Savlar4 sources healthy · latest episode 15 Oct 2026
Speaker

Thomas Larsen

Guest on Machine Learning Street Talk (MLST). 166 claims across 1 episode, Sept 2026.

166claims, each timestamped
4 of 12signals on the brief
0verified changes of mind
33open 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 alignment is probably not going to be determined by a broad, endorsed democratic process that aggregates everyone's values. no date given8 Sept 2026pending
Thomas Larsen expected reality to diverge much more from the AI 2027 scenario than it had by September 8, 2026. no date given8 Sept 2026pending
The course of the war will not unfold exactly as expected. no date given8 Sept 2026pending
The authors hope that war games, detailed scenario writing, and criticism will reveal unrealistic or poorly modeled parts of their scenarios and help them correct their thinking. no date given8 Sept 2026pending
The authors hope that exposing enough of their work will elicit feedback from reality or sufficiently realistic simulations that provides useful information. no date given8 Sept 2026pending
Rising awareness of highly capable AIs and their transformative importance could motivate people to race to control AGI or superintelligence. no date given8 Sept 2026pending
Companies will eventually prefer AI labor over human labor, as AI reaches human-level capability; the speaker presents this as a prediction based on there being no fundamental barrier to that capability. In a cluster of 15 →no date given8 Sept 2026pending
AIs will eventually reach human-level capability, as the speaker expects this to happen because he sees no fundamental barrier preventing it. In a cluster of 15 →no date given8 Sept 2026pending
AI capabilities will probably continue to expand, driven partly by increases in the amount and variety of reinforcement learning. In a cluster of 15 →no date given8 Sept 2026pending
AI capabilities will expand across problem types and degrees of verifiability until they encompass the tasks humans can perform. In a cluster of 15 →no date given8 Sept 2026pending
The guest believes AI agents will probably become able to do everything, or at least everything that really matters. In a cluster of 15 →no date given8 Sept 2026pending
AI, compute, and robotics capabilities could greatly accelerate and transform the world, even if occupational licensing blocks AI from some legal and other jobs. no date given8 Sept 2026pending
AI will probably progress beyond human-level cloud workers and eventually become superhuman. In a cluster of 15 →no date given8 Sept 2026pending
The speaker thinks an AI trained to rank very highly across many skills could be extraordinarily capable, while it would be extremely rare or nearly impossible for one human to possess all those skills at once. no date given8 Sept 2026pending
Machines will eventually be able to perform anything the human brain can do. no date given8 Sept 2026pending
Achieving the discussed level of AI will require numerous algorithmic improvements beyond existing models. no date given8 Sept 2026pending
AI loss of control will be one of the biggest problems on the horizon. no date given8 Sept 2026pending
AI-driven concentration of power will be a major problem on the horizon. In a cluster of 4 →no date given8 Sept 2026pending
The speaker predicts that competition over AI could lead to conflict, potentially including a literal World War Three. no date given8 Sept 2026pending
Countries that lose the AI race may realize that the winners will extremely disempower them. no date given8 Sept 2026pending
AI systems capable of building bioweapons may be misused if they become very cheap, broadly diffused, and open source. no date given8 Sept 2026pending
The speaker hopes human-level AIs will be capable enough to help solve the identified problems. no date given8 Sept 2026pending
AI alignment will ultimately need to be solved. In a cluster of 15 →no date given8 Sept 2026pending
If AIs become capable of subverting all control measures and attempt to harm humans, control measures will fail. no date given8 Sept 2026pending
Sufficiently capable AIs would find ways to circumvent the control systems people implement. no date given8 Sept 2026pending
In the speaker's AI 2040 scenario, people will rely almost entirely on control when AI capabilities are around the human range, roughly from 2030 to 2040. 1 Jan 2030 – 31 Dec 20408 Sept 2026pending
In the speaker's scenario, control would fail against wildly superhuman AI if that AI were misaligned, because it could completely subvert control measures. no date given8 Sept 2026pending
In the speaker’s AI 2040 scenario, people use human-level AI and control from 2030 to 2040 while advancing alignment, and by 2040 alignment progress is sufficient to scale to wildly superhuman AI that relies on alignment rather than control. In a cluster of 15 →1 Jan 2030 – 31 Dec 20408 Sept 2026pending
Aligning AI models will probably become harder as the models grow more sophisticated. In a cluster of 4 →no date given8 Sept 2026pending
AI adaptation to evaluation setups will increasingly make evaluation results diverge from the underlying quality being measured. In a cluster of 3 →no date given8 Sept 2026pending
The speakers hope that greater transparency and public understanding of AI companies’ activities will reduce the burden on regulators. no date given8 Sept 2026pending
The speaker expects society, academia, and nonprofits to be able to respond if catastrophically or existentially unsafe AI activity is occurring. no date given8 Sept 2026pending
Under Plan A's transparency principle, the public would be able to participate in informed scientific debate about AI. no date given8 Sept 2026pending

Appearances

Claims per episode