<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>RSI Progress updates</title><link>https://rsiprogress.com/</link><description>Source-backed evidence updates. Event dates and tracker changes are distinguished.</description><item><title>Corrected DGM model attribution: Claude 3.5 Sonnet (New) handles self-modification and SWE-bench evaluation; o3-mini handles Polyglot evaluation. Reported scores are unchanged.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-dgm-model-attribution</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Transcribed KernelGen Figure 85; the previous missing-result entry was a tracker omission, not an absent published score.</title><link>https://rsiprogress.com/evidence/kernelgen-1p-astra/</link><guid isPermaLink="false">correct-kernelgen-1p-astra</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Transcribed the explicitly labelled Revised MLE-bench score; preserved the separate historical medal-rate series.</title><link>https://rsiprogress.com/evidence/mle-revised-astra/</link><guid isPermaLink="false">correct-mle-revised-astra</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Replaced an incorrect missing-result label with the source’s published budget-curve evidence. Exact point values remain unavailable in the checked source.</title><link>https://rsiprogress.com/evidence/nanogpt-astra/</link><guid isPermaLink="false">correct-nanogpt-astra</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Replaced an incorrect missing-result label with the source’s published budget-curve evidence. Exact point values remain unavailable in the checked source.</title><link>https://rsiprogress.com/evidence/posttrainbench-lite-astra/</link><guid isPermaLink="false">correct-posttrainbench-lite-astra</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Corrected the tracker’s original RE-Bench metric definition to include normalized numerical scores; no upstream benchmark revision is implied.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-rebench-metric</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Corrected the local Code-Dev metric label and removed a copied run count not stated for Table 6; its score is unchanged.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-code-dev-metadata</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Removed January task-count and scaffold details not stated for the June assessment; retained METR’s non-robustness warning.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-th-sol-metadata</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Assigned the medium-effort configuration its own system identity instead of the all-effort aggregate identity; score unchanged.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-ai4ai-medium-subject</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Qualified the earlier metadata correction: METR discloses the ReAct harness family but not its exact revision, configuration or task count; estimate unchanged.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-metr-react-family</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Preserved the disclosed 41-bug and six-alignment-task components while removing an unsupported assertion that Figure 84 averaged exactly 41 tasks; scores unchanged.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">correct-debugging-population</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Cleared a metadata field that repeated the standard-error amplitude as if it were a confidence level; printed values and reported standard errors are unchanged.</title><link>https://rsiprogress.com/evidence/pb-o3-result/</link><guid isPermaLink="false">correct-standard-error-level</guid><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><description>corrected result</description></item><item><title>Indexed: Claude Opus 5.5 System Card</title><link>https://rsiprogress.com/research/opus55-card/</link><guid isPermaLink="false">indexed-opus55-card</guid><pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Measurements for understanding the pace of AI development inside frontier labs</title><link>https://rsiprogress.com/research/automation-measurements/</link><guid isPermaLink="false">indexed-automation-measurements</guid><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Added the source-attributed B0–L5 framework and historical domain assessment.</title><link>https://rsiprogress.com/research/last-ai-built-by-humans/</link><guid isPermaLink="false">update-autonomy-framework</guid><pubDate>Thu, 10 Sep 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: GPT-6 Astra System Card</title><link>https://rsiprogress.com/research/astra-card/</link><guid isPermaLink="false">indexed-astra-card</guid><pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed four exact comparator labels from the Astra-card debugging figure; this is historical evidence, not a new model release.