- From: Milan Regec <milan.regec@hey.com>
- Date: Wed, 27 Dec 2023 00:50:09 +0100
- To: chagnon@pubcom.com, Kavein Thran <kaveinthran@gmail.com>, Guy Hickling <guy.hickling@gmail.com>
- Cc: IG - WAI Interest Group List list <w3c-wai-ig@w3.org>
- Message-ID: <ef893f37fedaab5e74bc2df3595b6eaf7c26b432@hey.com>
Hi, please allow me to chip in, as I have some experience with ChatGPT. Firstly, to touch the quoted article. Joe Watkins sure might be an accessibility expert, what he most likely is not, is an AI expert. The whole piece should be considered a rather entertaining reading. What we see here is a naïve approach to an expert problem with a result to surprise no one. What version of ChatGPT he used? Feb 23 is history in terms of an AI development. Did he do some pre-training, pre- prompting or at least some custom instructions? (the answer is "no") My experience is that yes, ChatGPT can be beaten into submission to spit an accessible code, but if approached right it can be an useful assistant to either modify inaccessible forms with correct labels, create more sophisticated input verification and feedback, customize / apply general examples to a specific problem on a concrete website. And all that quite willingly. So, don't brush off this tool just yet. Learn how to effectively use it and prepare to shed some buck. If you just start asking it accessibility related questions of the bat, you will not have much luck. And if you are not an expert, it will mislead you and provide you with false, incorrect information. Just like it says in the disclaimer. This behavior is very much in line with the principles of how LLMS work. But if you provide it with background information, give it examples of what you want and stick to working with code rather then "general advising", it can be very useful even today and it can accelerate your learning immensely, provided you have already some solid, even if limited, background. There are some areas where I hit the ceiling pretty hard though. I have tried to build a custom GPT focused on simplifying the language on the website for the mentally challenged individuals. Specifically, I have tried if it can simplify instructions of the local government website of the Social services for the disabled people into the language a mentally disadvantaged can actually understand and follow. I (or it?) failed miserably. If anybody else is interested in the topic, please reach out. Peaceful Christmas days to everyone, Milan PS: I despise overlays just like most of you and maybe even some more. If website owners would put the same amount of time and money they spent of implementing overlays into addressing the most serious accessibility issues they have, the overall experience would be much better of everyone. On December 26, 2023, Kavein Thran <kaveinthran@gmail.com> wrote: > Hi,I appreciate the sharing and articles on AI limitations and bias. > Somehow, the framing of this post feels like discouraging to say the > least. The person who starts the discussion are just using the tool to > frame how they think about it, they're not, at least from my > understanding, saying that something is right or wrong. They are using > the chat bot to frame what's going on in their mind and coming back > to the community for better wisdom. I don't find issue with that. It's > about, using the tool more mindfully.The subsequent reply talking > about how AI can take over the world is an overstretch to a very > narrow discussion that we are having here about reflowable content. I > feel like the reply is a bit defensive and anxious.I wish we can have > more nuance discussion about technology and again, it's really good to > ask what we accomplish from the replies that we are going to give to a > person.The assumptions that underlie here is more about the person do > not know how to use chatgpt mindfully, I don't think it's true.. > > Regards, > Kavein > Kaveinthran (He/Him) > Curious, Native Blind > Part time Research Consultant in ADPAN > Disabled independent Human Rights Advocate > email: kaveinthran@gmail.com > twitter <https://twitter.com/kaveinthran> > My LinkedIn <https://my.linkedin.com/in/kaveinthran> > > On Thu, Dec 21, 2023 at 06:12 <chagnon@pubcom.com > <mailto:chagnon@pubcom.com>> wrote: > > Quote: But in both cases it still refused to back down completely > > and insisted it's solution should be used in addition to the correct > > solution! EndQuote > > > > > > > > Huh. > > > > Did it go something like, “I’m sorry, Dave. I’m afraid I can’t do > > that”? * > > > > > > > > (* From Stanley Kubrick’s 2001: A Space Odyssey) > > > > > > > > Thanks, Guy, for this fascinating information from the industry. The > > thought that creeps into my mind is that we’re eventually going to > > homogenize our knowledge down to what AI has approved, rather than > > what is truthful or accurate. Will we lose novel new ideas, > > differing points of view, thinking outside the box, etc.? > > > > > > > > We will achieve truthiness! > > > > Whoever controls the AI algorithms will control what we humans know > > and think. Gosh, are we moving backwards to Orwell’s 1984? > > > > > > > > —Bevi > > > > — — — > > > > Bevi Chagnon | Designer, Accessibility Technician | > > Chagnon@PubCom.com <mailto:Chagnon@PubCom.com> > > > > — — — > > > > PubCom: Technologists for Accessible Design + Publishing > > > > consulting • training • development • design • sec. 508 services > > > > Upcoming classes at www.PubCom.com/classes > > <http://www.pubcom.com/classes> > > > > — — — > > > > Latest blog-newsletter <https://mailchi.mp/e694edcdfadd/class- > > discount-3266574> – Simple Guide to Writing Alt-Text > > <https://www.pubcom.com/blog/2020_07-20/alt-text_part-1.shtml> > > > > > > > > From: Guy Hickling <guy.hickling@gmail.com > > <mailto:guy.hickling@gmail.com>> > > Sent: Wednesday, December 20, 2023 4:39 PM > > To: WAI Interest Group discussion list <w3c-wai-ig@w3.org > > <mailto:w3c-wai-ig@w3.org>> > > Subject: Re: Reflow > > > > > > > > You are quite write to check if what ChatGPT is true before using > > it. AI tools like ChatGPT are very questionable at the best of > > times. They are very useful as tools, but all they do is draw > > information from a range of other sources, and the information they > > gather are just a reflection of those unknown sources. So I would > > not class anything obtained that way as "definitive"; much better to > > track down the original sources and (if they are reputable), quote > > them instead. > > >
Received on Tuesday, 26 December 2023 23:50:17 UTC