A 27B ternary on Qwen3.6 would rewrite open source: flagship coding + reasoning in a few GB, blazing inference via additions on phones/laptops/anything. Ubiquitous local AI, zero cloud lock-in.
Yes, post-training ternary quantization (Bonsai-style or newer PTQ like CAT-Q) makes it doable for far less than 1-5M—hours on a handful of GPUs, not full retraining. The 8B exists because small teams shipped it; 27B just needs the same focused push on kernels + evals. No conflicts, pure upside. Who's funding the next Bonsai?
@Star_Knight12 Deepseek gave us the preview for free for 3 months brother, let another open source model get some shine, Deepseek is raising prices because they are releasing the actual Deepseek v4 and NOT the preview we've used since April.
@Vivek4real_ This dude brought us the Twitter that Elon exposed. Why are you listening to anything he has to say? He's saying the right things now because he got caught.
We already have the solutions to many of the worlds problems. The real bottleneck seems to be implementation at scale without being undermined.
I would focus the intelligence on `Systemic incentive misalignment and large scale collective action failures.` Solving (or dramatically reducing) incentive misalignment would unlock deployment.
Achieving technological post-scarcity would also make everything cheaper to solve so it's a good runner up.
@MKUltraMoney It's not terrible, it's a step up from recent efforts in the movie scene but it's like a mid pre 2010 film. Set your expectations accordingly.