🤔 This is interesting: Andrew Ng, Stanford professor and computer scientist who cofounded Google Brain, reported that GPT 3.5 achieves superior results than GPT 4 when wrapped in an agentic flow. 💡
This video is from one of Ng's presentations for Sequoia Capital earlier this year.
🦾🤖 Such agentic flow would not just give the LLM tools to search the web and execute code, but would have steps to examine its own work to improve it.
For example, an agentic flow would have steps such as:
- Plan the steps to fulfill the task
- Determine whether to search the web to get more information
- Execute the plan
- Get feedback on the results to identify problems
- Revise the plan to address the issues
- Iterate until completed
🔗 LINK:
https://lnkd.in/gjZs4jJs