Anne Griffin: disaster alarm, claude ai has much shorter context window, some gpts shuts off if given bad (harmful?) keyword, has connection to EV company owner in Carribean (unfair advantage) Ann Cantera: has connection to voiceflow employee (unfair advantage) ![[Pasted image 20230830192746.png]] Angie Moon: ![[moon23_nss_gps_ai_pd_demo.pdf]] - asked for three feedbacks: - 1. how to construct `choice_tree` well? who should I ask for its verification - given a reasonable choice tree (which can be verified from scholars and mentors like charlie, scott, bill), gpt is great at knowledge extraction (NLU (natural lang. understanding)) meaning it can learn the state transition from case studies - 2. evaluation - just as doctors are encourage to collaborate with ai, encourage human mentors to use this service - we can offer this service to mentors as "way for you to scale" - evaluation measure for `nss_gps` can be how often mentor using this to give advice - 3. can `b2` path finding algorithm (breadth first search) be replace by language-based algorithm? - maybe? try. few months ago, the accuracy for adding two numbers was 70%, but with wolfram-alpha plugin (api, function-call in python code), maybe its doable. similar case is knowledge graph with conversational wrapper. - I got the question from Sheela "What if an organization wants to go with a multi-pronged approach of patenting and bringing an innovative product to market quickly (i.e. protect IP and launch a product together), what will this tool direct the organization to?" - Angie's answer was, this is the gap between theory and practice, that is to say rooms for convergence. In academia, entrep. decision making is characterized by four: freedom (can choose between many paths), uncertain, constraints (limited resource), uncertainty, noisy learning. The first two leads to "test two and choose one" spirit. To use this framework, it might be better to reverse-engineering the level of alternative options so that they are mutually exclusive given the constraint of each lifetime of a startup...