<?xml version="1.0" encoding="UTF-8" ?>
<rss version="2.0">
    <channel>
      <title>SATs for LLMs</title>
      <link>https://mattdot.github.io/sats4llms</link>
      <description>Last 10 notes on SATs for LLMs</description>
      <generator>Quartz -- quartz.jzhao.xyz</generator>
      <item>
    <title>Liu et al. — Lost in the Middle: How Language Models Use Long Contexts (TACL, 2023)</title>
    <link>https://mattdot.github.io/sats4llms/sources/liu-lost-in-middle-2023</link>
    <guid>https://mattdot.github.io/sats4llms/sources/liu-lost-in-middle-2023</guid>
    <description><![CDATA[ Lost in the Middle: How Language Models Use Long Contexts Authors: Nelson F. Liu et al. ]]></description>
    <pubDate>Wed, 20 May 2026 18:48:04 GMT</pubDate>
  </item><item>
    <title>Grey Dynamics — &quot;Intelligence Failure: What, When, Why and How&quot; (2024)</title>
    <link>https://mattdot.github.io/sats4llms/sources/grey-dynamics-intelligence-failure-2024</link>
    <guid>https://mattdot.github.io/sats4llms/sources/grey-dynamics-intelligence-failure-2024</guid>
    <description><![CDATA[ Intelligence Failure: What, When, Why and How Publisher: Grey Dynamics Canonical URL: greydynamics.com/intelligence-failure-what-when-why-and-how/ Summary A practitioner-oriented overview of intelligence failure as a multi-stage phenomenon. ]]></description>
    <pubDate>Wed, 20 May 2026 17:53:14 GMT</pubDate>
  </item><item>
    <title>analysis-of-competing-hypotheses</title>
    <link>https://mattdot.github.io/sats4llms/concepts/analysis-of-competing-hypotheses</link>
    <guid>https://mattdot.github.io/sats4llms/concepts/analysis-of-competing-hypotheses</guid>
    <description><![CDATA[ Analysis of Competing Hypotheses (ACH) A SAT that identifies all alternative explanations (hypotheses) and evaluates evidence to disconfirm rather than confirm them. ]]></description>
    <pubDate>Wed, 20 May 2026 17:42:45 GMT</pubDate>
  </item><item>
    <title>catalog</title>
    <link>https://mattdot.github.io/sats4llms/catalog</link>
    <guid>https://mattdot.github.io/sats4llms/catalog</guid>
    <description><![CDATA[ Catalog This is the master catalog. Read this first before answering any query or starting any operation. ]]></description>
    <pubDate>Wed, 20 May 2026 17:36:06 GMT</pubDate>
  </item><item>
    <title>suprathermal</title>
    <link>https://mattdot.github.io/sats4llms/entities/suprathermal</link>
    <guid>https://mattdot.github.io/sats4llms/entities/suprathermal</guid>
    <description><![CDATA[ suprathermal GitHub user (github.com/suprathermal) and author of the ACH-Grounding open-source project — a working Python implementation of ACH as an LLM grounding mechanism. ]]></description>
    <pubDate>Wed, 20 May 2026 17:36:06 GMT</pubDate>
  </item><item>
    <title>log</title>
    <link>https://mattdot.github.io/sats4llms/log</link>
    <guid>https://mattdot.github.io/sats4llms/log</guid>
    <description><![CDATA[ Wiki Operations Log Append-only. ]]></description>
    <pubDate>Wed, 20 May 2026 17:36:06 GMT</pubDate>
  </item><item>
    <title>suprathermal-ach-grounding-2024</title>
    <link>https://mattdot.github.io/sats4llms/sources/suprathermal-ach-grounding-2024</link>
    <guid>https://mattdot.github.io/sats4llms/sources/suprathermal-ach-grounding-2024</guid>
    <description><![CDATA[ ACH-Grounding (suprathermal, GitHub) “LLMs hallucinate. Various strategies exist to cope with that. ]]></description>
    <pubDate>Wed, 20 May 2026 17:36:06 GMT</pubDate>
  </item><item>
    <title>H1 — ACH Reduces Confirmation Bias in Evidence Evaluation</title>
    <link>https://mattdot.github.io/sats4llms/synthesis/hypotheses/h1-ach-confirmation-bias</link>
    <guid>https://mattdot.github.io/sats4llms/synthesis/hypotheses/h1-ach-confirmation-bias</guid>
    <description><![CDATA[ Claim The multi-step ACH protocol produces conclusions that better match ground truth than single-prompt “what’s most likely?” queries on identical evidence. ]]></description>
    <pubDate>Wed, 20 May 2026 17:36:06 GMT</pubDate>
  </item><item>
    <title>Intelligence Cycle</title>
    <link>https://mattdot.github.io/sats4llms/concepts/intelligence-cycle</link>
    <guid>https://mattdot.github.io/sats4llms/concepts/intelligence-cycle</guid>
    <description><![CDATA[ The Intelligence Cycle A canonical five-stage framework describing how intelligence is produced, from consumer question to analytic product delivered to decision-makers. ]]></description>
    <pubDate>Wed, 20 May 2026 02:36:07 GMT</pubDate>
  </item><item>
    <title>mirror-imaging</title>
    <link>https://mattdot.github.io/sats4llms/concepts/mirror-imaging</link>
    <guid>https://mattdot.github.io/sats4llms/concepts/mirror-imaging</guid>
    <description><![CDATA[ Mirror Imaging The tendency to assume that adversaries, foreign actors, or other agents think, value, and make decisions the same way the analyst does. ]]></description>
    <pubDate>Wed, 20 May 2026 02:36:07 GMT</pubDate>
  </item>
    </channel>
  </rss>