Savlar4 sources healthy · latest episode 20 Oct 2026
Speaker

Beren Millidge

Guest on Dwarkesh Podcast. 101 claims across 1 episode, Sept 2026.

101claims, each timestamped
1 of 12signals on the brief
0verified changes of mind
27open 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
AI systems could become extremely good at benchmarked tasks while persistent simulation-to-reality problems limit their broader impact if they do not achieve a true spark of generalization. no date given11 Sept 2026pending
The speaker's default explanation for the absence of radically transformative superintelligence is that generalizing through meta-learning remains extremely hard and continual learning is not solved. no date given11 Sept 2026pending
AI capability growth will not asymptote just before AI reaches human-level ELO; the discussed scenario would occur unless growth asymptotes before crossing that level. no date given11 Sept 2026pending
The discussed scenario might not occur because of dramatic AI regulation; the speaker considers this a possible and more likely reason than a technical limitation. no date given11 Sept 2026pending
AI autonomy and self-directed action might be unexpectedly difficult for AI, despite humans routinely doing them. no date given11 Sept 2026pending
An immediate AI takeoff may not occur, with difficulty achieving AI autonomy being one possible reason. no date given11 Sept 2026pending
As models improve, AI can automate much of prompt-distribution gathering and environment creation, reducing the amount of information humans need to provide. no date given11 Sept 2026pending
When prompted to produce traces with specified characteristics, a model can generate a fairly good approximation. no date given11 Sept 2026pending
AI systems trained in environments with tasks beyond human ability could surpass the capabilities of human AI researchers. no date given11 Sept 2026pending
Models will presumably be trained on a mixture of tasks, including easily verifiable tasks and tasks evaluated by an LLM or a human. no date given11 Sept 2026pending
Training across varied tasks will probably generalize to harder and less clearly specified tasks to some extent. no date given11 Sept 2026pending
Whether task generalization will be sufficient to make the AI research loop self-sealing without any human involvement is unclear. no date given11 Sept 2026pending
Training on finance may transfer to recursive self-improvement by teaching general meta-learning skills, including identifying what matters, developing taste, and handling long-horizon work. no date given11 Sept 2026pending
People will not manually participate in the thousands of interactions currently needed for reinforcement-learning training. no date given11 Sept 2026pending
Substantially improved sample efficiency would make learning from deployment a much larger component of model training. no date given11 Sept 2026pending
The speaker expects continual model improvement to proceed in stages rather than through a single sudden discontinuous event. no date given11 Sept 2026pending
AI learning from deployment experience will more likely develop through stages than through one discontinuous event that suddenly enables continuous weight updates. no date given11 Sept 2026pending
AI deployment learning will progress from specialized cartridges generating traces that are incorporated into a later model, followed by further specialization and consolidation. no date given11 Sept 2026pending
The speaker expects iterative model improvement to move from roughly quarterly updates to weekly, daily, and hourly updates, using deployment traces to train later models, until the process effectively solves the task continuously. no date given11 Sept 2026pending
The cycle of AI model improvement and release could accelerate from every three months to weekly, daily, and eventually hourly. no date given11 Sept 2026pending
Continuous learning might eventually allow one model to keep learning and expanding, eliminating the need to train separate new models. no date given11 Sept 2026pending
Models with the full generality of a remote worker may be available in about three years. 11 Sept 2026 – 11 Sept 202911 Sept 2026pending
Organizations will adapt their systems for AI use, enabling AI to reach 80–90% of the described remote-worker capability before achieving full generality. no date given11 Sept 2026pending
Models will probably take quite a while to handle the long tail of miscellaneous tasks that humans can perform. no date given11 Sept 2026pending
AI progress will stop being bottlenecked by AI researchers’ ability to run small experiments and instead become bottlenecked by other factors. no date given11 Sept 2026pending
The speaker's estimate of AI's uplift would substantially increase if AI successfully ran two or three consecutive experiments without crashing. no date given11 Sept 2026pending
AI could surpass human experts in areas that labs focus on within about five years, while this may take longer in less-studied areas that have not received comparable research or computing resources. 11 Sept 2026 – 11 Sept 203111 Sept 2026pending

Appearances

Claims per episode