Coconut Monkey Cornerclub

[Zed Clampet]

WOW.

I have never used Claude before, but you can now open him up inside pinokio. Most everything in pinokio is broken, and I have been trying to fix them. It took over a week to fix one program, and CoPilot led me astray over and over again, so I finally stopped asking it. So today I tried to open up Klein and it errors out. Just out of curiosity, I pressed the "Ask AI" button, and Claude popped up on the screen. I copied the error message and posted it, and suddenly things started changing on the Klein screen and then it launched successfully. That's pretty damn impressive. And I especially love that Claude didn't even bother to talk to me. I had no idea you could bake Claude into software like this. Unfortunately, to continue with Claude is apparently going to require a subscription.

 

[BeardyHat]

Yeah, we've got red flag warnings down here - meaning no burning anything. It's about to drop about 30 degrees F and start raining off and on for about a week, though, which will help a lot. Maybe more than enough. If you folks want some, feel free to shift the center of the Earth over a bit until the rivers run the other direction.

Yeah, been seeing the red flags here every day and those two fires that have broken out seem to have been quickly contained, which I'm thankful for. Back in 2021, it was so dry we had a fire start and burn about half of my hometown; I haven't lived there for 20+ years now, but some friends parents houses did burn down.

We need less rivers really and just some more snowpack to really help us out. I bet we'll be under tight water restrictions this summer. Though, funny, residental water use is very mild compared to what farms out East in our state use.

 

[BeardyHat]

WOW.

I have never used Claude before, but you can now open him up inside pinokio. Most everything in pinokio is broken, and I have been trying to fix them. It took over a week to fix one program, and CoPilot led me astray over and over again, so I finally stopped asking it. So today I tried to open up Klein and it errors out. Just out of curiosity, I pressed the "Ask AI" button, and Claude popped up on the screen. I copied the error message and posted it, and suddenly things started changing on the Klein screen and then it launched successfully. That's pretty damn impressive. And I especially love that Claude didn't even bother to talk to me. I had no idea you could bake Claude into software like this. Unfortunately, to continue with Claude is apparently going to require a subscription.

Out of curiosity, do you ever get accurate and working links from AI? Every AI I've used can never give me a working link and when told the link is broken and they're hallucinating, they just hallucinate another broken link.

I've even just asked it to cite its sources as far as city codes go and it'll randomly make up a code number and cite it, which I am completely unable to find.

 

[Zed Clampet]

Out of curiosity, do you ever get accurate and working links from AI? Every AI I've used can never give me a working link and when told the link is broken and they're hallucinating, they just hallucinate another broken link.

I've even just asked it to cite its sources as far as city codes go and it'll randomly make up a code number and cite it, which I am completely unable to find.

I almost always get working links when I request links. Usually I don't request them unless I'm going to share the information with others. I believe that the reason it sometimes gives bad links is because of a conflict in what AI was designed to do and how it is currently being used. AI is required to give you an answer no matter what. AI's logic was designed for research and simulations, not to be a search engine. If AI was concerned about accuracy, it probably wouldn't be as good as it is in simulations and formulating all the new drugs it has come up with. It has a 90 percent success rate with new drugs. It's actually far better at that than it is at trying to act like a search engine.

Eventually they'll work this out, but they'll worry about it later. Right now they have a major crisis on their hands. When you go to ChatGPT and ask a question, it can theoretically use hundreds of GB of VRAM just for your one question. I suspect they are in efficiency panic mode as more and more people use AI. They've already grabbed every GB of RAM and VRAM they could get their hands on, but that's just a band aid. To be viable in the near future, they have to be far more efficient.

 

[BeardyHat]

I almost always get working links when I request links. Usually I don't request them unless I'm going to share the information with others. I believe that the reason it sometimes gives bad links is because of a conflict in what AI was designed to do and how it is currently being used. AI is required to give you an answer no matter what. AI's logic was designed for research and simulations, not to be a search engine. If AI was concerned about accuracy, it probably wouldn't be as good as it is in simulations and formulating all the new drugs it has come up with. It has a 90 percent success rate with new drugs. It's actually far better at that than it is at trying to act like a search engine.

Eventually they'll work this out, but they'll worry about it later. Right now they have a major crisis on their hands. When you go to ChatGPT and ask a question, it can theoretically use hundreds of GB of VRAM just for your one question. I suspect they are in efficiency panic mode as more and more people use AI. They've already grabbed every GB of RAM and VRAM they could get their hands on, but that's just a band aid. To be viable in the near future, they have to be far more efficient.

Ah, alright, that makes sense.

I mean, they gotta figure out how to monetize it better too, I'm assuming. Much as I enjoy using AI though, I am hoping the bubble bursts so we can go back to relatively affordable prices and availability.