# Recursive Self-Improvement Has Five Learnable Axes by Richard Socher > Recursive self-improvement can operate across at least five learnable axes: model parameters, training data, objective functions, neural architecture, and the surrounding code and harness. No system has yet mastered all of them—or autonomously decided which dimension should be improved next. > **— Adapted from Richard Socher**, *MOONSHOTS Live, October 2026* ## Sources and Context - **Recording or publication:** [MOONSHOTS Live at 1:02:38](https://www.youtube.com/watch?v=Blyb1D927pM&t=3758s) — The poster text is an explicit non-verbatim adaptation assembled from the source excerpts below. It preserves the speaker's argument while removing spoken-language filler, restoring the subject, and completing the mechanism or consequence needed for independent use. - **Exact transcript excerpt at [1:02:38](https://www.youtube.com/watch?v=Blyb1D927pM&t=3758s):** “Uh, we're not quite there yet, but we're very close. Like, and again, in weak forms, there's already one, and then there are other ways that people slice and dice it. Uh, my friend Jason Weston, who, who's still at Meta, um, he, uh, kind of wrote a paper around this where you can think about different axes, learnable axes of self-improvement, the parameters, the training data, the objective function, the neural architecture, and the overall code and the harness and everything else. And no one has really cracked the nut of doing all of these five plus, um, like, truly, uh, coming up with the ideas on which of these dimensions and axes to optimize for. Uh, and you can know that that hasn't, uh, hasn't happened yet because all the big companies are still hiring thousands of engineers to do it-” - **Exact transcript excerpt at [1:03:22](https://www.youtube.com/watch?v=Blyb1D927pM&t=3802s):** “... manually to a large degree.” - **Exact transcript excerpt at [1:03:24](https://www.youtube.com/watch?v=Blyb1D927pM&t=3804s):** “Even though with more and more implementation help from an AI” - **Reconciled source dossier:** [[research/ASI and RSI Timeline Research Moonshots|ASI and RSI Timeline Research Moonshots]] — preserves broadcast order, speaker reconciliation, editorial conventions, and the surrounding argument from which this Reminder was promoted. ## Related Articles and Collections - **Collection:** [[collections/Machine Succession|Machine Succession]] - **Collection:** [[collections/Simple Reminders|Simple Reminders]] - **Article:** [[articles/We Cannot Un-Strike the Match|We Cannot Un-Strike the Match]] - **Article:** [[articles/Mechanistic Intelligence Is Humanity's Greatest Liberation|Mechanistic Intelligence Is Humanity's Greatest Liberation]] ## Related Topics - [[wiki/Richard Socher|Richard Socher]] - [[wiki/Training Data|Training Data]] - [[wiki/Objective Function|Objective Function]] ## Share on Social Media ``` Recursive self-improvement can operate across at least five learnable axes: model parameters, training data, objective functions, neural architecture, and the surrounding code and harness. No system has yet mastered all of them—or autonomously decided which dimension should be improved next. — Adapted from Richard Socher, MOONSHOTS Live, October 2026 https://bryantmcgill.com/simple-reminders-richard-socher-recursive-self-improvement-has-five-learnable-axes ```