tspng
Joined 5 years ago
Joined 5 years ago
Looking at history, there is not really an argument for progress to stop in the long term. I think I get what you mean, that further advancement gets more and more complicated and the human mind gets to a limit what it can comprehend and reason about. But lots of research and progress have been augmented by technology in the past 70 years or so, and I don't see a reason why it wont continue along that path. I agree that there is a limit to everything. But it is almost impossible to forsee when and where that is going to happen.
In the context of AGI, is it inevitable eventually? Sure, I would agree, unless some catastrophic event puts back our technological advancement.
IMO, don't see the path to AGI with the current tech, though. All current SOTA agents are still LLM based with all their flaws (limited reasoning, generalization, incomplete world model, hallucinations, ...). At their core, they are still next token predictors with a limited context.
Most of the advances in AI in the past 2 years are in post-training and harnesses.
I'd expect a different core technology than just an LLM in order to get to AGI.
This could eventually be used for better in-display cameras where the pixels are used as a image sensors.
The researchers have published their results in Nature recently: https://www.nature.com/articles/s41586-026-10681-7
I did not yet test it with the copilot cli.
One issue I encountered. I cannot seem to create lists containing works like Todo, Done, .... No error message is shown. Creating lists with random strings always work though.