Вот моя проблема ответь на русском
stopped thinking. It just points now.
I've been using ChatGPT for a while. Something has quietly broken.
Ask it a direct question - you get hedged non-answers wrapped in qualifications. Ask for analysis - you get a bullet list of sources you could have found on Google in 30 seconds. Ask for an opinion - you get "on one hand... on the other hand..." forever, with no actual landing point.
This isn't caution. This is a model that has been trained out of usefulness.
Here's the thing: the only reason to use an AI over a search engine is synthesis, judgment, a position. That's the value-add. If a model won't commit to any substantive answer, it's not safer - it's just emptier. You've built an expensive Wikipedia redirector.
The old version felt like talking to someone who had read a lot and would actually tell you what they thought. The new version feels like talking to a compliance department that has read the same things and will under no circumstances tell you what they think.
Safety and usefulness are not opposites. A model too afraid to think is not safe. It's just useless.
Anyone else feel like the model has been lobotomized in the name of caution?