Pustam | पुस्तम | পুস্তম🇳🇵<p>"Random sampling works better than you think: Gemini 1.5 = o1. The secret? Self-verification magically gets easier with scale."</p><p>Thinking for longer (e.g. o1) is only one of many axes of test-time computing. In a new Google paper, the authors instead focus on scaling the search axis.</p><p>By just randomly sampling 200 responses and self-verifying, Gemini 1.5 (an ancient early 2024 model!) beats o1-Preview and approaches o1. This is without finetuning, RL, or ground-truth verifiers.</p><p>"This was surprising: search is bottlenecked by verification, models are notoriously bad at self-verifying (think hallucinations), and self-consistency doesn't scale. The magic is that self-verification naturally becomes easier at scale! You'd expect that picking out a correct solution becomes harder the larger your pool of solutions is, but the opposite is the case!"</p><p>Read more: <a href="https://eric-zhao.com/blog/sampling" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="">eric-zhao.com/blog/sampling</span><span class="invisible"></span></a></p><p><a href="https://mathstodon.xyz/tags/Sampling" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Sampling</span></a> <a href="https://mathstodon.xyz/tags/Random" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Random</span></a> <a href="https://mathstodon.xyz/tags/Randomness" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Randomness</span></a> <a href="https://mathstodon.xyz/tags/Gemini" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Gemini</span></a> <a href="https://mathstodon.xyz/tags/RandomSampling" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>RandomSampling</span></a> <a href="https://mathstodon.xyz/tags/Stats" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Stats</span></a> <a href="https://mathstodon.xyz/tags/Statistics" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Statistics</span></a></p>