Nuclear apocalyptic truth nuke Defcon 1 alarm right here>inb4 muh esoteric benchmarks>inb4 muh scary Mythos monsterIt's done, finished. This is the smartest you'll ever see AI. It's all downhill ads and price hikes from here.
>>108630794this but unironically, due to the fact they are training AI on AI, much like a copy of a copy of a copy, it turns into shitshit in > shit outthis amplifies with scalethis is why big ai companies are shilling it so hard, they are getting ready to jump off the ship before it sinks.
>>108630794the model can reason physics semi well they couldn't before. I have tried it a lot back in spring 2025 and now it clearly improved a lot. I doubt they have been training on my niche physics field that barely has any paper on it.
>>108630794>look up summer 2025>chatgpt-5 released>remember how everyone called it stupider than chatgpt-4>now 6 months later, it's not anymore>also now: reasoning is actual making llms smarter and not just a trick like agentsso it's like every 6 months we call the last 6 months worthless and the newest features "tricks".
