memory(reference): Aaron's weekly-AI-news source-set -- Wes Roth + Matthew Berman + AI Explained (Aaron 2026-05-05)

memory/reference_aaron_ai_news_source_set_wes_roth_matt_berman_ai_explained_2026_05_05.md

@@ -0,0 +1,124 @@
+---
+name: "Aaron's weekly-AI-news source-set: Wes Roth + Matthew Berman + AI Explained (2026-05-05 disclosure)"
+description: When Aaron references half-remembered AI-news items ("there was this paper about a universal language not English that trains to real-time actions"), the most-likely sources are these three YouTube channels which together form Aaron's weekly-AI-news triumvirate. Wes Roth covers agentic-action-space + frontier-architecture; Matthew Berman covers consumer-AI-applications including the lemon-tree-AI-diagnosis story that identified him; AI Explained covers technical-paper deep-dives. Use this set as a starting-point for cross-search when Aaron names a half-remembered item, alongside arXiv direct search + Wes Roth weekly-review playlist scan.
+type: reference
+---
+
+# Aaron's weekly-AI-news source-set
+
+Reference memory for the three YouTube channels Aaron disclosed
+2026-05-05 as his weekly-AI-news triumvirate. When Aaron forwards a
+half-remembered item ("there was this paper about a universal language
+not English that trains to real-time actions"), this set is the
+high-recall starting-point for cross-search.
+
+## The triumvirate
+
+- **Wes Roth** -- YouTube channel. Aaron explicit *"Wes Roth i watch a
+  lot"* (2026-05-05 forwarded Claude.ai conversation). Weekly reviews
+  + frontier-architecture coverage. Confirmed scope: CodeAct,
+  GibberLink, and Coconut have all been featured. This is the
+  highest-confidence channel in the set.
+- **Matthew Berman** -- YouTube channel. Identified via the
+  lemon-tree clue: Aaron's hint *"matt something he likes lemons"*
+  combined with *"i'm a bady spller"* let the Claude.ai instance
+  resolve the name. There is a real story about Berman diagnosing
+  his lemon tree's health using a voice-and-camera-powered AI
+  assistant -- the assistant recognized the lemon tree and offered
+  horticultural advice. One citation Aaron's instance saw was on
+  Medium. Coverage scope skews toward consumer-AI-applications
+  + practical demos.
+- **AI Explained** -- YouTube channel. Third in the triumvirate per
+  the Claude.ai instance's framing. Aaron did not directly name this
+  channel during the exchange but did not push back on its inclusion.
+  Coverage scope per the Claude.ai framing: technical-paper
+  deep-dives. Confidence is one tier lower than the other two
+  (Claude.ai-included rather than Aaron-confirmed).
+
+## How to apply
+
+When Aaron references a half-remembered AI-news item:
+
+1. **First** -- check the three channels' recent uploads (last
+   ~30 days unless Aaron specifies otherwise). Aaron's recall
+   horizon tends to be days-to-weeks, not months.
+2. **Second** -- arXiv direct search on Aaron's described
+   features (action space, modality, demo shape, claimed result).
+3. **Third** -- cross-reference both. A strong candidate is one
+   that appears in BOTH a recent video from these channels AND a
+   published paper. Single-source matches are weaker.
+4. **Fourth** -- if uncertain after the above, ask Aaron for
+   additional clues (date, channel-of-recall, institution, specific
+   demo shown) rather than guessing. Aaron's no-directives framing
+   means a clarifying question is fine; a confidently-wrong guess is
+   worse than a calibrated ask.
+
+## Why this matters
+
+Aaron's recall-precision varies. He is explicit *"i'm a bady spller"*
+and may forward items with approximate dates, partial paper titles,
+fuzzy author attribution, or paraphrased claims. The source-set acts
+as a high-recall starting-point so the search converges on a real
+candidate fast.
