I want 31B. The 12B 4-bit QAT is already small enough to run well enough on every device I use regularly, including phone and tablet, I don't need a 1-bit or ternary version of the 12B.
But, what I really want is for Google to release bigger Gemma 4 models, particularly a bigger MoE, like a ~70B or ~120B. Gemma 4 is the best all-rounder among the models I can self-host even though I've got a 128GB Strix Halo. A 4-bit QAT version of a 70B MoE would probably be the sweet spot.
A bigger Qwen 3.6 with a 4-bit QAT version would also be welcome, as the prior bigger versions aren't notably better than 3.6 27B, but I guess Qwen is done doing larger open weights models. They did release AgentWorld recently, a post-train of the 3.6 MoE, so they're still doing some open things.
I think I want to see more third-party testing of this ternary Qwen to know if crushing it to 1.56 bits kills it; there are tons of benchmarks of Qwen 3.6 27B, so it's an ideal candidate to figure out what the extreme compression does to it.
Sounds very interesting. So I’m interested in learning more about the environments goals and focus. Could you tell me more about the type of educational content you envision prioritizing and the audience it’s meant to serve? As well, I’d love to understand how artificial intelligence will specifically support users within this educational environment. Give a use case? How will the platform bridge the gap between education and the labor markets?
Are there goals for partnerships with particular companies or industries to support this? Oh and I’m also curious if the platform or whatever will focus on developing skills for specific professions or if it has a broader educational scope.
I'm a big fan of checklists (Checklist Manifesto is a great book).
I think one of the difficult things with checklists is by their default they assume happy path (step 1, then step 2). The reality is step 1 might spawn 13 options. I'd kill for a git-style (or mini-drakon-style) visualization of checklists that are slightly more workflowy.
We've found that there's some variance in the shape of workloads users have (some doing lead gen, so pricing per lead makes sense; and others uploading CSVs of their own leads and only want to use our AI research tooling, in which case usage-based pricing might be more appropriate), so we like to chat briefly to get a sense (at least until we figure out clean pricing tiers.)
