How do you feel about claims that Kimi K2 supports lossless long context? I've seen these in other articles about the model, but haven't seen anything official from Moonshot about it (just references to Chinese language interviews with Yang Zhilin where he doesn't seem to actually claim this).
I don't have the technical chops to fully investigate the claim, but Kimi K2 does seem to have some innovations that may help reduce cognitive dilution.
Love the verifier economy concept. How feasible would it be for each person to have their own ‘verifier’, such that they could scale their expertise to say validate work?
Your diagrams and discussions provide high level perspectives of the AI development terrain which shows that innovations that emerge from intrinsic motivation and self-less creativity can be disruptive.....rather than massive scale approaches based on extrinsic rewards and projections of dominance... The AI-hype race to AGI is not focused on solving real-world problems...there are so many real world problems that can be addressed by humans and weak-AI systems collaborating based on real-world constraints. The system level perspective as the logical next step is a little comforting given that holistic models are necessary for solving complex problems like climate change. The insight about interactivity as necessary for dynamic systems and the interdependency of the components starts to align with intelligence in living systems...dynamic, complex-adaptive systems. I hope humans can appreciate natural intelligence of living systems before we unleash destructive AI models that further disconnect humanity from the ecosystem of the living systems in the universe and destroy the environment beyond our capacity to mitigate the damage all in the race for claiming dominance in the race to AGI.
Thank you! I am new to Substack. Love the article!👏
Lots more here
How do you feel about claims that Kimi K2 supports lossless long context? I've seen these in other articles about the model, but haven't seen anything official from Moonshot about it (just references to Chinese language interviews with Yang Zhilin where he doesn't seem to actually claim this).
I don't have the technical chops to fully investigate the claim, but Kimi K2 does seem to have some innovations that may help reduce cognitive dilution.
Morning remarkable I've seen so far. I'm default skeptical
Love the verifier economy concept. How feasible would it be for each person to have their own ‘verifier’, such that they could scale their expertise to say validate work?
That would be the goal
Your diagrams and discussions provide high level perspectives of the AI development terrain which shows that innovations that emerge from intrinsic motivation and self-less creativity can be disruptive.....rather than massive scale approaches based on extrinsic rewards and projections of dominance... The AI-hype race to AGI is not focused on solving real-world problems...there are so many real world problems that can be addressed by humans and weak-AI systems collaborating based on real-world constraints. The system level perspective as the logical next step is a little comforting given that holistic models are necessary for solving complex problems like climate change. The insight about interactivity as necessary for dynamic systems and the interdependency of the components starts to align with intelligence in living systems...dynamic, complex-adaptive systems. I hope humans can appreciate natural intelligence of living systems before we unleash destructive AI models that further disconnect humanity from the ecosystem of the living systems in the universe and destroy the environment beyond our capacity to mitigate the damage all in the race for claiming dominance in the race to AGI.
True
Very interesting piece democratizing technical arguments !
Thank you
Loved the analysis. What other Verifiers you see in the market? Can we not just these better verifiers to improve other OSS models?