AI Links, 3/26/2025

Semi-analysis on robots; Eric Zhao on sampling the answer space; Ethan Mollick's experiment teaming AI with an individual; Shlomo Klapper on AI and humility

Semi-analysis writes,

These will be robotics systems manufacturing more robotics systems, and with each unit produced the cost will be driven down continuously and the quality will improve, only strengthening their production flywheel. This will repeat ad infinitum and as quality inevitably increases it will make it extraordinarily difficult for other countries to compete. Due to the fact that robotics is a general purpose technology, this will have horizontal impacts on all manufacturing sectors and all other currently advantaged industries as well–textiles, electronics, consumer goods, etc. At the moment, the West is caught flatfooted: South Korea and Japan have a birth rate crisis that is throttling their manufacturing capabilities, European industrial sectors are being eaten alive by China and their inability to generate power, and the US is focused on other markets and procuring cheap overseas production, all the while China’s manufacturing capacity has gotten stronger and robotics is catching fire.

Pointer from Moses Sternstein. This would be a situation in which China leads and the U.S. follows.

Eric Zhao writes,

Search over more solutions. This can be done serially, x didn’t work let’s try y, or in parallel, give it a few random shots and see if any work.

…having many solutions allows for easier verification by enabling meaningful comparisons. Models may be poor at identifying errors in their own reasoning, but are known to fare much better if pointed to the location of an error [9]: comparing solutions provides an easy way to exploit this because the diff (disagreement) between two similar solutions is bound to contain any errors not shared by both. This means that a solution’s diffs against a set of other candidate solutions provide a strong signal for where the solution contains errors. This, for example, plays a significant role in boosting performance on MATH where hallucinations are responsible for the majority of Gemini 1.5’s residuals.

Pointer from Alexander Kruel.

Ethan Mollick writes,

When working without AI, teams outperformed individuals by a significant amount, 0.24 standard deviations (providing a sigh of relief for every teacher and manager who has pushed the value of teamwork). But the surprise came when we looked at AI-enabled participants. Individuals working with AI performed just as well as teams without AI, showing a 0.37 standard deviation improvement over the baseline. This suggests that AI effectively replicated the performance benefits of having a human teammate – one person with AI could match what previously required two-person collaboration.

Shlomo Klapper writes,

AI’s greatest flaw isn’t just that it makes mistakes; it’s that it fails to recognize them.

…This isn’t a tech problem—it’s a social one. Our systems—clicks, likes, shares—reward bravado over humility. Yet technology could flip this.

We built AI to imitate us. But perhaps, paradoxically, we can learn from it instead. An AI that models doubt—pausing, weighing, admitting uncertainty—could force us to rethink our own habits of overconfidence. The real challenge isn’t just designing better AI. It’s designing a society that values uncertainty as much as truth.

substacks referenced above: @

Beyond Turing: The Next Test for AI

@

The Cybernetic Teammate

@

AGI Needs Brute-Force Search

@

Links for 2025-03-19

@

Healthcare makes all the jobs . . . but maybe manufacturing too? Other labor market desiderata