Devin Elliot

Every $10k in tokens is worth $1m in labor if you know what you’re doing

You won’t really know how to use AI until you’re pushing billions of tokens daily/weekly Anyone telling you about ai that doesn’t do that, is just guessing or parroting things they’ve heard

Anyone here with $100m+ in BTC outside of an exchange they want to sell? I have a motivated buyer

The most impressive thing about farcaster is how far behind everyone here is while believing they’re at some forefront. The only caveat being Quilibrium which is, conversely, light years ahead.

Was on a tech support call yesterday, the dude was using AI to look like someone else. If I wasn’t deeply confident in crypto, they might have gotten me

Most of what people attribute to an ai model is actually just the harness. Models have changed very little and cannot possibly ever do all the things that people are currently assuming they can

Anyone here randomly an expert in mining operations, specifically around material recovery from tailings or the like?

Guy who shows up to Internet late and then tells everyone else how it should work

Your data deserves the same level of security and care as your crypto does, even better when they’re in the same place, then everything gets easier

If you’re still manually checking code then you have yet to figure out how to build with AI

You can do things in a better way or just keep doing it like everyone else

Turns out you don’t need code at all, just the memory of it

After a week with Codex, it’s mostly garbage at anything new and novel. The best thing it can do is monitor a release pipeline Claude much better at communication but is now a hallucination storm Grok finally works in cursor but I’ll need a billion tokens with it to know if it’s valid or more mediocrity Composer 1.5 is the cheapest model in cursor and at this point the most reliable model. Its competent where codex is for a fraction of the price and less stupid

Any ai company that’s hiring, doesn’t actually know how to use AI yet

For any developers with skill la left in this place, here is a tensor compiler. This will allow you to turn brittle if/else/then logic into differentiable predicates. I’ve also provided the primitives for a new type of “database” which lets you run entirely in memory, gives you embeddings/vectors natively and a graph natively, no extra code to enable those features and if you need to handle critical data transfers then you also have a native crdt library built in. Everything is in rust https://github.com/monetic-labs/an-tensor-compiler

Building things and getting them run in ways people want to use them is very hard, but that doesn’t excuse ignoring all the signs of what not to do, besides being hard they come with responsibility. Most of that responsibility can be managed by simply being honest with yourself instead believing your own confabulations

Lulz

Deport liberal white women

Maybe you should stop typing nonsense and go lift some weights while eating a burger