ChatGTP to analyze X-rays!

RE: ChatGTP to analyze X-rays!

Alasdair Hotston Moore — Tue, 26 Nov 2024 08:53:44 GMT

[quote userid="11308" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245928#245928"] often gives some, erm, interesting abnormalities[/quote]

Sadly as do veterinarians in my experience. I doubt it’s any worse than a typical practitioner. I also think it’s likely better than posting a screen shot to social media and relying on the opinion of strangers

RE: ChatGTP to analyze X-rays!

Lucy Fleming — Wed, 13 Nov 2024 09:38:02 GMT

[quote userid="9179" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245842#245842"]

Sarah Keir said:
Also all the AI algorithms are heavily biased (after all, coded by white, cis, males) and using them without this knowledge and trying to combat it is going to set back humanity by hundreds of years.

this is a very strange comment given the number of people involved in coding AI and the number of countries involved in generating AI technologies

[/quote]

My understanding is it's not the coding that's an issue, it's the information being used/ fed into them to generate the responses. That's what has tended to be biased historically, skewing what you get out of the programs. I imagine it's improving but it's a lot of historical bias to overcome.

RE: ChatGTP to analyze X-rays!

Anthony Dennison — Tue, 05 Nov 2024 16:37:55 GMT

There is already an AI/AGI system for radiographs - it's called SignalPet. It's something that has been put on all IVC practices. We don't really use it, often gives some, erm, interesting abnormalities

RE: ChatGTP to analyze X-rays!

Chris Milligan — Mon, 28 Oct 2024 23:29:44 GMT

AI is a loaded term, and a lot of people working in the field dislike it for various reasons. I like it as it's a pithy shorthand and - honestly - who really knows what intelligence is. There's also a tendency to assume content is 'algorithm driven', but this isn't the case. The algorithms used in the current iterations of large language models are very broad-brush guiderails, and the responses systems now generate are iterative rather than generative which is much closer to how people project their own thoughts. The difference is our AI models currently only work across a fairly narrow range of intelligence parameters, which is why they're so bad at even basic mathematics at the moment. Reasoning is a separate field of generative AI - if we can get reasoning algorithms to interact with LLM's then we're well on the way to a much more capable system. I have no doubt this will be the next milestone these admittedly extremely opaque companies hit in the next few years.

[quote userid="9515" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245839#245839"]Also all the AI algorithms are heavily biased (after all, coded by white, cis, males) and using them without this knowledge and trying to combat it is going to set back humanity by hundreds of years.[/quote]

this is a very strange comment given the number of people involved in coding AI and the number of countries involved in generating AI technologies. I'm sure, for instance, Meia Tegmark and Cynthia Breazeal who wrote much of the preliminary coding for early large language models, and who are both heavily involved in supervising AI safety efforts, would take exception to your comment

RE: ChatGTP to analyze X-rays!

David Mills — Mon, 28 Oct 2024 23:03:33 GMT

[quote userid="9179" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245830#245830"]it is AI, it's not AGI[/quote]

It really isn't AI. Despite what ChatGPT claims to be.

RE: ChatGTP to analyze X-rays!

Dinu Catilina — Mon, 28 Oct 2024 21:28:02 GMT

[quote userid="9515" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245839#245839"]Also all the AI algorithms are heavily biased (after all, coded by white, cis, males) [/quote]

Only that the tech industry and IT are two of the most progressive and inclusive for the LGBTQ community. Not many right rednecks know computing.

RE: ChatGTP to analyze X-rays!

Sarah Keir — Mon, 28 Oct 2024 20:07:57 GMT

The AI that we can use at the moment is called generative AI i.e. that it should learn, but it really isn't. It is regurgitative i.e. it spews out information gleaned from the internet and even makes it up when it can't find it!

Also all the AI algorithms are heavily biased (after all, coded by white, cis, males) and using them without this knowledge and trying to combat it is going to set back humanity by hundreds of years.

RE: ChatGTP to analyze X-rays!

Chris Milligan — Sat, 26 Oct 2024 01:13:24 GMT

good point! I have as little to do with all things repro as possible so hadn't twigged :D

RE: ChatGTP to analyze X-rays!

Michael Woodhouse — Fri, 25 Oct 2024 22:44:06 GMT

Sorry, but in vet med, AI is taken. We need to call it something else......

RE: ChatGTP to analyze X-rays!

Francisco Gomez — Fri, 25 Oct 2024 20:09:42 GMT

[quote userid="9179" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245830#245830"]

I'm looking at getting a group of us together in the veterinary field to discuss innovation and share AI playtime results if you're interested?

[/quote]

Absolutely. Count on me.

RE: ChatGTP to analyze X-rays!

Chris Milligan — Fri, 25 Oct 2024 16:18:27 GMT

[quote userid="8991" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245812#245812"]It's not AI. It's a series of algorithms. [/quote]

it is AI, it's not AGI.

[quote userid="11493" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245811#245811"]I'm an AI enthusiast. [/quote]

I'm looking at getting a group of us together in the veterinary field to discuss innovation and share AI playtime results if you're interested?

[quote userid="3222" url="~/f/non-clinical-questions/31070/chatgtp-to-analyze-x-rays/245826#245826"]It is good fun using it, but obviously not trusting it for diagnostics or anything[/quote]

it's horses for courses in some regards - depends what you need/want to use it for.

ChatGPT can be moulded to do most things if you give it the right parameters. If you're really patient you can almost get it to stop repeating fundamental errors like Francisco mentioned. I haven't tried GPT for rads, though other services do a pretty darn good job in my experience. I run pretty much all my rads through it when I have time to see what it throws up. More than once now it's beaten me to a metastatic disease diagnosis later confirmed by specialist review.

Bard and GPT are phenomenal for writing notes, discharge records, client handouts in the veterinary field. If you want a conversation with it Meta's AI is hard to beat as long as you instruct it not to just make lists and apologise for everything. There's a lot of functionality behind the scenes that you can unlock. Meta will respond to you in any language if asked, for instance, but will then back delete it and tell you it isn't trained in that language. Same with image creation but it will do it if you push hard.

If you want it to calculate stuff and tell you how it got there, however, well that's the next peak to summit!

RE: ChatGTP to analyze X-rays!

Stigen — Fri, 25 Oct 2024 11:38:22 GMT

It is good fun using it, but obviously not trusting it for diagnostics or anything. But it does help in differentials, and it does make a good text to use in cases of NAD

RE: ChatGTP to analyze X-rays!

David Mills — Thu, 24 Oct 2024 22:00:36 GMT

It's not AI. It's a series of algorithms.

RE: ChatGTP to analyze X-rays!

Francisco Gomez — Thu, 24 Oct 2024 21:18:04 GMT

I'm an AI enthusiast. I enjoy it using it. I have used it to help me structure blogs, emails, writing code (check out my poster in Porto's neuro ophthalmology conference thegovet.github.io/.../ ) and even now I'm building a remote control robot with it.

But AI makes catastrophic mistakes, will apologise for them, and repeat the mistakes again. Just like the predicted text in your phone. The important thing is learning how, when, and what to use it for, and be honest if you're using it.

I recommend Google's AI essentials course. It taught me a few things.