Page Summary
Active Entries
- 1: Helion details
- 2: "Narrow AGI" this year?
- 3: Tao on coordinate vs coordinate-free math reasoning
- 4: "Aging as a loss of goal-directedness"
- 5: New integrated mode for GPT-4 in ChatGPT+
- 6: Китайский новый год начнётся 10-го февраля
- 7: Automating the Search for Artificial Life with Foundation Models
- 8: "Anatomy of a Formal Proof"
Style Credit
- Style: Neutral Good for Practicality by
Expand Cut Tags
No cut tags


no subject
Date: 2021-04-24 08:19 pm (UTC)1) About reconciling V-values and flat tensors. We have a full-scale well-developed schema, but what about a lightweight compromise? We can just say that "leaves are matrices", that's one possible option. (Would it be fruitful to consider sparse matrices in this context?)
2) Reproducing "perceptual 'illusions' in software is a fruitful topic.
2a) Still want to go from DeepDream effects to actual subjective perception. Conjecture: yes, this is what we "subjectively see" (and in "consensus reality" some "meta-performance and/or equilibrium characteristic" of the network is optimized, while in "alternative perception", perhaps something else is optimized, or an "effective network" is different enough).
[This still does not answer the riddle of what represents subjective perception, and how.]
To be continued