I don't think transformers are the answer. I've thought about this for a long time, longer than most. My interest in AI goes back to ELIZA, and in the early 2000s I was already modifying open-source chatbots in Visual Basic, experimenting with giving them something closer to human cognitive memory. The question of how machines actually remember and reason has been with me ever since.
The more I understand about how biological intelligence works, the more convinced I am that the field is optimizing the wrong thing. Attention mechanisms and scale are impressive, but they're not how brains learn, and I think that gap matters more than the benchmarks suggest.
Most of my serious research has been about closing that gap. I study biological neural computation, how organic systems wire themselves, adapt, and generalize, and try to translate those principles into architectures that don't exist yet. I built an analog neural network from discrete electronics, no digital gates, to get closer to the hardware reality of synaptic computation. That wasn't a side project. It was part of a longer investigation.
I also work on the practical side: training and fine-tuning LLMs, building inference infrastructure, shipping products. I published on ASRL and Progressive LoRA Merging, and independently developed reflection-based chain-of-thought reasoning before it became common. Those are stepping stones, not the destination.
I've been building on the web for over twenty years and making games on the side. Most interested in the problems where the right answer hasn't been written down yet.
