hutter Prize's reliance on general purpose instruction sets. 2) Scaling laws are merely empirical curves -- not laws of nature. 3) Reliance on "validation test data" can no longer be relied upon in an era of such enormous conflicts of interest over "benchmarks" that people will be suspicious of adversarial leaks. 4) The largest increases in market cap in AI have been coding assistants and no coding assistant has come close to the the kind of skills required to compete for a Hutter Prize.

[Matt Mahoney]

Could Anthropic's J-space be incorporated into a Hutter prize entry? Anthropic admits in their paper that they don't completely understand where it comes from. Claude's neural network has at least 88 layers. Back propagation does not work on deep neural networks, so how does that work?

They describe the J-space as the set of neurons that Claude is thinking about as intermediate steps when reasoning that don't make it to the final output. It looks a lot like the way humans think, except that the global workspace or short term memory or transformer attention mechanism or access consciousness can hold several dozen words vs 5 to 9 words in humans. Human short term memory uses neural feedback loops, where in Claude it uses a combination of long propagation chains and an attention mechanism that works like a scratch pad.

Another difference is that LLMs are trained in multiple passes, and then the weights are frozen to avoid leaking data between users. This requires a large context window (200K to 1M tokens) so that it can forget everything you said in the conversation and then play it all back before the next question. Obviously a compressor would work more like the brain or a private LLM, training in a single pass on every input with no need for a context window. This should require a lot less computation.

This should be possible. The current entries based on CMIX don't have multiple layers or a J space. The top LTCB entry NNCP might because it is based on a transformer network running on a GPU, but it is closed source and the paper doesn't give details.

-- Matt Mahoney, mattma...@gmail.com

--
You received this message because you are subscribed to the Google Groups "Hutter Prize" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hutter-prize...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hutter-prize/c1b49367-55e6-4309-9133-70f621036569n%40googlegroups.com.

[Alex Wissner-Gross]

Hi James and Matt,

I've been on your mailing list for years! Happy to chat!

Best,  Alex

---
Alex Wissner-Gross, Ph.D.
www.alexwg.org

To view this discussion visit https://groups.google.com/d/msgid/hutter-prize/CACFhWPOb6M%2BSpLeV%3DBUU8FV_RyF101SB3XuTRS-jsk0i_ndUdw%40mail.gmail.com.

[James Bowery]

On Friday, July 10, 2026 at 5:34:34 PM UTC-5 ale...@physics.harvard.edu wrote:

Hi James and Matt,

I've been on your mailing list for years! Happy to chat!

Thanks for responding!  Indeed, let's chat regarding what you said at this time code:

"this so-called J-space and I can talk if we want a little bit more mathematically about what it actually is"

The most interesting thing to me is the connection to not just compression but to lossless compression.  One of the problems I've run into with differential geometric lossless compression of macrosocial measurements (to a low dimensional manifold bottleneck) is not just the pullback to the original precision (requiring residual offsets to be pulled back as well), but estimating the (positive rather than differential) information content of the model's parameters.  Jacobians sound like they may be they way to approximate the parameter complexity.  However, these are merely statistical models, that I'm using to impute the huge gaps in the data panels available from the US Census so I can do dynamics hence extrapolation rather than mere interpolation (imputation).

Hopefully, this can be related back to language model "J-space" without too much difficulty.