Jump to content

Talk:ChatGPT

Page contents not supported in other languages.
Page semi-protected
From Wikipedia, the free encyclopedia

chat.openai.com now redirects to chatgpt.com

?The infobox is now out of date, and because I have no idea how to change the website paramater of the infobox because the Website paramater has for some reason not been filled in could someone else please change it Someone-123-321 (I contribute) 02:54, 4 May 2024 (UTC)[reply]

The URL is pulled in automatically from Wikidata, so I opened a request over there to change it. We technically could override it here with an infobox parameter, but I don't think it's worth it since the URL redirects and will update automatically when it's changed on wikidata. Jamedeus (talk) 04:43, 4 May 2024 (UTC)[reply]
There's no hurry. The old link(s) still work(s). InfiniteNexus (talk) 21:44, 4 May 2024 (UTC)[reply]

I turned the copyright concerns part of "Reception" into a subsection of the said section, as it is evident that there may be more copyright challenges to the platform. One of these is not mentioned in the article: the Authors' Guild lawsuit vs. OpenAI. Relevant sources mentioning it: from ABC News, from Variety, and from WIRED. JWilz12345 (Talk|Contrib's.) 19:22, 12 May 2024 (UTC)[reply]

Update of the introduction

The introduction describes ChatGPT almost exclusively as a chatbot. Some readers may conclude that it can just read and generate text. I think it would be good to update it, maybe with a paragraph that gives an overview of the current features and potentially talks a bit about GPT-4o, since it's clearly the best model right now and it's multimodal. No need to use complicated words like multimodality, since many readers may not be familiar with the technical jargon. Alenoach (talk) 22:46, 21 May 2024 (UTC)[reply]

That's not ChatGPT, that's GPT-4. ChatGPT is "exclusively" a chatbot. InfiniteNexus (talk) 18:05, 22 May 2024 (UTC)[reply]
Doesn't the term ChatGPT also include things like speech-to-speech, and perhaps image generation via DALL-E 3? As I understood it, ChatGPT is more like an AI application that can include many subsystems for different things. Alenoach (talk) 18:32, 22 May 2024 (UTC)[reply]
That still describes a chatbot. DALL-E is a separate thing; multimodality describes GPT-4. InfiniteNexus (talk) 18:58, 22 May 2024 (UTC)[reply]

Deleting "virtual assistant" from the lead section

ChatGPT is not a virtual assistant. In addition to not having the typical virtual assistant features as Apple's Siri or Amazon Alexa, Google Assistant, etc., there is no mention of it being one in OpenAi's What is ChatGPT? FAQ either, which, at the time of writing this, states it was "Updated over a week ago." The newer integrations of ChatGPT with other applications and websites only integrate the chatbot and text generation abilities (mostly because it's the most trained proprietary LLM out there). Ara (talk) 19:44, 20 September 2024 (UTC)[reply]

"Updated over a week ago" does not mean "Updated under two weeks ago". Besides that, I agree with your point. –Gluonz talk contribs 00:24, 21 September 2024 (UTC)[reply]
Correct. Which is why I specified that those are the words stated on the FAQ page at the time I created this topic, and made no claims about the actual updation date. Thanks. Ara (talk) 04:52, 21 September 2024 (UTC)[reply]
Understood. I had simply been clarifying, as "Updated over a week ago" could reasonably imply "Updated under two weeks ago". –Gluonz talk contribs 14:11, 21 September 2024 (UTC)[reply]
Not an issue. However, the current revision of the page by you requires alteration (or citations). ChatGPT is an LLM. The text generation abilities of the LLM is what set apart ChatGPT from the non-AI chatbots. Cluttered looking text might be preferable to inaccurate information. In the page I linked above, the answer to the second most frequently asked question states ChatGPT is "a language model trained to produce text." No mention of the term "chatbot" on the page, or on any official OpenAI description/research. AraCh (talk | contribs) 15:41, 21 September 2024 (UTC)[reply]
I prefer Gluonz's version. ChatGPT is not exactly a LLM, it's the chatbot, the service. It relies on LLMs (or more precisely its GPT models), but it's not a LLM itself. Alenoach (talk) 15:45, 21 September 2024 (UTC)[reply]
It is literally an LLM. The full response to the FAQ I mentioned on this page is "ChatGPT is fine-tuned from GPT-3.5, a language model trained to produce text. ChatGPT was optimized for dialogue by using Reinforcement Learning with Human Feedback (RLHF) – a method that uses human demonstrations and preference comparisons to guide the model toward desired behavior."
You could even ask ChatGPT whether it's an LLM to confirm (I just checked, it confirms it's an LLM), and I could also link some of the most cited papers about this if needed. AraCh (talk | contribs) 16:02, 21 September 2024 (UTC)[reply]
As I understand it, ChatGPT is more than that. It can leverage different LLMs. And it includes things like the interface and the plugins. It's more like the system, of which the LLMs are components. Saying that ChatGPT is a LLM looks like an oversimplification. Alenoach (talk) 16:16, 21 September 2024 (UTC)[reply]
LLMs aren't simple at all - they literally kickstarted the AI frenzy. Regardless, current version of the page seems ok, factually correct and verifiable enough for any reader. It's especially important to keep this page as verifiable and NPOV as possible, due to the high page viewership, and more importantly to not accidentally contribute to the incessant negative media coverage of everything AI.
Also, I considered making a topic about this but since you guys are frequent editors I'll leave this here - I noticed a lot of the citations for specific research references in the article link to news and other websites. If there are any specific papers you would like cited or verified, you can @ me or drop a message in my talk page.
Thanks for taking the time to respond. AraCh (talk | contribs) 17:03, 21 September 2024 (UTC)[reply]
The mention of "language model-based" seems unnecessary in the first sentence in this version, because it's already implied by the term "chatbot". Also, as ChatGPT is becoming increasingly multimodal, it becomes less accurate. The previous phrasing, with "Based on large language models, ..." in the second sentence sounded a bit less redundant though and introduced the technical term "large language models". Alenoach (talk) 18:30, 21 September 2024 (UTC)[reply]

"Medicine" subsection

There is a lot of content in the "Medicine" subsection. It's probably interesting for health specialists, but for the average viewer, it's probably too detailed. It is also partially outdated (already!) because the accuracy of a model like o1-preview is likely very different from the one of a model like GPT-3.5.

The "Law" subsection may also be a bit too detailed.

Perhaps we could migrate some of its content to the specialized article "Artificial intelligence in healthcare", and try to have a relatively short, well-structured and up-to-date overview here? What do you think? Alenoach (talk) 17:42, 21 September 2024 (UTC)[reply]