We annotated 119K medical images with two frontier VLMs (Qwen 3.5, Kimi K2.5), cross-validated at 93% agreement, and produced 110K training records, all for under $500. Fine-tuning 3 small models (2-3B params) improved all benchmarks: best model reaches +15.0% average exact match. Everything is open-sourced: datasets, adapters, and code.
The 2024 Nobel Prize in Chemistry went to the creators of AlphaFold, a deep learning system that solved a 50-year grand challenge in biology. The architectures behind it (transformers, diffusion models, GNNs) are the same ones you already use. This post maps the protein AI landscape: key architectures, the open-source ecosystem (which has exploded since 2024), and practical tool selection. Part II (coming soon) covers how I built my own end-to-end pipeline.
After six experiments and dozens of failed attempts, I learned something I did not expect: activation steering, the technique Anthropic uses for AI safety, completely fails for one of the most common tasks in production LLM deployments: generating valid JSON.
And I don't mean "fails to help." My steering-only approach achieved 24.4% valid JSON, compared to 86.8% from the completely untrained base model. Steering made the model worse than doing nothing at all.
Here's what I learned, why it matters, and what actually works when you need guaranteed structured outputs from decoder-only language models.
In other words they managed to fake it until they make! Like most visionaries in silicon valley, lie now, tweet about it, prompt it through fake influencers with their mouth open on YouTube, get that VC money without any due diligence, hire smart people and force them to do it!