I think that’s because it’s a Chinese Project. Same thing with Ant Design Components, which are really awesome but not as well known as they should be.
I think it being Chinese is part of the reason as some of the examples in the early days were Chinese only, which could deter some people. It is certainly more complex (for a good reason in my opinion), I can see why it is not well known since I think the vast majority just wants to create simple charts. However, with echarts, it really can meet Enterprise needs.
Yeah that might be why. A couple years ago I was trying to find "this cool charting library" I came across and I could not get it to surface in Google.
It seems good, but their docs websites are absolute trash (though they've seemingly gotten somewhat better recently - they were previously completely unusable).
You’re right that it’s analogous in concept, but strategy distillation happens at a higher level: it encodes and transfers successful latent reasoning patterns as reusable “strategies,” without necessarily requiring direct gradient updates to the original model weights.
I can see where you’re coming from, but not really. Unlike an RNN, the main transformer still processes sequences non-recurrently. The “sidecar” model just encodes internal activations into compressed latent states, allowing introspection and rollback without changing the underlying transformer architecture.
I can assure you it’s not a joke. Compute power is increasing at a ridiculous pace, and highly capable models are getting smaller and smaller, now at the 30b parameter size and under. So even if it wouldn’t be pragmatic now, it could become highly relevant in 4 or 5 years if trend lines continue at anything like the recent pace.
Interesting, it certainly wouldn’t take up much additional space, but I wonder if it would have any real impact, since it seems somewhat orthogonal to finding a faithful low-dimensional encoding of the activations.
I recently started thinking about what a shame it is that LLMs have no way of directly accessing their own internal states, and how potentially useful that would be if they could. One thing led to the next, and I ended up developing those ideas a lot further.
Transformers today discard internal states after each token, losing valuable information. There's no rollback, introspection, or replaying of their reasoning. Saving every activation isn't practical; it would require way too much space (hundreds of megabytes at least).
The insight here is that transformer activations aren't randomly scattered in high-dimensional space. Instead, they form structured, lower-dimensional manifolds shaped by architecture, language structure, and learned tasks. It's all sitting on a paper-thin membrane in N-space!
This suggested a neat analogy: just like video games save compact states (player location, inventory, progress flags) instead of full frames, transformers could efficiently save "thought states," reconstructable at any time. Reload your saved game, for LLMs!
Here's the approach: attach a small sidecar model alongside a transformer to compress its internal states into compact latent codes. These codes can later be decoded to reconstruct the hidden states and attention caches. The trick is to compress stuff a LOT, but not be TOO lossy.
What new capabilities would this enable? Transformers could rewind their thoughts, debug errors at the latent level, or explore alternative decision paths. RL agents could optimize entire thought trajectories instead of just outputs. A joystick for the brain if you will.
This leads naturally to the concept of a rewindable reasoning graph, where each compressed state is a node. Models could precisely backtrack, branch into alternate reasoning paths, and debug the causes of errors internally. Like a thoughtful person can (hopefully!).
Longer-term, it suggests something bigger: a metacognitive operating system for transformers, enabling AI to practice difficult reasoning tasks repeatedly, refine cognitive strategies, and transfer learned skills across domains. Learning from learning, if you will.
Ultimately, the core shift is moving transformers from stateless text generators into cognitive systems capable of reflective self-improvement. It's a fundamentally new way for AI to become better at thinking.
For fun, I wrote it up and formatted it as a fancy academic-looking paper, which you can read here:
This is great, I was debating whether I should do my latest project using the new OpenAI Responses API (optimized for agent workflows) or using MCP, but now it seems even more obvious that MCP is the way to go.
I was able to make a pretty complex MCP server in 2 days for LLM task delegation:
I wrote this article detailing my recent experience helping my dad navigate some complex health issues with AI models. In the process, I came up with a good set of prompts and procedures for getting really high-quality diagnostic feedback and analysis from the best frontier models. Hopefully others will find it useful.
