“In practical terms, this means never facing a blank page. In fact, don’t even try to pound out a first draft. Just throw a bunch of relevant notes into a blank document and experiment with them. Delete some. Then rearrange and reword what remains and give the piece a title. Look for connections between note-islands and explain how they are related.” – Tiago Forte
See: William Zinsser
Related:
- 1-1a5b4.3 build-launch-measure-learn and iterate - don’t restart
- 1-1a5b4.4 Real-time calibration (i.e., recalibration) lets you adapt to the changing landscape
- 3-1c2e1 Connecting-relating ideas is a way of creating knowledge
Codes will be rewritten (20250729)
- karpathy on yc
- software 1.0 = code which programs computer
- software 2.0 = weights which programs neural net
- github (in software 1.0) = huggingface (in software 2.0)
- software 3.0 = prompts which programs LLM
- LLM = programmable neural net
- a huge amount of software will be (re-)written
- llms are like OSs
- you can run cursor on gpt, claude, or gemini
- early hints of personal computing v2
- stacked apple minis at home
- llms flipped how technology diffuses in society
- consumer → corporations → government
- meaning, the latter two are lagging behind
- consumer → corporations → government
- llm psychology
- e.g., it doesn’t consolidate knowledge over sleep
- it’s like memento
- e.g., it doesn’t consolidate knowledge over sleep
- llm apps with partial autonomy
- AI generates → human verifies
- the latter can be made easier and faster with e.g., GUI
- how information is presented matters (9-2b1.2 Marshall McLuhan - ‘It is the medium itself that is the message, not the content’)revisit
- the former can be kept on a tight leash to increase the probability of successful verification
- i.e., via specialized AIs
- the latter can be made easier and faster with e.g., GUI
- AI generates → human verifies
- vibe-coding
- the easy part: coding
- the hard part: domain, authentication, payment, etc
- computer is telling you what to do
- new category of consumer-manipulator of digital information (i.e., new target to build for) emerged—and why codes will be rewritten (7-1d2a The relation between technology and the problem-situation (i.e., culture and mind) is reflexive)
- humans (= build GUIs for them)
- computers (= build APIs for them)
- new: agents ← computers (= build for these human-like computers)
- docs for llms >>> docs for people
- e.g., using markdown
- actions for llms >>> actions for people
- e.g., cURL >>> click
- e.g., MCP (model context protocol) server which provides AI agents a set of tools to call the API (e.g., Stripe)
- Gitingest >>> github
- devin deepwiki turns repo into more llm accessible format
- docs for llms >>> docs for people