← sources

How long do we have until RSI is achieved?

source: reddit (reddit:ChatGPT)published 2026-10-25 23:02ingested 2026-10-26 02:36

retained text

We’ve broken through every wall that LLMs have hit. And while I didn’t believe we’d get through them, we somehow kept doing it.

So why wouldn’t RSI be just another problem to solve rit?

Yes, it is a very difficult problem to have it create novel new R&D for current human AI researchers. And they may be lying internally when they do get asked about this. Just to prop up their stock or make them look better to their competitors. Which is a real concern, cause they’ve done it in the past.

But I don’t know, it feels different like we’re actually getting close with how much these companies are using their models, whether or not it’s forced or not by their staff. There is genuinly A small amount of recursive self improvement happening internally in these companies.

And even if they only improve the humans making the next model faster than without using it. That’s a form of RSI that will help speed up the inevitable goal we’re striving for.

So I’m curious, what is your guy’s theories and predictions of when we will reach it to a considerable percentage.

And when we do, will it be released to the public, or will it be highly commoditised and restricted to only the elite like the cybersecurity models are.

Cause if you gave every single human an RSI enabled agent that could learn and improve itself in whatever way they use or wanted. Even with heavy guardrails and no jailbreak to be seen, which is unlikely, but this is hypothetical.

Even with all that in place, you would still have massive concerns that can never be patched or foreseen.

I’m thinking with in the next 4 to 6 months, we will have achieved a canonical classical RSI LLM model.

Thoughts?

document fetched 2026-10-26 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xfgjl8/how_long_do_we_have_until_rsi_is_achieved