Victor Laynez

I've complained about this for sure... @Apple makes it impossible to get your photos off their phones. Zero way to validate a bulk transfer without wasted hours doing this manually. This is moat behavior and why people avoid these products. This should be a one button click and come back later for as long as this poor filesystem takes.... ... and no, I should not have to signup for iCloud to get my photos off my phone. Apple fanchilds, don't start. :)

Any guidance for folks using Qwen 3.8 27B locally for light coding agent and online researching / scraping? How far are you pushing this model locally?

So my eyes are decent but I wonder with receipt printers, given the flexibility of printing in different fonts, sizes, and even QR/graphics…. The total price has to be the size of a grain of rice?!?! Looking at you restaurants!!

Ladies & Gentleman, A moment accesable around the world and read by 7, in the second round of advanced competition, we have a reversal. Some will like it, some will hate it, some might be scared, but most will simply scroll past it... Qwen 3.5 9B wins the day. $1

I will not get into the AI benchmark game... but IF I did I would create pay-per-view and charge $1 and find soemone's lawn to perform it in live. :) Interesting bit here is that this is with a single 1080Ti. A second will be added soon. Bonsai 8B vs Qwen 3.5 9B.

I'm building lamma.cpp for two old cards in an old machine for Qwen 3.8 27B and prepare for Qwen4 27B. Of course I have opencode helping me out. An EE traveling SO far outside of my domain experience right now. :) We'll see if the machine lasts the night.

Picking up a 2nd 1080Ti. Any reason to think I can’t run Qwen 3.8 27B with lamma.cpp?

Trying DeepSeek-V4.1-Flash and it does seem to be more precice. I dont know what heappened to V4.0 flash recently but it got verbose, unsure, and did I say verbose? :)

<< MAJOR MILESTONE >> I think I've developed slightly enough experience with AI tools to have more tasks than tokens. Token anixety has now begun. #tokenanxiety

I like the new OpenClaw 2.0 so far. A smidge of a pain to get migrated but such is life. Matrix / Element X, all plug-ins, and the 4 agents are alive. Good work! @openclaw WebGUI feels more intuitive. More soon as I try it out more.

Is OX-Alpha down? It looks like it's not responding. Free period over perhaps?

I asked my agent (Alfred) to look back over the year of session data and offer an image of what it thought it looked like:

Any of my followers using Qwen 3.8 27B? Thinking about picking up a second 1080Ti to attempt to run it. No tensor, but 22GB VRAM might do it?

I've been having fun with local AI and forcing it into other domains/hobbies :) I see an analogy here with early microntroller design. I'm on interrupts now, soon efficent periperals, after that accelerators per fucnction. I don't think this is an OS, I think its a 16b uC.