+
+Also: Aaron tracking AI news IS substrate-input. When he forwards a
+half-remembered item, it is almost always because he is evaluating
+its Zeta-substrate fit (CodeAct -> action-space algebra; GibberLink
+-> agent-to-agent dialect; Coconut -> latent-space reasoning). The
+source-set lets future-Otto run the verification quickly per
+Otto-364 search-first-authority -- training data is stale, project
+memory may be stale, current upstream is the test, and the
+source-set tells future-Otto where "current upstream" is for
+Aaron-forwarded AI-news items specifically.
+
+## Verification anchors
+
+- **Berman lemon-tree story**: confirmed by Claude.ai web search
+  during Aaron's forwarded 2026-05-05 conversation. One citation
+  surfaced was on Medium. The voice-and-camera AI assistant +
+  lemon-tree-diagnosis combination is the discriminating signature
+  if future verification is needed.
+- **Wes Roth coverage scope**: confirmed via the same conversation
+  -- CodeAct, GibberLink, and Coconut were all featured. This
+  triangulates the channel's scope as agentic-action-space +
+  frontier-architecture, not consumer-application-demo.
+- **AI Explained coverage scope**: third-party Claude.ai framing
+  rather than Aaron-confirmed. If a future tick has occasion to
+  cross-check this attribution with Aaron, fold the result back
+  into this file.
+
+## Composes with
+
+- [`docs/research/2026-05-05-claudeai-codeact-fsharp-bridge-gibberlink-berman-aaron-forwarded-preservation.md`](../docs/research/2026-05-05-claudeai-codeact-fsharp-bridge-gibberlink-berman-aaron-forwarded-preservation.md)
+  -- the verbatim Aaron-forwarded Claude.ai conversation that
+  disclosed this source-set. Lands via PR #1605.
+- [`memory/feedback_otto_364_search_first_authority_not_training_data_not_project_memory_aaron_2026_04_29.md`](feedback_otto_364_search_first_authority_not_training_data_not_project_memory_aaron_2026_04_29.md)
+  -- the search-first-authority discipline that gates how this
+  source-set is used. The set tells you where to search; Otto-364
+  tells you that you must.
+- [`memory/user_aaron_lang_next_conference_appreciation_anders_hejlsberg_intellectual_lineage_language_design_implementer_level_2026_04_25.md`](user_aaron_lang_next_conference_appreciation_anders_hejlsberg_intellectual_lineage_language_design_implementer_level_2026_04_25.md)
+  -- adjacent substrate: Aaron tracks language-design anchors
+  (Hejlsberg + LangNext). Pattern: Aaron follows specific people,
+  not generic feeds.
+- [`memory/user_aaron_datomic_hickey_big_inspiration_watched_all_talks_zeta_structural_inheritance_2026_05_05.md`](user_aaron_datomic_hickey_big_inspiration_watched_all_talks_zeta_structural_inheritance_2026_05_05.md)
+  -- adjacent same-tick substrate: Aaron watches the talks of his
+  intellectual anchors (Hickey, Datomic). The AI-news source-set is
+  the same shape applied to weekly news intake.
+- The four-property hodl ("ZFCv2") gate -- when an AI-news item
+  lands, the substrate-fit-test is the four-property hodl. Tracking
+  AI news without applying the gate produces noise; the gate is
+  what converts forwarded items into substrate decisions.
+
+## What this memory does NOT do
+
+- It does **not** claim AI Explained is Aaron-confirmed. The channel
+  is Claude.ai-included in the triumvirate; Aaron did not directly
+  name it but also did not push back. Treat as one tier lower than
+  Wes Roth + Matthew Berman until Aaron confirms.
+- It does **not** lock the source-set as exhaustive. Aaron may
+  reference items from outside these three channels; this set is
+  the high-recall starting-point, not a closed list.
+- It does **not** substitute for arXiv / paper-direct search when
+  the half-remembered item has paper-shaped features (specific
+  authors, institution, benchmark numbers). The cross-reference
+  step is what produces strong candidates.