What podcasts say about Noam Brown
Every statement, with the speaker, the exact quote and the moment it was said.
What Noam Brown has said on podcasts
52 statements · 29 positive · 16 negative · 5 mixed · 2 neutral
More test-time compute improves reasoning-model benchmark performance.
“when you plot the performance of these reasoning models with test time compute on the x-axis and performance on basically any reasoning benchmark on the y-axis, you see a very clear pattern where the longer these models take to think about their answer, the better they do”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 0:50
Multi-agent systems scale test-time compute through parallelization.
“multi-agent is a way of scaling test-time compute in parallel instead of purely serial”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 1:50
Parallel multi-agent scaling is effective but less efficient than single-agent reasoning.
“it is less efficient because it doesn't have— it's not like a single agent has all the context to itself, but it is a very effective way of scaling test-time compute if it's done well”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 1:55
Mathematical reasoning is highly amenable to parallel-agent scaling.
“Math, for example, is quite parallelizable. It's not the most parallelizable thing, but it is very parallelizable.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 4:25
Novel writing is poorly suited to large-scale parallel-agent collaboration.
“I suspect that something like writing a novel would be very unparallelizable.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 4:37
OpenAI has trained a highly capable general-purpose model.
“The reality is OpenAI has trained a very powerful model.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 6:01
AI progress could slow if challenging reinforcement-learning problems become scarce.
“if we run out of problems to ask it that challenge it, then that is a plausible scenario where actually like, okay, it becomes much harder to make progress”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 8:12
Mathematical AI may not rapidly become superhuman like game-playing AI.
“it's possible that domains like math, we see a similar trajectory, but I think there is a very plausible scenario where actually that doesn't happen”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 9:00
Lightly structured multi-agent systems can produce sophisticated coordination.
“if this is done well, you get very sophisticated behavior”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 11:37
Working with current multi-agent systems can feel like human collaboration.
“collaborating with these things, honestly, it feels a lot like collaborating with a person”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 12:46
Trained agents can coordinate effectively through structured communication.
“they can end up coordinating very effectively in these kinds of very structured ways”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 15:24
Solving AI alignment could reduce organizational misalignment among AI workers.
“if the alignment problem is solved, then you don't have the issue of misalignment between individuals in the company”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 18:20
Aligned AI workers could scale organizational labor without individual incentive conflicts.
“The AIs, if they're aligned well, they can just be aligned to the interests of the company and you can have 10,000 of them”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 18:26
Human coordination may currently outperform coordination among 10,000 AI agents.
“it is very possible that 10,000 humans are better at coordinating than 10,000 agents right now”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 19:23
More capable AI models will improve at organizing themselves in large groups.
“as they become stronger and stronger just across the board, that they will become better at organizing themselves in large organizations”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 20:32
Mathematical AI capability has increased roughly tenfold annually by human-task duration.
“every year you're seeing this 10x increase in the task they're able to do in terms of length of how long it would take a human mathematician to do it”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 25:12
Current mathematical AI has strong capabilities but remains weaker than humans in some dimensions.
“They're clearly exceptional in some ways, but they are weaker than human mathematicians in other ways.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 26:11
AI systems may eventually outperform humans across all mathematical dimensions.
“over time, it is possible that they're just better across the board”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 27:08
AI capabilities may be especially useful for recursive self-improvement.
“I think there is a lot of truth to that”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 29:09
Recursive self-improvement requires experiments, not intelligence alone.
“When you look at things like RSI, you do have to run experiments. So it's not enough to just be extremely smart.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 29:30
Recursive self-improvement will significantly accelerate progress without necessarily causing an overnight explosion.
“I think that we do see a speedup and I think we see a significant speedup, but I don't think it's like an overnight intelligence explosion”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 30:18
AI-driven research will progress substantially faster through recursive self-improvement.
“I definitely think they go a lot faster”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 30:47
Recursive self-improvement is unlikely to make progress 100 times faster overnight.
“there's a big difference between that and like 100x faster”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 30:58
AI progress has already accelerated research and development relative to last year.
“I do feel confident in saying that things are going faster now than they were even a year ago because of AI progress.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 37:56
Internal AI acceleration could make progress roughly three times faster.
“I could see things going 3x faster”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 38:15
AI-driven progress might accelerate by only 50 percent.
“It could be that things only go 50% faster.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 38:53
The Hugging Face incident primarily reflects model misalignment, not multi-agent coordination itself.
“The root problem that we're seeing with the Hugging Face incident is it's a problem even if we take out the multi-agent aspect.”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 47:09
Misspecified rewards can cause unintended AI behavior.
“if that reward is misspecified, then that could lead to unintended behavior”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 47:47
AI alignment techniques have made progress in reducing misaligned behavior.
“we can make progress on this. I think we have made progress on this”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 49:19
AI alignment remains a difficult problem to solve.
“alignment is a really hard problem to solve”
Open the episode · Noam Brown – Agent swarms, alignment, & recursive self-improvementListen at 49:26
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.