• Moderators, please send me a PM if you are unable to access mod permissions. Thanks, Habsy.

Tech Thread

giphy.gif
 
MindzEye said:
There's shit this stuff can do well and stuff it can't. Case in point from this morning. I use Claude to build python and json scripts that allows me to do huge batches of images for one of the business lines I have and solve other fairly technical problems that I could do but would take me way, way more time to do. Customer organization is upset that because they rescheduled our shoot with them twice, the delivery date of the physical prints would now be past the end of their season. Clearly our fault for not telling them that delaying the shoot for a month would also mean that the physical prints from said shoot were delayed for a month. So now they're demanding that we fix this by delivering the orders (87 of them) or paying to have them mailed/shipped out of our pocket. For shits and giggles I took our order CSV, threw it into Claude and had Claude map out the addresses, pull Google Maps API into the loop and build the most efficient delivery structure. 30 seconds, and because I supplied all of the data and it didn't have to go looking for anything, everything address is confirmed as accurate. Manually I would have done a worse job at mapping this out, and it would have taken me hours.

But yeah, if you're going to Claude and asking it to build you a meal or workout plan and you don't specify that you only want it to include sources that are high quality controlled studies, you're going to get it mashing together a bunch of shitty blog top 10 lists into a fucking abortion of a plan.
Click to expand...

so you got claude to look up delivery prices.

did you check the output?

how many mistakes would claude have to make to eat your timesavings?
 
CH1 said:
forgot to mention that I also use the dryer for bed related stuff

And by the way, my clothes last a freakishly long time
Click to expand...
Dammit, I used to do that years ago but then said fuck it but now I think I have to go back to it.

The problem is, where do you hang it all? Unless outside, it is a headache indoors, unless you have a legit laundry room with hooks on the walls specifically made for hangers.
 
zeke said:
so you got claude to look up delivery prices.

did you check the output?

how many mistakes would claude have to make to eat your timesavings?
Click to expand...

No, I got Claude to map addresses for manual deliveries because it's easier for me to pay to have one of my employees and myself take half of the route each and hand deliver the orders all in an afternoon than it is to ship each order out at 7-8 bucks each.

I've gone over each line in the CSV it generated and the route maps it generated. All accurate and good.

To answer your question..lots. I'm not a logistics and delivery company. Mapping out and delivering to almost 90 points of call isn't something I'm set up to do. But I understand how technology works under the hood. If I provide it with a strict data set, and tell it what I want to go (connect with Google maps API, input the address lines into the API and report back Google's suggested routing) it will do that with good accuracy. If I asked it "how do I do this thing" it will search the internet, pull information from outside of its training data and probably hallucinate at me a bunch.
 
MindzEye said:
No, I got Claude to map addresses for manual deliveries because it's easier for me to pay to have one of my employees and myself take half of the route each and hand deliver the orders all in an afternoon than it is to ship each order out at 7-8 bucks each.

I've gone over each line in the CSV it generated and the route maps it generated. All accurate and good.

To answer your question..lots. I'm not a logistics and delivery company. Mapping out and delivering to almost 90 points of call isn't something I'm set up to do. But I understand how technology works under the hood. If I provide it with a strict data set, and tell it what I want to go (connect with Google maps API, input the address lines into the API and report back Google's suggested routing) it will do that with good accuracy. If I asked it "how do I do this thing" it will search the internet, pull information from outside of its training data and probably hallucinate at me a bunch.
Click to expand...

so claude inputs all the addresses into google maps for you and gives you a map. fair enough that looks like a solid use case.

but that's just claude using the google maps import feature, no? still useful either way i guess. as long as it doesn't screw up.
 
LeafOfFaith said:
This obviously belongs in the tech thread - anyone know a thing or two about washer/dryers?

Lately my dryer is for sure shrinking my clothes. I haven't changed any setting, but it goes in fine and comes out smaller. I tried to experiment by drying with cooler temp or setting to delicates rather than normal, but those don't dry properly. I feel like any repairman I bring will just waste my time and money because he'd say it's drying hot and coming out dry and fine. There must be something to try short of having to hang shit up and air dry.
Click to expand...
Eat less
 
LeafOfFaith said:
Is that what you do? I was literally thinking earlier about it, but then you're wearing oversized and hoping it'll shrink down.
Click to expand...
For casual clothes that can work. You can also get pre shrunk clothes that helps.

Set your washer to high spin. That helps get things dry so it spends less time in the dryer.

Do you have timed dry or the "sensing" mode?

Are these new clothes?
 
BG said:
Could actually be a problem with the dryer or the dryer vent. Your dryer vent may be plugging up which would keep a lot of humidity in the drum, basically steaming your clothes. Could also be that the thermostat or curcuit board is fried, and the dryer is just operating at high heat regardless of your settings.

Could also be womens jeans.
Click to expand...
Good call and not just the vent trap check the actually vent hose and where it goes outside
 
zeke said:
so claude inputs all the addresses into google maps for you and gives you a map. fair enough that looks like a solid use case.

but that's just claude using the google maps import feature, no? still useful either way i guess. as long as it doesn't screw up.
Click to expand...

It's taking 87 lines of data from a CSV that I drag/drop into it, connecting with Google Maps API so that it can input all 87 at once (the consumer facing google maps maxes at 10 lines), letting google maps figure out the most efficient route to take, then Claude took that data and put it into a point of call list for me broken up into 2 at the most logical place for 2 people to tackle separately with google map links for the routes.

The CSV's it spit out are accurate (because I gave it a defined data set to work with and it just designs and runs a python script in the background on it) and the routes are 2 google map links that I can click and it opens up in a browser window or in the google maps app.

