turlockmike

Isn't this just RSS with extra steps?

This is where you can use adverserial systems.

1. Agent writes a minimal test against a spec. 2. Another agent writes minimal implementation to make test pass only.

Repeat

This is ping pong pair programming.

The role of an engineer is to produce software that probably works. If you are producing more bugs it's because you're skipping provability. AI is also really good at writing tests and doing test-driven development You can get 100% branching coverage. You use a secondary LLM to review the work and make sure everything follows best practices.

LLMs Can accelerate you if you use best practices and focus on provability and quality, but if you produce slop LLMs will help you produce slop faster.

I built a tool specifically for dealing with remote mcp servers via the cli. No config files needed. FUSE (and Bash) is all we need. This way we can still get benefits of discoverability of mcp endpoints, but can use cli tools and bash to do things easier.

https://github.com/turlockmike/murl

When producing code is cheap, you can spend more time on verification testing.

Force the LLM to follow a workflow, have it do TDD, use task lists, have it write implementation plans.

LLMs are great coders, but subpar developers, help them be a good developer and you will see massive returns.

I believe in a very practical definition of AGI. AGI is a system capable of RSI. Why? Because it mimics humans. We have some behaviours that are given to us from birth, but the real power of humans is our ability to learn and improve ourselves and the environment around us.

A system capable of self improvement will be sufficient for AGI imo.

MCP is just JSON rpc with one useful command (list-tools). Turns out that's probably all we ever needed.

I definitely exaggerated, but a great example of the opposite case is CERN. Physicists knew the limits of what could and couldn't be tested with it, but they hyped it up to convince governments to spend money anyway. The JWST is a much more fruitful project, but how many of those are there compared to projects that just focus on getting grant money. We should build more telescopes and fewer participle accelerators, but the grant money doesn't match the need since it's hard for politicians to understand.

This boils down to a fundamental question. Why do we spend any time doing science to begin with? Historically scientists were drawn to the field in order to improve human understanding of our reality. These individuals often died quite poor and unknown, but advanced us forward. Now popular science is the goal and getting huge money grants. The goal is no longer the pursuit of knowledge, it's a money game. Like journalism. The only useful science done at the moment is at tech companies who will use it to build better products.