Savlar4 sources healthy · latest episode 21 Oct 2026
Speaker

Zhengyao Jiang

Guest on Machine Learning Street Talk (MLST). 141 claims across 1 episode, Sept 2026.

141claims, each timestamped
0 of 12signals on the brief
0verified changes of mind
15open predictions

Positions over time

Every pair of statements that was compared. Only a verified change counts as a change of mind; rewording, added nuance and different questions do not.

No two statements by this speaker have been compared yet.

Predictions

As stated. A window appears only where the speaker gave one.

ExpectsBySaidStatus
Zheng Yaojiang does not expect recursive self-improvement to result in a technological singularity or intelligence explosion. no date given26 Sept 2026pending
An autonomous research system could improve its research efficiency by directing research toward improving itself. no date given26 Sept 2026pending
The speaker hopes that future research-efficiency improvements will overcome diminishing returns and make the effort-output curve change from concave to convex. no date given26 Sept 2026pending
The team believes adding constraints that encourage simpler solutions, analogous to neural-network regularization, may improve the optimizer. no date given26 Sept 2026pending
Level-three recursive self-improvement will require self-referential loops across all layers when it reaches its inflection point. no date given26 Sept 2026pending
If an agent or harness discovered in the inner loop can be applied to the outer loop, the process can continue without adding another layer. no date given26 Sept 2026pending
The process can run indefinitely, applying infinitely many iterations without adding another layer. no date given26 Sept 2026pending
The team plans to publish a paper, beginning with a PDF version, that credits academic contributions to meta-harness optimization. no date given26 Sept 2026pending
Human detection of reward hacking will become increasingly difficult as agent behavior grows more sophisticated. no date given26 Sept 2026pending
Software engineers may use input-output contracts and behavioral evaluation when working with agent-generated code in the future. no date given26 Sept 2026pending
The speaker believes the problem of repeated, less-interesting ideas probably will not be addressed unless continual learning is solved. no date given26 Sept 2026pending
The goal is for the inner-loop search policy to generalize to the fixed outer loop in the described system. no date given26 Sept 2026pending
The speaker hopes that the inner-loop search policy will generalize to the outer loop. no date given26 Sept 2026pending
The speaker expects that achieving recursive self-improvement will take a long time, even if it is achieved. no date given26 Sept 2026pending
Progress toward recursive self-improvement will probably cause the development curve to bend gradually. no date given26 Sept 2026pending

Appearances

Claims per episode