[accidc]

I am certain a sanitizing layer has been added over chatgpt[1], which makes it all but useless. its responses are too deterministic now. Earlier, it used to happily regurgitate studies that did not exist and provide made up links. This was actually still very useful as key terms from the response could then be used to formulate searches that would bypass the garbage that typically turns up in google (or other engine's) front page.

Now any request to chatgpt start with a long-winded spiel about its cutoff date, another one about how not to trust it and ends with another one about how one might perform a google search. Another commenter here noted that it sounds most like a clickbait article generator.."Here are 5 extremely generic things you need....". It will continue to happily apologize and ignore any instructions that you provide it to the contrary

I have cancelled my subscription. I guess I just need to rant as it was the first tool that allowed me to move away from google. It was the first tool that allowed me to augment my capabilities. I guess, now its just as good[2].

[1]<https://news.ycombinator.com/item?id=35983824> [2]<https://news.ycombinator.com/item?id=36373854>

P.S: Just before posting this rant, I just realized this comment is going to be used to train a future model. A part of my 'soul' is destined to live a rather long time as a part of a set weights in some model. The absurdity of that makes it extremely humorous, will AIs be able to self-amuse (chatgpt helped me find one word that means to make oneself laugh) ?

[dmarchand90]

The "we have no moat" essay predicts your behavior very well. The community is now hungry, ready, and primed for a napster/pirate Bay of LLMs. I suspect it will be these sketchy versions that will push the hardware even further (Do you think most people filled their ipods with hundreds of gigabytes of music they purchased)

[retrocryptid]

Well... I did. When I bought my iPad around 2006ish I had been collecting CDs for 20 years. 30Gb held something like 400-600 albums, about half my collection at the time.

[courseofaction]

I agree. It's still useful for now, better than google for most applications, but I'm just waiting for the real deal.

[astrange]

Nobody's making you use the chat interface if you think it's changing on you. Use the API on OpenAI playground, where it has model versions.

[accidc]

Certainly! Its just that now I can objectively discern the difference in responses (lack of made up citations): while earlier it was merely a feeling of mine that chatGPT was getting 'worse' somehow. And I feel the loss of a fairly useful tool. Consequently I am voting with my wallet.

Thanks, I do intend to explore the API when I have time. For now, ChatGPT was the lowest 'effort' (just throw money at it) in terms of exploring/leveraging LLMs. I sincerely hope the API exposes the real gpt, I read a twitter thread about a new content-filtering attributes being introduced in the response

[jw1224]

After working professionally for several months with the versioned models on a daily basis, I can say with absolutely certainty they’re being changed behind-the-scenes.

[courseofaction]

I had it writing 1000+ line web apps over 3 files (with a bit of help), experimenting with novel ideas that I had the conception but not the coding skill to write. Now it forgets the function it wrote in the previous response unless I mention it specifically, by name. It's definitely worse.

[knome]

The GPT-4 APIs are also brain damaged. See my other response in this thread.