Machine Learning improving education
Historically with over 2000 courses to study from, our students were lost. We understand the dangers of feeling lost on one’s path, and the wasted time one can save by having someone else guiding them. Unfortunately it is impossible to give customized hand selected recommendations for each one of the students.
I custom designed a 2 step word2vec + seq2seq neural network recommender which greatly increased course conclusion rates and student retention. The research and implementation was made in multiple steps as to ensure the statistical significance of such findings.
2. OpenAI Panel with NYU Faculty Members
As a final project of the NYU Creative Programming course I implemented an AI generated interview with all IDM faculty members listed on the website. It might be incompatible with a few browsers due to the usage (and hacking) of html’s text to speech API.
All questions were generated by an artificial intelligence (OpenAI) which was fed with the faculty members background present in their NYU webpages. The input data was a copy and paste so I could analyze the capabilities of OpenAI under a limited data situation.
Then, all answers from the faculty members were also generated by OpenAI as if OpenAI were the member themselves.
Finally, the voice was generated with the text to speech api from the browser itself.
You can run the OpenProcessing code here. Or view the demonstration below.
I also explored DeepFake approaches to generate the answer video but most deep fake algorithms require input videos from the target person. The one I had access to that requires only a single photo from the faculty member did not generate good enough results.
A demonstration on how the panel works. Further forks could use translation APIs to automatically translate and text to speech the panel. Or it could deep fake a video of the panelists.