Slow productivity, reading retention, and why LLMs cannot become conscious

Executive overview

Two threats face anyone trying to live deeply: distraction pulling you away from meaningful work, and overwork burning you out. The slow-and-steady approach — writing a paragraph a day like Joan Didion rather than grinding six hours like Murakami — is sustainable and produces equally good work. But slowness only works when paired with consistency, concentration, and external accountability to beat perfectionism.

The episode also covers practical email and reading systems, academic publishing as an implicit-knowledge game, and a sharp critique of AI consciousness claims.

Slow, steady, and focused beats fast and heavy — as long as you finish.

The slow productivity framework

• Murakami: 4am start, 6 hours writing, 10km run — treats writing like clocking into a factory
• Didion: thinks during the day, writes a paragraph in the afternoon, has a drink, edits — equally acclaimed
• Pace of production does not determine quality of output; what matters is the final work
• Three caveats: steadiness (no-progress days compound into never finishing), concentration (slow does not mean distracted — full attention when working), anti-perfectionism (slow pace makes it easy to over-refine and never ship)
• Beat perfectionism with external stakes: commit to showing work to someone on a fixed date, then move on
• Reframe with "this just has to be good enough — the next one can be great"

Email and communication systems

• Skip out-of-office replies for delays; only ~5% of senders will notice or complain — explain to them individually
• Provide an emergency phone number as a release valve; removes the complaint that you're unreachable
• Project-based email addresses (e.g. projectname@domain.com) shift sender psychology — they feel part of a process, not a one-on-one conversation, and tolerate slower replies
• Email addresses defaulting to individuals is a historical accident from time-shared mainframes, not a design choice
• For inbox zero: drag all email to a processing folder, sort into project subfolders, work oldest-to-newest within one subfolder at a time to reduce cognitive switching
• Track emails requiring a response on a waiting-for list; only add items that are on a critical path

Reading retention methods

• For novels and journalistic nonfiction: retention is not the goal — reading is valuable as an experience in itself
• For nonfiction you want to retain: page-marking — bracket or underline key passages, fold the page corner; re-read marked pages in ~5 minutes to reload a book's ideas
• For deep integration: write a brief summary and personal reaction after each chapter in a document; this cements ideas better than highlighting alone
• Match method to goal: most books need only page-marking; note-taking reserved for books that will shape your own thinking or projects

Academic publishing as implicit knowledge

• Publishing success in academia is not an hours game — it requires knowing exactly what referees and venues want
• Elite institutions succeed partly because students learn the unwritten rules: which venues matter, what signals rigor, what kills a paper
• Signaling tactics matter (e.g. citing advanced combinatorics theorems to demonstrate seriousness), not just the quality of ideas
• For those outside that network: co-author with someone already publishing in target venues — learn the implicit rules directly
• Do not write your own story about what matters; get the real answer from someone inside the studios or institutions you want to join

Creative workflow and burnout prevention

• Following inspiration leads to feast-or-famine: 12-hour sessions followed by zero-output days that destabilise the whole project
• Factory mindset: set fixed work hours, clock in and clock out — stop even when you have energy left
• Consistent moderate output aggregates over time; erratic bursts do not
• Add external checkpoints — show work to someone on a fixed date — to prevent perfectionism from extending creative projects indefinitely

Why LLMs cannot become conscious

• A large language model is a static collection of matrix multiplications; it has no state, no memory, no actuation, no world model
• It produces a probability distribution over the next token; nothing changes between runs
• Consciousness or autonomy would require: a world model that updates from experience, values/incentives, and actuation (the ability to take action and observe outcomes) — tightly looped together
• DeepMind's Dreamer (now v3) is an example of that architecture: reinforcement-learning agent with world model, goals, and actuation — the type of system worth watching for control concerns
• AI company welfare research into LLM consciousness is characterised as a PR strategy; engineers at these companies know their models cannot meet the architectural requirements
• Real near-term AI breakthroughs: replacing search engines, accelerating software development — not emergent sentience