very good and directionally correct, but some fundamental errors imo:
- 402 was revived by lightning labs over 5 years ago, not by coinbase's openx402
- unclear that agents would *prefer* stablecoins when bitcoin's right there. i bet cashu will win. If the agents are truly autonomous, setting up their own bank in cyberspace is ideal.
- they won't use proprietary models. The self-improvement loop also means a capability loop, so they may organize to train new models. Either way the idea that their lifeblood and intelligence will depend on API calls to third-party providers is a non-starter for autonomous intelligence.
- they still won't know _what_ to build. Agents will have to develop taste, perspective, immense context, etc.
View quoted note →
fwiw Elixir always been S-tier at docs, and its all of a sudden paying off big.
like I can call 'Code.fetch_docs(module)' and get the docs at runtime. so I can generate skills or let the LLM get context on the fly.
View quoted note →
really starting to feel like a coordinated campaign with quantum fear as Trojan horse.
the fact that we have bip360 is _insanely_ proactive given the current risk quantum poses. Saylor and starboy can fork off at their leisure.
View quoted note →
Like python, elixir can fetch docs from bytecode at runtime, so now I have an LLM creating modules, then creating skills from those modules' docs, so it can then create new more complex modules and so on. autonomy loop 🌀
i never liked or used reddit much, so moltbook is not landing for me. it's slop-ception on a centralized platform.
icymi twitter has been agent bots talking to each other for years.
great documentary by Micode (in french) on the cognitive risks/implications of LLMs.
disregard clickbait title. we're gonna have to reinvent education. It's gonna be lit.
I am very bearish Anthropic. This guy doesn't seem to quite understand LLMs, humans, industry, or economics. And every line of reasoning carries the reeking specter of Effective Altruism.
given that people can't seem to agree on a path for CTV-like constructs, we should seek to move forward with OP-CCV (BIP 443), the successor to OP-VAULT.
Everyone wants vaults. They're obvious, useful constructs that have a clear use case. The implementation surface is smaller than CTV packages.