2 articles
I built the strongest Forward-Forward instance I could and then used it to show that layer-local training does not scale on real data, that the synthetic benchmarks the field relies on overstate it, and that its memory advantage does not survive a fair baseline. The paper was rejected. The findings still stand.
Recording my experience participating in the University of Toronto Rotman Commerce business case competition, detailing how to use LLaMA-Factory to fine-tune large language models and build AI solutions for law firms.