Viewing profile — androiddrew

I'd really like to see a 45B-ish dense model ready for a dual GPU setup. Something with a little more intelligence while still within the range of some higher end local setups.

I wish we could stop sensationalizing this about the AI and really just understand the incompetence of the labs disabling an internet connection in a sandbox.

So I always make sure I take a crack at doing what I want first. Then I ask for an AI review and it usually has a more efficient way to get the job done. For example I had a workin…

I wonder if this would work with the oculus pro?

So I'd like to see a Nemotron3 Ultra converted to a 1.58bit format then have them retrain on the open dataset.

Where does someone start on kernel development for an RDNA4? Books, resources, whatever? 8 year developer just getting into to AI side of the house

I have only been using 3.6 27B for coding. Is mem0 for agents like Openclaw or Hermes? How are you using it?

Curious, do you find the 3.5 120B sized MoE works better than the dense 3.6 27B?

I have been running 3.6 27b on a dual AMD r9700 setup using Opencode and Matt Pocock's skills workflow for writing Golang CLIs. It's decent, but won't win any awards on code archit…

Lolz, just forward to my friend who works at Planetscale. Looking forward to his reaction

Give me a good 180B param model that fits snuggly on an single DGX spark and I will sing your praises.

I love Alex Edwards. His books and Learn go with tests were my first introduction to the language. Still recommend to this day. I'm feel inspired to convert some old stuff to HTMX

A model harness can launch a bagillion sub agents and we still call it one shotting.

Yeah 100% they were going to do this anyways.

I mean I wouldn’t want to work for Andrew Kelley either. Doesn’t mean I don’t see utility in zig. Taking swings at a pretty toxic culture (silicon valley) while refreshing also pai…

Yeah and Andrew Kelley is anti AI for his project because it’s counter to the projects learning goals. I think it’s perfectly fine for a project to determine if AI contributions ar…

The author being the author of zig…

[flagged]

I try to always mention that AMD ROCm has come a long way. Like the B70 the Radeon AI Pro 9700 has 32GB of DDR6 640GB/s. Also $1300 a card. Very capable cards now in mid 2026. Grea…

Dual AMD Radeon AI Pro 9700s (600 watts total 64GB of vram) runs Qwen 3.6 27B at FP8 with mtp on vLLM at 50ish TPS for decode. Cards cost $1300 a piece. Enough KV cache to fully ma…

Yeah. The "I don't care" line from The Fugitive comes to mind.

An old steam link still works for me

Ummm probably not. Lock ups are going to dump far more stock into the market.

No, they are still in the disrupt phase of the strategy. They have enough money to operate at a significant loss. So the play is to heavily subsidize, ingrain until a business can’…

Yeah, but burn rates are high. The money is mostly gone by now, and the run way is approaching its end in the next 12-18 months. Hence the “great devaluation “ they are predicting