@RayFroZn @unusual_whales I struggle to outweigh the steel-man of innovation incentive against the gini coefficient we’re currently experiencing in every major city.
@RayFroZn @unusual_whales There’s an exponential relationship between time & wealth accumulation at that level. You could make an argument there’s value for society to have capital allocators, but our country was most prosperous when we had the highest marginal tax rate.
@abrar_gist@theo Also, lazy MCP tool design generally flood the context and make agents worse. If the tools match 1:1 with endpoints, you’ve done something wrong.
@abrar_gist@theo When you’re able to use natural language to prompt & parse data, and then chain tools, it starts to make more sense. They enable a bunch of low-complexity tasks to be automated without strict input data contracts. Everything can be fuzzy except output.
@realMikeBlue@RossBarkan he just participated in a primary — and is running in the general where people will vote between him & other candidates
definitionally, that’s democracy.
@crutchcorn@tarekkh1997 unfortunately I think that’s also just what happens when you hit peak market saturation — you get every use case possible to solve, and that’s hell to architect around.
the most commonly used libs have the most use cases they serve, and you’re most likely to encounter em.
@nstfkc@mfpears@catalinmpit understanding LLM tokenization & context helps significantly with task chunking / sequencing
knowing how to articulate a task for an agent using keywords yields much better results for agentic programming (but applying specificity incorrectly is a footgun)