Mercutio
Fatwah on Western Digital
My favorite part of that picture is the way Bill's mouth is just a little bit open in the most simian way.
I'll change it some day when somebody pies Elon Musk. I don't think he'll take it as well as Mr. Gates.
Something Random
- Thread starter ddrueding
- Start date
My favorite part of that picture is the way Bill's mouth is just a little bit open in the most simian way.
I'll change it some day when somebody pies Elon Musk. I don't think he'll take it as well as Mr. Gates.
Mercutio
Fatwah on Western Digital
The biggest gamer I know is a 70-something year old divorced woman. She's been getting her hair cut based on her headset needs for about the last 15 years. She told me last night that she's been advised to stop gaming because of ongoing joint pain in her hands and neck, so she's asked if I want the gaming PC I built for her in August of 2025 back. What I did not know about her is that she's independently wealthy. Her ex-husband was a commodities trader on Chicago's Mercantile Exchange (which I did know; I just didn't put together that the divorce went very well for her). She has an eight-figure net worth. She's giving me her gaming rig. I just have to make an 80 mile drive to get it.
Ryzen 9800X3D, 64GB DDR5, 6TB of NVMe, an RTX5080 and a 49" Samsung G9. Even the audio setup is amazing; Beyerdynamic headsets and Mackie studio monitors don't grow on trees and most of these are unobtainium parts. I'm talking about a PC that was $5k last summer, before all the pricing shit happened.
Hello car down payment.
Ryzen 9800X3D, 64GB DDR5, 6TB of NVMe, an RTX5080 and a 49" Samsung G9. Even the audio setup is amazing; Beyerdynamic headsets and Mackie studio monitors don't grow on trees and most of these are unobtainium parts. I'm talking about a PC that was $5k last summer, before all the pricing shit happened.
Hello car down payment.
Mercutio
Fatwah on Western Digital
Guys, it was just a regular Ryzen 9800X not an X3D, and she decided to keep her giant monitor. Shucks.
AI Bullshit updates:
nVidia and Perplexity.ai released a new tool over the weekend for on-device AI via RTX GPUs. It's supposed to offer Agentic tools. I have access to Perplexity Pro; I found a 2-year subscription (thanks, G2A) for a total of $50, hoping I'd be able to try it, but it doesn't include any compute tokens, which are apparently necessary to mess with it. I suppose I'm just going to have to get around to installing some kind of local AI model if I want to try a "do this for me" tool.
Scripting tools have been my best use of LLMs generally. I have access to "pro" Gemini and Claude. I've tried some other cases, like "what's making this noise that's coming from my car in this video?", a domain where I really could use expert knowledge, or "how could I better organize the space in these photos?" and I haven't seen it do anything terribly helpful for either query. But if you ask it to build a file organizing tool with multiple metadata search hooks, rules for 300 file types and multiple decision paths, man, they'll spit that out with about 20 minutes of conversation.
AI Bullshit updates:
nVidia and Perplexity.ai released a new tool over the weekend for on-device AI via RTX GPUs. It's supposed to offer Agentic tools. I have access to Perplexity Pro; I found a 2-year subscription (thanks, G2A) for a total of $50, hoping I'd be able to try it, but it doesn't include any compute tokens, which are apparently necessary to mess with it. I suppose I'm just going to have to get around to installing some kind of local AI model if I want to try a "do this for me" tool.
Scripting tools have been my best use of LLMs generally. I have access to "pro" Gemini and Claude. I've tried some other cases, like "what's making this noise that's coming from my car in this video?", a domain where I really could use expert knowledge, or "how could I better organize the space in these photos?" and I haven't seen it do anything terribly helpful for either query. But if you ask it to build a file organizing tool with multiple metadata search hooks, rules for 300 file types and multiple decision paths, man, they'll spit that out with about 20 minutes of conversation.
jtr1962
Storage? I am Storage!
You've helped so many people. It's nice to see something great like this come your way!
sedrosken
Florida Man
As much as I view AI as a copyright infringement machine, I do have Claude Pro (at least for the rest of the month, unsure if I'll renew) and I have used it for automating some stuff like writing out SQL to make new items on a batch basis (when I need lots of a similar thing, I'll make the first one and have it extrapolate) for an Asheron's Call server I'm currently working on some custom content for. It's also been useful tracking down why certain things won't build from source when I try to build them. I'm probably boiling off a small lake to do it. I've tried OpenWebUI and LM Studio which as I understand it are your main options for local AI models, and I do prefer OpenWebUI as it's a lot closer to what I need, but with a Turing-gen quadro with "only" 16GB of VRAM, useful options are pretty thin on the ground.
Mercutio
Fatwah on Western Digital
Given how much of a PITA my car has been, everything I just got is getting dumped immediately on someone who can actually use that rig. Even if I gave somebody a $2500 bargain deal, that's still a sizeable chunk of extra cash I can put in to my purchase.
Also, I'm finding that car salesmen LOVE to call on their personal cell numbers. My phone has been on silent for a couple weeks now since it turns out I hate talking to these people.
