// The confidence cliff in self-improving autoresearch // Autoresearch loops are still quite brittle. Here is a nice paper offering some ins…

// The confidence cliff in self-improving autoresearch // Autoresearch loops are still quite brittle. Here is a nice paper offering some insights into why this might be happening. Self-improving autoresearch loops keep judging their own ideas long after the judgment stops working. In public AutoSOTA logs the fraction of helpful modifications falls from 70% in the first two iterations to 43% by iteration six. On 296 same-baseline modification pairs from 39 paper-derived tasks with outcomes hidden, an LLM judge given rationales but no prior-attempt history reaches 79.5% accuracy where strict consensus returns a verdict. The capability is real before it degrades. Across the full 366-pair benchmark, selective accuracy falls from 82.8% to 56.9% as successful changes accumulate. The judge stays just as willing to decide. Rehearse is a small loop change. Propose several ideas, compare them before execution, run the most promising, and judge against a focused memory of similar past attempts and their outcomes. Late selective accuracy recovers to 83.5%, and across 4,000 budgeted training runs on nanochat, image classification and time-series forecasting the endpoint improves under the same budget. Paper: https://arxiv.org/abs/2607.27687 Track more trending AI papers in our academy: https://academy.dair.ai/

Source: DAIR.AI (X) | 2026-08-04