What do you want to learn on the AI water coolers next meetup and long-term, how are your AI projects coming along?

43:31I'll report from the field. So the one thing I like to see now, I personally don't vibe code myself, but the one thing, but there are people out there who love doing it, is something like Aftershoot. So I spoke about this a little bit earlier. So Aftershoot is essentially, it's a photo app that you throw your photos in and it just analyzes your photos.

43:51It goes, I think these are the best ones and these are the worst ones. People are blinking in these. These ones are blurry. so that kind of stuff depending on how much you pay depending on how much and you can actually go in and fix stuff so you can actually there's lots of stuff you can do in that app I'm interested to see people doing add-ons or plugins for

44:07these kind of apps that will do very bespoke stuff so people who can vibe code or people who are developers themselves doing rapid development using some form of vibe coding and that's kind of where I'm hoping to see some of this AI because this is one it's an AI app but also using AI and vibe coding to add plugins to do very particular features to particular photographs you know those

44:29kind of things so I'm hoping that's kind of where we we get to I don't know I know it will because there are people out there who will want these tools they go well this tool doesn't exist I can vibe code it and developers who are already creating these kind of apps they can rather than take three weeks of vibe code something they can take three hours or maybe just three days to get something

44:46that's product ready so which is a really cool place we're at now here in in 2026 uh john yeah so i've really enjoyed uh what i've seen other people kind of showing off i haven't been able to make it for the uh formal session uh but the end like little demos that you can kind of see in the different ideas that are coming out of the community are really interesting

45:08um yeah so i i look forward to it every friday and i will try to get in there actually on time so i can participate with the group and not just kind of do the show and tell afterwards words. Great guy. Yeah, I'd love to see how to hand off between a GitHub project where you can, if you're using up all your tokens on one frontier model, how to pass that work to another one

45:38cleanly. Yeah, because it's one of those things where once you learn from other people's workflows, flows there's the hence the you know the community water cooler where it's just like what are you doing because i just put on this uh lm studio bionic and i have a 3090 card 3080 card in here and i was asking it like hey what what might a 5090 card cost and it's like uh as of now 2026

46:03the nvidia has not officially released or launched the geforce rtx 1590 so you know this this is quen 3 vl 8b instruct so when you download these these local models they can be very old and so if you're asking it well the first question i asked is what's the uh uh how many people are what's the capital of spain and it said the capital of spain is madrid i'm like okay cool it's working because

46:30you could see i had some errors and then how many people live there and you know so for a question like that it gets it right like population of madrid 6.7 million people gross domestic product output, blah, blah, blah. But then I ask it for something like what's the price of an Nvidia 5090. And it just doesn't even know that it is shipping.

46:47So that's the other thing is just how do you maximize the use of using your own local LLM so you're not burning up tokens. Cause if you have one of these cards and you're just doing basic stuff, like some of the stuff that we're doing with the short studio, I'd love to just pass it onto my local machine while it's camping in the middle of the night, let it not

47:03cost me anything except my electricity bill. well and if memory serves you've got some solar panels so um i'm guessing that's not the same as my power bill because i don't have any john what you got i was gonna say uh make sure guy that you're using the llms for what they're supposed to be used for if you need real-time information it has to be something that's online and then has some sort of rag model that's pulling in that

47:29real-time information to go alongside the data set it's trained on um but yeah i i find it really really interesting what we can do now with the local models and uh starting to test some um at the workplace uh haven't gotten too far with the yeah but um it's really interesting because we have a partnership with uh cisco and so they're giving us some access to some gpus that we can