|
What is special about this assistant is that it can leverage interfaces (web search, Python, etc) directly within its messages. For example, as it realizes that it needs to calculate 1 + 1, it might type "[PYTHON] print(1 + 1) [/PYTHON]", and the output "[OUTPUT] 2 [/OUTPUT]" gets inserted right after, before the assistant continues streaming its answer. The seamless insertion of code and output within the conversation - as opposed to the more standard approach (ChatGPT Plus, AutoGPT) where the assistant has to classify your prompt, select appropriate actions from a menu, and then use them to prompt other external agents - turns out to produce a really useful assistant that is both surprisingly autonomous and aware of its own limitations. Use it as you wish (and at your own risk of course). Here are a few examples of things I have successfully used Saola for in the past few days: - I asked it why some app kept telling me to "close Chrome" after I had already closed it. Saola quickly realized that the app was seeing processes from another session. - I asked it to help me set up a new Python package I was building. It moved files around, (after confirming that I agreed), it asked me some necessary questions and read Python files to derive dependencies for the pyproject file. - I asked it to find and successively open a bunch of spreadsheets for me from different places, so I could do some edits and print each of them to a PDF. At every step, it would check that I saved the PDF to the right place. - It helped me brainstorm a solution for a math problem, verified particular cases as asked, and provided ideas from the patterns it observed. |