</title><link>https://rsiprogress.com/evidence/debug-gpt6-sol-card/</link><guid isPermaLink="false">backfill-debugging-figure84</guid><pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: AI4AI-Bench: Benchmarking LLM Agents in Algorithmic Design for Recursive Self-Improvement</title><link>https://rsiprogress.com/research/ai4ai-paper/</link><guid isPermaLink="false">indexed-ai4ai-paper</guid><pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed the paper’s separate Claude Opus 5 medium-effort result, distinct from its across-effort mean.</title><link>https://rsiprogress.com/evidence/a4-opus-medium-mean/</link><guid isPermaLink="false">backfill-ai4ai-opus-medium</guid><pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Gemini 3.7 Flash Frontier Safety Framework Report</title><link>https://rsiprogress.com/research/gemini37-report/</link><guid isPermaLink="false">indexed-gemini37-report</guid><description>newly indexed historical evidence</description></item><item><title>Indexed METR’s conditional, non-robust GPT-5.6 Sol time-horizon estimate with its uncertainty.</title><link>https://rsiprogress.com/evidence/th-1-1-sol-gpt56/</link><guid isPermaLink="false">backfill-th11-gpt56-sol</guid><pubDate>Fri, 26 Jun 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: We are Changing our Developer Productivity Experiment Design</title><link>https://rsiprogress.com/research/productivity-redesign/</link><guid isPermaLink="false">indexed-productivity-redesign</guid><pubDate>Tue, 24 Feb 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed late-2025 raw cohort estimates, preserving METR’s selection-bias warning.</title><link>https://rsiprogress.com/evidence/productivity-returning/</link><guid isPermaLink="false">backfill-productivity-late25-cohorts</guid><pubDate>Tue, 24 Feb 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Added Google’s February 2026 numerical RE-Bench assessment.</title><link>https://rsiprogress.com/evidence/gemini31-rebench-score/</link><guid isPermaLink="false">backfill-gemini31-rebench</guid><pubDate>Thu, 19 Feb 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Time Horizon 1.1</title><link>https://rsiprogress.com/research/time-horizon-revision/</link><guid isPermaLink="false">indexed-time-horizon-revision</guid><pubDate>Thu, 29 Jan 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Completed the printed TH1/TH1.1 comparison-table rows omitted from the tracker’s historical snapshot.</title><link>https://rsiprogress.com/evidence/th-1-0-th-codexmax/</link><guid isPermaLink="false">backfill-th11-comparison-table</guid><pubDate>Thu, 29 Jan 2026 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity</title><link>https://rsiprogress.com/research/productivity-trial/</link><guid isPermaLink="false">indexed-productivity-trial</guid><pubDate>Thu, 10 Jul 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Recent Frontier Models Are Reward Hacking</title><link>https://rsiprogress.com/research/reward-hacking/</link><guid isPermaLink="false">indexed-reward-hacking</guid><pubDate>Thu, 05 Jun 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents</title><link>https://rsiprogress.com/research/dgm-paper/</link><guid isPermaLink="false">indexed-dgm-paper</guid><pubDate>Thu, 29 May 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms</title><link>https://rsiprogress.com/research/alphaevolve-report/</link><guid isPermaLink="false">indexed-alphaevolve-report</guid><pubDate>Wed, 14 May 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: PaperBench: Evaluating AI’s Ability to Replicate AI Research</title><link>https://rsiprogress.com/research/paperbench-paper/</link><guid isPermaLink="false">indexed-paperbench-paper</guid><pubDate>Wed, 02 Apr 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed omitted original PaperBench and distinct Code-Dev results from the 2025 paper.</title><link>https://rsiprogress.com/evidence/pb-deepseek-result/</link><guid isPermaLink="false">backfill-paperbench-original-tables</guid><pubDate>Wed, 02 Apr 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed the paper’s three-paper agent result and human best-of-three reference with different time accounting; no upstream benchmark change.</title><link>https://rsiprogress.com/#activity</link><guid isPermaLink="false">backfill-paperbench-human-subset</guid><pubDate>Wed, 02 Apr 2025 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: Evaluating frontier AI R&amp;D capabilities of language model agents against human experts</title><link>https://rsiprogress.com/research/rebench-report/</link><guid isPermaLink="false">indexed-rebench-report</guid><pubDate>Fri, 22 Nov 2024 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed: MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering</title><link>https://rsiprogress.com/research/mle-paper/</link><guid isPermaLink="false">indexed-mle-paper</guid><pubDate>Wed, 09 Oct 2024 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item><item><title>Indexed original MLE-bench GPT-4o scaffold comparisons from the 2024 paper.</title><link>https://rsiprogress.com/evidence/mle-4o-mlab-result/</link><guid isPermaLink="false">backfill-mle-original-scaffolds</guid><pubDate>Wed, 09 Oct 2024 00:00:00 GMT</pubDate><description>newly indexed historical evidence</description></item></channel></rss>