Michal Zygmunt

@davepl1968 Double RTX6000? How much did you spend and how your setup looks like? You put them in normal desktop computer? It's some dedicated server etc?

OK, I installed OpenCLAW. I set it up with a backend AI server on a Dell 7875 workstation with dual Blackwell RTX6000 GPUs running DeepSeek R1 32B. And it can fall down to Qwen 2.5 for easy stuff. I created an agent to scour eBay for things I might be interested in, compile a summary, and mail it to myself. But I could have done that in shell script... so... what now? What are folks doing with it that's interesting? What are you having your agents do?

Interesting into disclosure

Good one

@stevesi Great post Steve!

You would find the same level of performance on the same "typical" hardware (the 486dx was the latest and 16MB was 8x the system requirements) on each release of Office through 2007 (95, 97, 2000, XP/2002, 2003, 2007). There was a major rearchitecture to share code in Office 97 which changed the multi-app scenarios for the better. In 2002 we tuned for the NT kernel (even though by 2000 all self-hosting was NT, customers had not moved yet so we still supported W95). Somewhat confusingly we still supported 9x on Office XP as well which genuinely annoyed people internally. O2003 was NT only. O2003 ran on W2000+. As a rule of thumb, Word, Excel, PowerPoint remained roughly 2-6MB working set that entire time, including min footprint when app was minimized. Even the move to 32 bits did not double system requirements. Outlook definitely changed the equation as did running multiple products at once. For a while system requirements were "per app". Then in 97 Outlook was called out separately as requiring more. Since it was running all the time it did change the memory profile of the suite. With a browser running all the time and consuming a lot of memory (Win95+ along with the networking stack) we started to see competition for memory on low end (4-8MB) systems. The move to NT reduced this but increased total system footprint. AMA...I can't tell you how carefully this was measured all the time and debated constantly. Win95 in particular was a huge "battle" between Windows and Office over what the system requirements should be and what constituted typical use in order to support those. Consider this essay on the box :-) For more of this epic experience see -> …rdcoresoftware.learningbyshipping.com/p/035-windows-…

@davepl1968 I also hate UI. Much better to concentrate on backend code:-)

I cranked out this very complicated nested-tables management UI with Codex in about 3 hours. I absolutely HATE creating browser UI, so I'm hugely impressed and relieved by what Codex/Claude is able to do in plain CSS and HTML!

@davepl1968 Nobody invested in optimizing this rdp protocol in same way as Microsoft did for Windows.

Ok, I installed a GUI! But why is it so slow remotely? I'm using RDP and it draws really glacially compared to Mac Screen Sharing or Windows RDP?

I've been running Linux since the early 90s and have never had a mouse connected. The GUIs are terrible. Spectacular operating system headsless, though! But if you can't do it by ssh, it's not happening.

This short film made with Seedance 2.0 is absolutely insane. The realism looks like a real movie — no one can tell it's AI.

@davepl1968 Cool!

I'm oddly proud and stoked about this one - live video preview w/ audio of the Robotron AI training on the web. You can watch it live here: davepl.com:8796 I started simple yesterday, a canvas I updated with a snapshot of the game that I pumped over a socket from an emulator session. That evolved into a video tag with a properly encoded video stream, and today I spent a full damned day almost adding audio and keeping it in sync with the game (which runs at an uncorked and unpredictable framerate). Assuming my machine can handle the load, it's available for viewing pleasure now at the link above :-) You have to click on the preview to enable the audio, and you can turn it off/on at the bottom of the page. Thanks to @dsobeski for beta-testing it!

@davepl1968 He looks younger then you! And no beard 🤣

I'm older now than my Dad ever was. But here's what he'd look like taking a selfie, if they'd had the iPhone in 1959.

This, my friends, is called timing.

Joe Lycett Parking Fine🤣🤣🤣🤣🤣

@davepl1968 Your setup looks crazy nice!

@lahcim2000 I've got a dedicated 20A for it, but am having a second one added before I bring up the extra drives!

The DEC homelab grows by another couple of racks. Today's additions include the upright TS11 tape and three RA60s, an RA81, and two RK05s (not pictured).

@davepl1968 What hardware your are running this on Dave?

My AI performs 10,000+ frames PER SECOND of inference AND training. Every move made by 100 simultaneous Tempest games at 100fps is managed by the central server, responding on average within 3ms. The data exchange is between Python and LUA, not the fastest languages out there, so I used standard TCP/IP sockets and a pure binary interface. No REST, no HTTP, no JSON. Just speed.

@lcamtuf @InfZakladowy Haha. So true

Pamiętacie, jak drogie były kiedyś komputery i oprogramowanie? 30 lat temu Microsoft Office kosztował trzy pensje minimalne, dziś - mniej niż ćwierć jednej pensji. Nowy podcast o historii rynku pecetów w Polsce już za kilka dni, zapisz się na newsletter! informatykzakladowy.pl/newsletter/

Good system

@stevesi Thoughts on this article @stevesi : citriniresearch.com/p/2028gic ?

AI is currently in the phase equivalent to: * creating HTML w/notepad * editing Windows registry/managing Mac INIT/CDEV * sharing favorite macros/shortcuts * dropping into a CMD shell All of what we use today and think of as a user model will change dramatically in 2-3 years.

Interesting read and AI future and one of potential impacts on world economy: citriniresearch.com/p/2028gic

Cool

Interesting new chip