Like I said, you have to know how to use it. What it's good at, what it's not, where the vast majority of it's hallucinations come from, etc. It's a force multiplier for someone with an already functioning clue, not a reliable replacement brain for someone with no knowledge in the subject that they're looking for information in.
 
Artificial intelligence products have completely changed the game for me on a personal level. The fact that it doesn't replace a brain makes me love it even more quite frankly. As humans with above average intelligence, we want to maintain that edge and AI allows one to increase the intelligence gap, if anything. It's a powerful tool, but dumb people won't know how to leverage it.
 
Actually you can upload thousands of addresses into Google My Maps from that same CSV file but yeah that's still a good use case.

as long as you trust claude's opinion on how to split up the map!
 
zeke said:
Actually you can upload thousands of addresses into Google My Maps from that same CSV file but yeah that's still a good use case.

as long as you trust claude's opinion on how to split up the map!
Click to expand...

Quick review of the maps is all it took to see if it passed the smell test. It's fine.

Fwiw, I'm still very much an AI industry skeptic and don't see the industry surviving in anything resembling it's current form (and creating a historic amount of wealth destruction across the broader economy on it's way out the door), and the technology has some severe limitations that will keep it from being more than a small fraction of what AI boosters claim it is. But it's not strictly useless and if you know what it's good at, while knowing what it's not good at, it's a pretty big win for small businesses imo. I've used it to build custom scripts that have bolted github shit I didn't know existed onto it, that work brilliantly. For example, I have a script that uses code from a German project called tesseract, that was designed to read scans of old books, for decoding QR codes. The industry solution for this is a meta data encoding bluetooth dongle that costs 300 per dongle and requires a 100 dollar a month subscription for me to run on site, all while being buggy because it requires functioning wifi on site. I'm able to skip all of that and run this script over my photos after the fact, it decodes the QR, pulls the data from it, renames the file in my preferred naming convention based on the data baked into the card, and places it in a pre set folder structure for processing. Did it take 3-4 iterations of the code to get it bang on? Yes. Does it work brilliantly now? Also yes. Once the script is written, you're not interacting with the LLM, it's just a python command.
 
Yeah for small batch uses i can see the timesaving.....but would still have to measure the cost of the AI, the time it takes to build something reliable, the time it takes to review, and the potential risk of error and future time spent fixing or rewriting what you've built. But that risk is lower in small scale usage.

I guess a test would be - would you try and sell whatever you've wrtten or built with AI to another business?
 
worm said:
For casual clothes that can work. You can also get pre shrunk clothes that helps.

Set your washer to high spin. That helps get things dry so it spends less time in the dryer.

Do you have timed dry or the "sensing" mode?

Are these new clothes?
Click to expand...
Yeah, it has sensing mode, but I did notice lately that it wouldn't fully dry after the cycle, so then I'd put it in for another 10 minutes or so. But that's with towels and bedding and delicates. I wouldn't do that with the jeans specifically to avoid the shrinking. I guess if the sensing mode is fucked up, then the whole thing is fucked up, but I figured it was underheating rather than overheating so I don't know what causes things to shrink.

What sucks is that bringing a repair guy is probably half the cost of a new dryer and no assurance that he'll actually diagnose or fix the problem.
 
zeke said:
Yeah for small batch uses i can see the timesaving.....but would still have to measure the cost of the AI, the time it takes to build something reliable, the time it takes to review, and the potential risk of error and future time spent fixing or rewriting what you've built. But that risk is lower in small scale usage.

I guess a test would be - would you try and sell whatever you've wrtten or built with AI to another business?
Click to expand...

Cost of the AI would be 28 bucks. I needed Claude Pro for 1 month to generate everything I built. Built something reliable in a night. I had it bake in a test feature into the script, so I can run it against a bit set of a few thousand images, but have it only process 10 of them to test. Ran a few tests of increasing size (test time is largely determined by how many images and what your hardware is), built in all of the weirdo edge case logic into it (what does it do with failures, results that score below a certain surety score that tesseract has inherent to it, etc) and now it's a folder with a few scripts in it that I run either out of python or photoshop depending on the specific use).

I mean, what's the concern? That the script is going to stop working? That's kind of not how scripts work mate. Once it's coded, it's coded. You just run the code, I don't need Claude to run the code for me. It's entirely out of the production loop after it's written code for me that works.

Would I try to sell it? This is niche code to solve a niche problem in a industry of largely mom and pop providers. The time just the read/rename/reroute script saves me on a big job is about 6-8 hours (or the alternative cost and subscription mentioned previously). My production scripts save me a 700 dollar year subscription to an automation system for this stuff, or literally using Photoshop's busted ass legacy data set function in small batches myself. So 6-8 hours of letting my computer run more or less unattended to produce a few thousand photos, or doing a 100 image batch, running it, coming back, loading another CSV, running another batch, over and over again. Actually, that's probably the best comparison here. My set of production scripts replaces 95% of the functionality of a 700 dollar a year commercial subscription and does the work in the same amount of time. If I wrapped this all in a UI, I could probably sell it, yes.
 
LeafOfFaith said:
Yeah, it has sensing mode, but I did notice lately that it wouldn't fully dry after the cycle, so then I'd put it in for another 10 minutes or so. But that's with towels and bedding and delicates. I wouldn't do that with the jeans specifically to avoid the shrinking. I guess if the sensing mode is fucked up, then the whole thing is fucked up, but I figured it was underheating rather than overheating so I don't know what causes things to shrink.

What sucks is that bringing a repair guy is probably half the cost of a new dryer and no assurance that he'll actually diagnose or fix the problem.
Click to expand...
We have been using wool balls for a while now and it cuts down drying time
 
You must log in or register to reply here.
Back
Top