Last month or so just been getting worse and worse results from chatgpt - quality went down... Claude on the other hand keeps surprising me. What's going on?
9 comments
Because the only place where "free" AI is viable is when the model is running on your own computer. It's not sustainable at all, from a business perspective, to let millions of connections absolutely slam a huge AI model 24/7 for free - even with usage limits and throttling.
https://en.wikipedia.org/wiki/Jevons_paradox
Better to write a solid inference engine, tell people to run it themselves, and externalize that cost.
You gave no details at all, but ChatGPT seems better for me.
What is your use case? Which tier are you on? What sorts of tasks are you using it for that you notice quality is down?
The models are being tweaked for purposes other than usefulness. Its a product after all. It's not really a tool in the conventional sense of the word.
Sure you can use it as a tool, but you can also use it as an experience, or even a companion. But that requires it to be "not boring".
Quite frankly, "useful" is the least profitable route for this technology. It's all about the experience. People pay way more for that.
Thanks, I think your comment is spot on — and sadly, it reflects a familiar pattern: products start out as reliable tools, then shift toward engagement and ‘experience’ once monetization takes priority. From a business perspective it makes sense, but for users who valued the usefulness first, it feels like a step backwards
It’s hard to understand what you’re actually asking here. Right now your post is too vague — you hint at a problem, but don’t give enough detail for anyone to meaningfully respond. On HN, it really helps if you spell out exactly what you mean, provide context, and show what you’ve already tried or thought about. Otherwise it just comes across as expecting people to do the work of figuring out your question for you. Could you clarify your definitions, give concrete examples, and be more precise about what kind of feedback you’re looking for?
Write a comment to an Ask HN post chastising the poster for not adequately explaining what they mean. Nope, works fine for me.
It’s not. Your expectations are getting higher.
I’ve found ChatGPT to be getting better and better over the past month or so.
I can't tell you, but over the last year, as a consumer, I see consistent oscillation up and down in quality. Just today I noticed a big change in Perplexity, where it took an unusually incorrigible stance and even flagrantly lied to me.
If any value comes of this, it's to always remember to not get too comfortable, friendly or reliant, because these tools are not, have not, and never will be truly for us, but will continuously erode in some critical way for the little consumer while improving for the big ones.
Maybe something open source will save us little people and provide something stalwart. I hope so. But I won't expect anything from the big brands beyond ephemeral wows here and there, until they stamp them out and reserve them for the elite.
Probably not helpful. Sorry for the rant. I certainly do see quality variations and I'm not certain why, other than being convinced it's never noble intention.
OpenAI bleeding money running their models. The routers in gpt5 exist to save them money, not improve inference
[dead]