"With the shift towards this multi-agent collaboration and orchestration world, you need a neutral substrate that has money, identity, cryptography, and web-of-trust baked in, to make everything work."
Pablo & Gigi are getting high on glue nostr.
Books and articles mentioned:
In this dialogue:
- vibeline & vibeline-ui
- LLMs as tools, and how to use them
- Vervaeke: AI thresholds & the path we must take
- Hallucinations and grounding in reality
- GPL, LLMs, and open-source licensing
- Pablo's multi-agent Roo setup
- Are we going to make programmers obsolete?
- "When it works it's amazing"
- Hiring & training agents
- Agents creating RAG databases of NIPs
- Different models and their context windows
- Generalists vs specialists
- "Write drunk, edit sober"
- DVMCP.fun
- Recklessness and destruction of vibe-coding
- Sharing secrets with agents & LLMs
- The "no API key" advantage of nostr
- What data to trust? And how does nostr help?
- Identity, web of trust, and signing data
- How to fight AI slop
- Marketplaces of code snippets
- Restricting agents with expert knowledge
- Trusted sources without a central repository
- Zapstore as the prime example
- "How do you fight off re-inventing GitHub?"
- Using large context windows to help with refactoring
- Code snippets for Olas, NDK, NIP-60, and more
- Using MCP as the base
- Using nostr as the underlying substrate
- Nostr as the glue & the discovery layer
- Why is this important?
- Why is this exciting?
- "With the shift towards this multi-agent collaboration and orchestration world, you need a neutral substrate that has money/identity/cryptography and web-of-trust baked in, to make everything work."
- How to single-shot nostr applications
- "Go and create this app"
- The agent has money, because of NIP-60/61
- PayPerQ
- Anthropic and the genius of mcp-tools
- Agents zapping & giving SkyNet more money
- Are we going to run the mints?
- Are agents going to run the mints?
- How can we best explain this to our bubble?
- Let alone to people outside of our bubble?
- Building pipelines of multiple agents
- LLM chains & piped Unix tools
- OpenAI vs Anthropic
- Genius models without tools vs midwit models with tools
- Re-thinking software development
- LLMs allow you to tackle bigger problems
- Increased speed is a paradigm shift
- Generalists vs specialists, left brain vs right brain
- Nostr as the home for specialists
- fiatjaf publishing snippets (reluctantly)
- fiatjaf's blossom implementation
- Thinking with LLMs
- The tension of specialization VS generalization
- How the publishing world changed
- Stupid faces on YouTube thumbnails
- Gaming the algorithm
- Will AI slop destroy the attention economy?
- Recency bias & hiding publication dates
- Undoing platform conditioning as a success metric
- Craving realness in a fake attention world
- The theater of the attention economy
- What TikTok got "right"
- Porn, FoodPorn, EarthPorn, etc.
- Porn vs Beauty
- Smoothness and awe
- "Beauty is an angel that could kill you in an instant (but decides not to)."
- The success of Joe Rogan & long-form conversations
- Smoothness fatigue & how our feeds numb us
- Nostr & touching grass
- How movement changes conversations
- LangChain & DVMs
- Central models vs marketplaces
- Going from assembly to high-level to conceptual
- Natural language VS programming languages
- Pablo's code snippets
- Writing documentation for LLMs
- Shared concepts, shared language, and forks
- Vibe-forking open-source software
- Spotting vibe-coded interfaces
- Visualizing nostr data in a 3D world
- Tweets, blog posts, and podcasts
- Vibe-producing blog posts from conversations
- Tweets are excellent for discovery
- Adding context to tweets (long-form posts, podcasts, etc)
- Removing the character limit was a mistake
- "Everyone's attention span is rekt"
- "There is no meaning without friction"
- "Nothing worth having ever comes easy"
- Being okay with doing the hard thing
- Growth hacks & engagement bait
- TikTok, theater, and showing faces and emotions
- The 1% rule: 99% of internet users are Lurkers
- "We are socially malnourished"
- Web-of-trust and zaps bring realness
- The semantic web does NOT fix this LLMs might
- "You can not model the world perfectly"
- Hallucination as a requirement for creativity
