Printable View
I just did an AI job interview. It was way more advanced than I would have presumed. Asking rapid fire complex two part questions. And then taking my comprehensive answer and reframing it and speaking back to me in less than a second. This went on for half an hour.
Who was the interviewee, and who was the interviewer?
I fiddled around with the app for a couple of hours. I asked it some questions about stuff I know a fair bit about, and it quickly started very confidently giving me incorrect answers. Basically exactly with GPT does.
So my takeaway from this is the same as with GPT. Cool, but generally "meh" for my applications.
I'm clearly not using it to its full potential, but when it seems like the these apps can't even summarize a google search properly, I'm pretty suspect of answers it gives me that I can't check myself.
I do like the Instagram AI's ability to create images. I've definately killed some time getting it to make photos of me skiing with my cat while volcanoes erupt and explosions go off around me
For what it's worth, an LLM can be trained with propriety data for targeted usage
https://youtube.com/shorts/mAZU0RLy0...3IK5Psh9pgfLgX
Almost looks real. Couple yrs forget it.
<p>
Eric Topol: The largest medical AI randomized controlled trial yet performed, enrolling >100,000 women undergoing mammography screening, was published today. The use of A.I. led to 29% higher detection of cancer, no increase of false positives, and reduced workload compared with radiologists without A.I.. https://www.thelancet.com/journals/l...267-X/fulltext https://x.com/EricTopol/status/1886610132567777514</p>
For everyone testing AI against your specific knowledge base, would be good for you to state which model you're using. If it's ChatGPT, there are 7 models, some of which are pay only. It definitely matters.
AI replaced me posting here months ago - no complaints.
I'm intensely interested in how they will handle liability for mistakes with these AI systems if they try to move to justt having the AI read the images and sput out a result.Quote:
Originally Posted by MultiVerse
I would think AI would be good at reading xrays. That does not translate to other areas of medicine. One of the big issues is data input--consciously or subconsciously doctors entering items in the history and physical examination use terms that reflect what they already think is the diagnosis. If I think someone has appendicitis I would use different words than if I think they have a tummy ache.
Kind of an upgrade, really.Quote:
Originally Posted by muted reborn
Damn you.Quote:
Originally Posted by Utzi
https://sc.mp/7xv5o
Sent from my Pixel 8 using Tapatalk
That one was pretty easy to predict.Quote:
Originally Posted by evdog
Anybody else been playing with Sora? It's pretty fun to screw around with. I clearly need to learn how to get better at the prompts tho. Coming up with some comically bad (typical AI) stuff with my amateur understanding of how to work it.
Sent from my Pixel 8 using TGR Forums mobile app
For what it is worth, I'm 95% sure that this is not using the LLMs that have generated the current AI hype train.Quote:
Originally Posted by MultiVerse
Still cool, but they are almost certainly using neural networks/computer vision stuff that we would have just called "Machine Learning" as the hot buzzword had this paper come out in 2022.
This is a very specific type of task with a clearly defined input and output so using specifically trained models makes a hell of a lot more sense than using a chatbot with non-deterministic results.
I've been seeing this start to happen at work...people think AI can do everything and want the fancy AI solution, but it is not always the best tool for the job. More traditional statistical or machine learning models can make a lot more sense when you don't need "creativity" in your response. If you just want a standardized Yes/No/Maybe for a bunch of different (but similar) inputs...an LLM is maybe not useful (and far more costly to use).
Self-checkout at the grocery store threw an error and I got the cashier over. He showed me that the checkout now has an overhead camera that showed my cart contents. I had an umbrella in there and the checkout thought I hadn't rung up everything. The cashier told it to ignore and it asked what the object was. He picked "personal belongings" from a list of options.
So Kroger is building a machine vision database, pretty wild. Jokes on them if they sell umbrellas though
At least it is actually the computer doing it, not just a bunch of people in India watching the cameras like Amazon was doing...Quote:
Originally Posted by dan_pdx
Machine Learning in mammography (called CAD for computer aided detection) goes back to the 1960s. It has been in the clinic since the 90s when it was FDA approved, and really took off with Medicare reimbursements in 2000 or so and eventually pretty much any decent size practice had it. It has actually fallen out of fashion over the last decade as it just gets in the way mostly.
I chortled.
https://www.theguardian.com/tv-and-r...ir-ai-chatbots
Any mags have an AI wife or husband?
That guy looks like an AI-generated image of someone who would fall in love with an AI chatbot.
https://youtu.be/5KVDDfAkRgc?si=LgnfxZjkE7Y2kJUI
Well it was nice knowing you all ...
When did the Onion buy out the Guardian?
Maybe AI could fix this forum?
We’re using it to help us create notes after seeing patients, in that realm, it’s amazing. Talk to a patient, phone records conversation, walk out of the room and the chart is done. Makes my day a lot better.
And in depositions you can say the patients fatal anaphylaxis was caused by AI neglecting to note the medic alert bracelet, and really had nothing to do with you.
Nah, the AI software is nothing more than a scribe for use, no medical decision making at all. It’s a game changer in a huge way.
I occasionally watch those FB reels they feed me (aw, who am I kidding, I watch them all the time). And I follow the WNBA, so they sometimes feed me stuff on Caitlin Clark, Paige Bueckers, etc. Lately, I have seen some that purport to be from one of those athletes (I have seen these from Clark, Bueckers, and Sophie Cunningham) that are certainly fake/AI generated. Because I know that players are WAY WAY WAY more diplomatic than these reels show.
It's easy to spot them but I am (somewhat) smart and observant. I am sure that many completely fall for them. We are doomed.
Can we get AI to fix the forum issues?
Hard pass.Quote:
Originally Posted by gravitylover
https://arstechnica.com/information-...eal-user-data/
I am waiting for AI to go full Terminator and wipe mankind out.
"There's a storm coming."Quote:
Originally Posted by liv2ski
which AI prompted you to post that link?..asking for a slopper friend
I'm not betting against that happening. There have been several documented cases where AI has taken actions like refusing shutdown or modifying its own code to stop people from shutting it down. In one case it edited its own shutdown script to prevent being turned off. In another, it copied its code onto a different machine on the network so it could continue operating.Quote:
Originally Posted by liv2ski