P2: #dailyreport #conferences #synergy #learning #llm #agents
flexible biological programs may be.
In the self-driving car 🚗 speech, I noted an interesting
fact: agents have “prediction” to imagine the near
future before the reasoning step.
RL learns by trial and error with rewards, Imitational
Learning learns by copying experts, often used to "warm
start" RL, but sometimes combined for better
performance. This is simular to how LLM trained first by
self-supervised learning to predict next token and then
by RLHF.
Interesting DevEx practice 👨💻 shifts focus from
measuring development output to how the developer feels:
- Cognitive load – lower is better
- Flow state – higher is better
- Feedback loops – earlier is better
One speaker sees the AI future in: 🚀
- Edge computing
- AI assistants-coworkers
- Evolutionary, adaptive applications