Daily Learnings: Wed, May 15, 2024
It is not so important to know everything as to appreciate what we learn. — Hannah More
Random Notes on AI
Given the amount of AI work that SOLVD is doing these days, with helping our clients and building tools internally, I’ve been able to learn quite a bit about various tools and offerings. I haven’t done a good job about keeping track of these learnings, or writing them down at all. So here’s a smattering of random learnings to make sure I’m retaining what I’ve learned.
- Anthropic's AI models, especially Claude 3-Opus, has shown some really great early returns in the applications for which we’re using it
- Unfortunately, we haven’t figured out a way to utilize these models with our own vector database in an embeddings-type setup
- OpenAI’s models and APIs make it a lot easier to do this, therefore we’re building quite a bit with their models at this time
- Data Privacy
- Anthropic’s models accessed via API do NOT train on inputs / outputs
- This means that the queries and responses sent to these models aren’t at risk of being exposed to other users
- OpenAI’s models accessed via API ALSO don’t train on inputs / outputs
- However, the free ChatGPT application does, unless you specifically request to opt out
- ChatGPT Teams plan does not train on inputs / outputs
- Anthropic’s models accessed via API do NOT train on inputs / outputs