Rant time 😈
I implemented and shipped support for MCP at Clerk, I contributed to the MCP spec and SDKs, and I was involved in steering the direction of the protocol in the early days, even attending in-person meetings about it with a variety of industry players.
I am now very strongly against MCP and advise against using it both as a consumer and as a provider. Unfortunately, I don't think there's any going back at this point, but I feel sadly confident that it will waste countless hours of everyone's time over the coming years, when a much easier and better answer is right under our noses.
Let me tell the full story.
Dabbled over the weekend writing code like an ancient developer from 2023 without any AI at all and it was both sad and happy to see how much better my code was in every way.
@wesbos To be fair, if we assume it would take you 1 second to write ls yourself, you only need to make at least 12*60*60 cents an hour for you to be saving money by outsourcing to AI. Let's just ignore that you spent longer than 1 second writing /agent ls of course...
@RazorOil How do we know that "Venezuela has the world's largest oil reserves"? Is it true that when Chavez came to power he had PDVSA arbitrarily increase the estimated reserves in their reporting?
@GergelyOrosz Claude code very often writes bad tests. Anyone who’s worked in a complex code base with nontrivial requirements knows this. And this is before taking into consideration the overhead of running a very large test suite and the optimisations that become necessary.
You can't go back in time... But perhaps look at this book.
It's accessible to non-math people but also has insights for VERY mathy people who don't understand real physics. There's even a page that explains "tensors" - might be useful for would-be AI engineers ;-)
If you're bored, flip to a random page of this book and think about it. A physicist might have to flip a lot of pages to find something interesting, but most people will learn stuff (about the universe!) right away by doing this.
https://t.co/ZXDDiRAoyY
@hsu_steve Amazing. I was literally looking through old bookmarks and Amazon lists yesterday trying to find this book. I only vaguely remembered the cover. Thanks.
This is a great talk by Chollet about the path to AGI and why scaling alone (and not even scaling + CoT reasoning) will get us there. Video link below.
He discusses abstraction, symbolic reasoning, continuous vs discrete optimization (program search), and many other important topics.
Note he keeps using the term "Fluid Intelligence".
$trillions and geostrategic dominance are on the line so AI researchers have to be realistic about what "intelligence" is. They end up recreating the methods of psychometrics (now applied to AIs instead of humans) that have largely been driven out of the academy by woke stupidity.
@VicVijayakumar Hard to believe these are real. As in, the objective is to performatively vibe code for Twitter where something like “ai deleted my database” is a sure to go viral, rather than actually building a product.