B21 digest: 2025-01-14 to 2025-02-25#

What this batch is#

Fourteen posts from early 2025 (Stargate, the rescinded Biden AI Executive Order, DeepSeek R1, the International AI Safety Report, Antiqua et Nova, Deep Research, Evo 2). Six posts are Modem Futura promotions and are set aside, as the user asked. Two posts contain a lot of AI text: “AI at a Crossroads” was written entirely by ChatGPT, and “The Artisanal Intellectual” is a Deep Research draft that Maynard line-edited. In those, only his framing prose counts. The weight falls on AI and scholarship, with one strong risk-and-governance post (Evo 2) and one risk-methods post (WEF).

Main ideas#

1. Human-only scholarship may become an anachronism (2025-02-04, 2025-02-09, 2025-02-16). This is the batch’s dominant thread and its most important new position. After using Deep Research, he suspects that reasoning research agents “will eventually make non AI-augmented scholarship and research look intellectually limited and somewhat quaint”. He says a Deep Research “dissertation” is “frighteningly close” to the real thing, and that we “will have to critically rethink the purpose and value of a non-AI augmented PhD”. Institutions will soon need these tools just to “get into” the research game, “And if you find that a little disconcerting, you probably should.” His case rests on cross-disciplinary synthesis: the AI goes past “the restrictive training and perspective that comes with disciplinary boundaries”. This ties the claim to his long-standing complaint that single disciplines are too narrow for technology transitions. He is candid about the flaws: citations, sources, discernment, originality and reliability.

2. The artisanal intellectual (introduced 2025-02-09, developed 2025-02-16). He coins it in a footnote. AI will “relegate human-only research to a class of artisanal intellectualism where the primary purpose is the provenance and process, not the product.” The show notes for the 2025-02-11 podcast episode define it as “someone who thinks without using AI”. By 2025-02-16 the meaning has shifted: the artisanal intellectual is “me in this context”, the human craftsperson who checks, judges and builds on AI output. The concept is unsettled and can mean three things: a niche, prestige-by-provenance mode of scholarship; a stance of refusing AI; or a craft of human judgement inside augmented work. He calls it “a bit of a throwaway” that grew on him. The co-written article adds deskilling and a possible new inequality (only well-resourced scholars can afford to work artisanally); these are endorsed, not authored.

3. What AI is: emulated understanding, a symbol-language engine, but a real step change (2025-01-26, 2025-02-09, 2025-02-23). He holds two views together: - Deflationary about minds. ChatGPT has only a “simulated understanding of the world” and lacks “spatial intelligence”. AI cannot write a real dissertation because that “would require independent intent and understanding”. Evo 2 is “a DNA-based stochastic parrot”, and text models “emulate” understanding without being “grounded” in it. - Inflationary about capability. “the generative AI of today is most definitely not the generative AI of 2022.” Reasoning agents are “PhD-level”, and Evo 2’s ability to “parrot” biology “far exceeds anything humans are capable of”.

His unifying frame is language. Text, code and DNA are all symbol sequences tied to functional outcomes. “conversational coding” means “creating functional machines through conversations with AI”. This language-centred view of AI is worth tracking.

4. Risk, governance and the political turn (2025-02-23). The Evo 2 post is the batch’s governance statement: - Asilomar 1975. He uses it as a literal and structural precedent (“a landmark in establishing the foundations of responsible and beneficial genetic manipulation”). - The developers’ choices. He praises the team for leaving pathogen genomes out of training and for red-teaming. - Wider consequences. He argues that unintended consequences go “way beyond harmful viruses” and need cross-disciplinary transition expertise. - The political shift. He names it directly: responsible AI is “going out of fashion at lightening speed” as “permissionless innovation” takes its place, in an AI game of “go fast and break things in the hope that someone else will clean up the mess.” - Continuity. Permissionless innovation stays a critical term, as in earlier batches; what is new is his sense that it is winning.

5. Expert perception of risk underrates AI (2025-01-19). In the WEF risk rankings, AI sits 32nd over two years and 6th over ten. He reads this not as sanity but as mainstream experts who “do not grasp how disruptive the technology may turn out to be”. Aggregated expert opinion “tend[s] to regress to the mean”, which guards against speculative risk but “devalue[s] risks that are poorly understood by a broad base of mainstream experts”. The disruptions will be ““I told you so” moments” for insiders and “blindsides” for others. He also frames AI as an enabling technology whose risks appear under other headings, such as misinformation, which ranks 4th. His PhD sub-question on “thinking at the edge of the distribution” makes the same point.

6. Writing as craft and identity (2025-01-30). Publishing a ChatGPT-written analysis, he says AI “can write better than me, faster than me”. He holds that his writing’s value lies in “the very human piece of me”, idiosyncrasies and flaws included. He breaks his rule because the insights matter “irrespective of whom or what wrote them”. His prompt contains a rare self-description: not “polarizing … preachy … an ideology or an agenda”, with “human wellbeing and flourishing at the heart of my work.”

Concepts appearing#

Artisanal intellectual; human-only versus AI-augmented scholarship; process versus product; reasoning research agents; navigating advanced technology transitions (in his prompts: “new models, frameworks etc” for “transformative, convergent and synergistic technologies”); polycrisis; conversational coding; simulated understanding; DNA as language; permissionless innovation versus responsible AI; Asilomar; regression to the mean in expert risk perception; AI as an enabling technology; writing as craft.

New or changed#

• Practice. By his own account, he publishes a wholly AI-written analytical article for the first time (“I’ve resisted doing this for so long”). AI-generated scholarship and a co-written article follow, all fully disclosed. He writes about the tool in unusually relational terms (“synthetic research partner”, “felt bereft”).
• Position. This is his strongest statement so far that AI augmentation of intellectual work is inevitable and desirable (“hard to imagine a future of effective intellectual labor that is not AI-enhanced”). He even calls his own seven o1-pro papers not worth pursuing. The concerns (deskilling, inequality, loss of craft) are acknowledged but secondary here, and are mostly voiced in the co-written text rather than his own.
• Governance mood. His sense that responsible AI is losing ground to permissionless innovation is new and explicit.
• Continuity. His WEF critique carries on from 2024. This time he says clearly that experts are underrating AI, not merely failing to foresee it.

Most important posts#

1. 2025-02-23 evo-2-dna-ai: risk, Asilomar, permissionless innovation, DNA as language, and the nature of AI.
2. 2025-02-09 can-ai-write-your-phd-dissertation: the origin of the artisanal intellectual, the limits of AI (no intent or understanding), and rethinking the PhD.
3. 2025-02-04 openai-deep-research-ai-scholarship: human-only scholarship as anachronism, plus his own framing of advanced technology transitions in his prompts.
4. 2025-01-19 wef-global-risks-2025: his methodological critique of expert risk perception and the claim that experts underrate AI.
5. 2025-02-16 the-artisanal-intellectual-in-the-age-of-ai: the concept developed and redefined (hybrid provenance, so use his intro and notes).
6. 2025-01-30 ai-at-a-crossroads: his postscript on writing as craft and his self-description (the body is ChatGPT’s).