Rendered at 10:12:28 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
skybrian 18 hours ago [-]
Storing configuration and “memory” outside the LLM in prompts and skills written in English and in tools written in ordinary programming languages makes them far easier to modify by ordinary programmers who aren’t AI experts, and we can use the coding skills of LLM’s to modify them too.
It seems a lot more “open” and accessible than it would to store memories in an opaque neuralese that we have to do mechanistic interpretability research on to get even a partial understanding.
There’s a lot of churn in the agent harness space but I expect to see innovation come out of it, too.
cyanydeez 13 hours ago [-]
I'd expect a turtles all the way down harness, where you take your grandiose model and use it to train a smaller model to do the specific scoped work you want to do, and then use that to build a harness model that constrains output to the given paradigm (langauge/framework/fullstack, etc) and from there you can try to approach the singularity.
But the singularity doesn't exist.
cush 19 hours ago [-]
This kind of continuous fine tuning is being actively researched very heavily, but you might not see much about it at a conference because the labs don’t have enough to show their cards yet.
bantunes 23 hours ago [-]
I assume there's just too much money in optimizing for the thing that works now and making it faster than in risking it all on unproven approaches.
mempko 21 hours ago [-]
Companies are spending about $1 trillion this year in capex in the US! World wide estimate is about $2.52 trillion in AI spend in 2026 according to Gartner. There has never been a bigger spend in tech. It's so much spend that the software industry has to basically double in revenue in the next couple years to keep up.
Most of that spend is on infrastructure, GPUs, ASICS, and everything else that goes into a datacenter.
It seems a lot more “open” and accessible than it would to store memories in an opaque neuralese that we have to do mechanistic interpretability research on to get even a partial understanding.
There’s a lot of churn in the agent harness space but I expect to see innovation come out of it, too.
But the singularity doesn't exist.
Most of that spend is on infrastructure, GPUs, ASICS, and everything else that goes into a datacenter.