Deep learning networks mimic what happens in biological neurons with tweaking of weights, but living systems have a bottom-up stream of information about the environment through emotion, and original Pavlovian conditioning (which modern reinforcement learning is based on) involves unconditioned stimuli like pleasure and pain—yet we don't understand how to incorporate the biology of emotion into AI systems.

Fine-tuning models through tools like LoRA (Low-Rank Adaptation) and access to open-source models allows individuals to customize large language models on consumer-grade hardware, creating a tension between democratic access and safety risks; companies and people worry this could enable malicious actors to fine-tune models for dangerous purposes.