I was able to set up an Ollama cluster among four of my game systems. In about a half hour, I had a setup where I can interact with at a prompt and it'll farm work out to the other instances. I have that agent currently chewing on application of my data organization scripts and in particular better outcomes for metadata tagging, like more reliable selection of (comic- or e-) book covers and closer-to-human fuzzy title and author lookups leading to more predictable, concrete file tags. I went through a MASSIVE amount of pain doing this for my music data. I'm just seeing what these systems can accomplish with books and comics.
FWIW, I think Gemini makes better code than Claude or Perplexity, but I REALLY like how Claude gives me an interactive editor. Perplexity writes code I can tell at a glance isn't going to run on the first try, but it gives better feedback in terms of the reasoning it's using in its responses than either of the others I've tried. That may be down to me not forcing it to use one model or other.
Gemini, being the thing hooked in to Google, also sometimes REALLY surprises me with things it knows because it's the back-end of Google Assistant now.
Also, I'm finding that car salesmen LOVE to call on their personal cell numbers. My phone has been on silent for a couple weeks now since it turns out I hate talking to these people.
I was able to set up an Ollama cluster among four of my game systems. In about a half hour, I had a setup where I can interact with at a prompt and it'll farm work out to the other instances. I have that agent currently chewing on application of my data organization scripts and in particular better outcomes for metadata tagging, like more reliable selection of (comic- or e-) book covers and closer-to-human fuzzy title and author lookups leading to more predictable, concrete file tags. I went through a MASSIVE amount of pain doing this for my music data. I'm just seeing what these systems can accomplish with books and comics.
FWIW, I think Gemini makes better code than Claude or Perplexity, but I REALLY like how Claude gives me an interactive editor. Perplexity writes code I can tell at a glance isn't going to run on the first try, but it gives better feedback in terms of the reasoning it's using in its responses than either of the others I've tried. That may be down to me not forcing it to use one model or other.
Gemini, being the thing hooked in to Google, also sometimes REALLY surprises me with things it knows because it's the back-end of Google Assistant now.
Last edited:
sedrosken
Florida Man
In retro build news, my Pentium MMX is finally stable. Biostar MB-8500TTD paired with the custom J.3 patch to fix several bugs with the IDE controller -- and an EEPROM with an actually writable config memory portion -- has this thing rock solid, even at a 75MHz bus. 83 can boot into Windows and run some benchmarks but isn't truly stable the way a 75MHz bus is. Makes sense -- at 83MHz the PCI bus is running at almost 42MHz, that's way out of spec. Even with cards that behave fine with that overclocked bus (my G200 PCI, a 3C-905C) I'm willing to bet the chipset is less than happy at that speed. Still, 262MHz out of a 233MHz CPU isn't bad. Even if I think it would have benefitted from a third multiplier pin a la Super 7 to be able to set a 4x multiplier -- the CPU feels like it's got more left in the tank, but I just can't physically push it high enough.
I'd still rather have a fast Pentium MMX machine than a slow, deliberately hobbled K6-2 or III machine. I have an actual sentimental attachment to the Pentium MMX -- a 200MHz MMX was my very first CPU, and it feels good to have a working machine around one again.
The Matrox G200 is a surprisingly capable card. The TNT2 M64 I also have in PCI was only marginally faster in this machine -- but then, both are probably being held back by the bus and by the CPU. I find it especially fun that I was able to find the VRAM expansion for my G200 online for less than 20 bucks because it was part of a server parts surplus -- the G200 really had a weirdly long history in enterprise. And something I'm running on there can actually make use of the extra VRAM, even -- the Dark Majesty client for Asheron's Call runs here, but at normal settings was complaining about low available texture memory. And there's an actual playable server that talks to the Dark Majesty client now, Aetherium.
Yes, I had to swap EEPROMs to get this thing to be completely stable -- the MX28F1000P that Biostar included would never write out the DMI table and would make anything that tried to access it (Speedsys, WMI, DXDiag) hang the machine. An AT29C010A is much better behaved. Current working theory on VOGONS is Biostar goofed up and bundled a chip that needs 12V to program -- the AT29C010A doesn't and has that pin internally disconnected.
I'd still rather have a fast Pentium MMX machine than a slow, deliberately hobbled K6-2 or III machine. I have an actual sentimental attachment to the Pentium MMX -- a 200MHz MMX was my very first CPU, and it feels good to have a working machine around one again.
The Matrox G200 is a surprisingly capable card. The TNT2 M64 I also have in PCI was only marginally faster in this machine -- but then, both are probably being held back by the bus and by the CPU. I find it especially fun that I was able to find the VRAM expansion for my G200 online for less than 20 bucks because it was part of a server parts surplus -- the G200 really had a weirdly long history in enterprise. And something I'm running on there can actually make use of the extra VRAM, even -- the Dark Majesty client for Asheron's Call runs here, but at normal settings was complaining about low available texture memory. And there's an actual playable server that talks to the Dark Majesty client now, Aetherium.
Yes, I had to swap EEPROMs to get this thing to be completely stable -- the MX28F1000P that Biostar included would never write out the DMI table and would make anything that tried to access it (Speedsys, WMI, DXDiag) hang the machine. An AT29C010A is much better behaved. Current working theory on VOGONS is Biostar goofed up and bundled a chip that needs 12V to program -- the AT29C010A doesn't and has that pin internally disconnected.