@MachineScholar I want to really thank you for your cost analysis and helping me to understand this. I’m a bit overwhelmed with all the new information myself, but the young computer scientist interns seem to suck up the information like a sponge. They may be thinking 8x faster than me…
I have one intern working on the AI plugin for two different Discourse communities. We’re paying the interns, but they are cheap and they’re certainly enthusiastic. The intern mainly doing the AI work is at a University of California computer science program and I often wonder what the on-campus discussions are like in such a young group where the future is so clearly their future to create.
I also wonder what your own research environment is like? You seem to be deeply involved in the technology. What a great time to be involved. So exciting.
I’ll likely start a new topic on my next question. The intern is implementing Google Custom Search and GitHub Token access for the AI bot. I’m not quite sure what these are. However, I’m hoping that the AI bot can access GitHub repos to look through documentation… I’m not sure what’s possible. I also don’t know if Retrieval-Augmented Generation (RAG) is used in the Discourse AI plugin.
Regarding the efficacy of DeepSeek R1 versus o1, a different intern was talking to me about using it for their CS projects using the Web app UI (using ChatGPT Plus). So, the test was super informal, but the enthusiasm for DeepSeek by one of the interns was big.
The intern that is actually working on the AI implementation has been much more reserved about the differences between the LLMs. They are primarily providing cost and usage tables with limited comments thus far on usage differences. We will be making all the LLMs available to the community and ask them to assess. So, it’s smart of the intern to keep their opinion low at the moment.
Thank you again for your help on my journey.