I refuse to pay for AI, so I only use the freely available GPT, and I can’t comment on coding or other technical capabilities, but I feel like in the last couple of months, it has actually become dumber.
Half a year ago Musk was saying this is a hard take-off for AI, what means AI would improve itself and you basically can’t stop it anymore. Now he says he wants a pause. The hard take-off seems to have fallen flat.
The one thing that hasn’t plateaued yet is the ability of open models running on a gaming PC to match OpenAI and Anthropic are doing. I just watched a video on the Codacus YouTube channel where the guy is running Qwen 3.8 Flash, a 177B model, on his RTX 3060.
Or the technology has plateau’d and this is their cover story. I don’t know, but these two people are known liars.
Model collapse is finally landing.
But…but… GPT6 Astra is AGI! I swear!
GPT6 couldn’t even make me the most simple blueprint for merging 8 belts into 4 in Factorio. AGI it is most certainly not.
I refuse to pay for AI, so I only use the freely available GPT, and I can’t comment on coding or other technical capabilities, but I feel like in the last couple of months, it has actually become dumber.
For some time now I think its been the case that frontier models perform worse (or at least more inconsistently) at procedural activity.
My first thought also.
Half a year ago Musk was saying this is a hard take-off for AI, what means AI would improve itself and you basically can’t stop it anymore. Now he says he wants a pause. The hard take-off seems to have fallen flat.
LLM’s plateaued quite some time ago. Everything since has been an attempt to keep the hype going.
AI has never improved, nor will it ever. There’s nothing to worry about.
Yeah they spent a few years fine tuning but the gains were limited from the start.
The one thing that hasn’t plateaued yet is the ability of open models running on a gaming PC to match OpenAI and Anthropic are doing. I just watched a video on the Codacus YouTube channel where the guy is running Qwen 3.8 Flash, a 177B model, on his RTX 3060.
At what quantisation? You’re most likely better off running 27b dense or 35 moe
They’re out of money and this is their cover story.