“The age of the idea guys has begun.”
Articles mentioned:
- LLMs as a tool for thought by Amelia Wattenberger
- Micropayments and Mental Transaction Costs by Nick Szabo
- How our interfaces have lost their senses by Amelia Wattenberger
Talks mentioned:
- The Art of Bitcoin Rhetoric by Bitstein
Books mentioned:
- Human Action by Ludwig von Mises
- Working in Public by Nadia Eghbal
In this dialogue:
- nak
- Files
- SyncThing (and how it BitTorrent Sync became Resilio Sync)
- Convention over configuration
- Changes & speciation
- File systems as sources of truth
- Vibe-coding shower thoughts
- Inspiration and The Muse
- Justin's LLM setup
- Tony's setup (o1-pro as the architect)
- Being okay with paying for LLMs
- Anthropomorphising LLMs
- Dialog, rubber-duck debugging, and the process of thinking
- Being nice and mean to LLMs
- Battlebots & Gladiators
- Hedging your bets by being nice to Skynet
- Pascal's Wager for AI
- Thinking models vs non-thinking faster models
- Sandwich-style LLM prompting, again (waterfall stuff, HLDD / LLDD)
- Cursor rules & Paul's Prompt Buddy
- Giving lots of context vs giving specific context
- The benefit of LLMs figuring out obscure bugs in minutes (instead of days)
- The phase change of fast iteration and vibe coding
- Idea level vs coding level
- High-level vs low-level languages
- Gigi's "vibeline"
- Peterson's Logos vs Vervaeke's Dia-Logos
- Entering into a conversation with technology
- Introducing MCPs into your workflow
- How does Claude think?
- How does it create a rhyme?
- How does thinking work?
- And how does it relate to dialogue?
- Gzuuus' DVMCP & using nostr as an AI substrate
- Language Server Protocols (LSPs)
- VAAS: Vibe-coding as a service
- Open models vs proprietary models
- What Cursor got right
- What ChatGPT got right
- What Google got right
- Tight integration of tools & remaining in a flow state
- LLMs as conversational partners
- The cost of context switching
- Conversational flow & how to stay in it
- Prompts VS diary entries
- Solving technical vs philosophical models
- Buying GPUs & training your own models
- Training LLMs to understand Zig
- Preventing entryism by writing no documentation
- Thin layers & alignment layers
- Working in public & thinking in public
- Building a therapist / diary / notes / idea / task system
- "The age of the idea guys has begun."
- Daemons and spirits
- Monological VS dialogical thinking
- Yes-men and disagreeable LLMs
- Energy cost vs human cost
- Paying by the meter vs paying a subscription
- The equivalence of storage and compute
- Thinking needs memory, and memory is about the future
- Nostr+ecash as the perfect AI+human substrate
- Real cost, real consequence, and Human Action
- The cost of words & speaking
- Costly signals and free markets
- From shitcoin tokens to LLM tokens to ecash tokens
- Being too close to the metal & not seeing the forest for the trees
- Power users vs engineers
- Participatory knowing and actually using the tools
- Nostr as the germination ground for ecash
- What is Sovereign Engineering?
- LLVM and the other side of the bell-curve
- How nostr gives you users, discovery, mircopayments, a backend, and many other things for free
- Echo chambers & virality
- Authenticity & Realness
- Growing on the edges, catering to the fringe
- You don't own your iPhone
- GrapheneOS
- WebRTC and other monolithic "open" standards
- Optimizing for the wrong thing
- Building a nostr phone & Gigi's dream flow
- Using nostr to sync dotfile setups and other things
- "There are no solutions, only trade-offs"
- Cross-platform development
- Native vs non-native implementations
- Vitor's point on what we mean by native
- Does your custom UI framework work for blind people?
- Ladybird browser & how to build a browser from scratch
- TempleOS
- Form follows function & 90's interfaces
- Lamentations on the state of modern browsers
- Complexity & the downfall of the Legacy Web
- Nostr as the "new internet"
- Talks by Ladybird developer Andreas Kling
- Will's attempt of building it from scratch with Notedeck & nostr-db
- Justin's attempt with rust-multiplatform
- "If it doesn't have a rust implementation, you shouldn't use it."
- Native in terms of speed vs native in terms of UI/UX
- Engineer the logic, vibe-code the UI
- From Excalidraw to app in minutes
- What can you one-shot?
- What do you need to care about?
- Pablo's NDK snippets
- 7GUIs and GUI benchmarks for LLMs
- "Now we're purpose-building tools to make it easier for LLMs"
- "Certain tools really make your problems go away."
- Macros and meta-programming
- Zig's comptime
- UNIX tools and pipes
- Simple tools & composability
- Nostr tools for iOS & sharing developer signing keys
- Building 10 apps as one guy
- Simplicity in a community context
- Most people are on phones
- Most people don't install PWAs
- Zapstore & building our own distribution channels
- Web-of-trust and pushing builds quickly
- Improving homebrew by 10x
- (Micro)payments for package managers
- Guix and bitcoin-core
- Nix vs Guix
- Reproducible builds & web-of-trust
- Keet vs "calling an npub"
- Getting into someone's notifications
- Removing the character limit was a mistake
Recorded at 889,812.