1) Agentic behaviour and importance of access t RAG and, search tool use.

2) system prompt

3) llm-as-a-judge: customizable judgment and writing policy, tools such as any-llm, oss, etc. to say how tow rite policy and evaluation of guardrails too.

4) disclaimers in the humanitarian context. Adding disclaimers detection and making sure classifiers of topics are accurate in both languages and have that. Or deployers can do that. 

5) resources for language: going beyond the training data narrative: tokenization, guardrails , tokenizes, evaluation benchmarks, etc. It’s easy for companies to say the issue is because of lack of data therefore

6) companies law enforcement and border control compliance by handing out data and education for deleting information. Also need for these AI labs for having transparency reporting about handing out data to alw enforcement.