Speakers
General questions on Data & AI
- What will you do with my data?
- How do you secure my data?
- You must deal with skepticism (data you plan to use, the freshness of the corpus, how frequently will you use it?)
- How will you deal with bias?
- In Europe, it's very hard to get approvals and you must disclose a lot of information.
- Compliance, cloud and due diligence on the use case & vendors.
- Think about timelines in terms of testing pilots, releasing products that will have a long term impact.
- The capability of building software, disrupting your technology and reducing cost.
- Customers are asking: what will you do with GenAI?
- Talk about efficiency
- Talk about usage of GenAI
- Reassure them about your business value
Leveraging existing models and optimizing cost of fine-tuning on customer data
- Design the product in a way where you're not dependent on a model to derive business value.
- Open sourced models usually suffice in the beginning. Accumulate data and fine tune for specific task versus generic training.
- It is a challenge to start with no data but large models can do 60-70% of the task. Ask your customer for permission to use for training your models.