Allyson Ettinger on GPT-3

Описание: How can a teacher know if a student actually wrote their book report, or if a computer did it? Are AI writers coming for journalists’ jobs? What does it mean when a language processing model can write its own computer code upon request? These are all questions currently sparked by GPT-3, a free online natural language processing artificial intelligence by Open AI.

This isn’t your dimestore chatbot. GPT-3 takes advantage of a whole new method of artificial intelligence research, called neural nets, to create plays, write code, and even roleplay as a historical figure. But what are the limitations to this kind of AI? In this episode of Carry the Two, University of Chicago professor Allyson Ettinger walks us through how GPT-3 manages to sound so human and where and how it fails in interesting ways.


Find our transcript here: https://docs.google.com/document/d/1v...
Curious to learn more? Check out these additional links:

Use natural language processing to talk with a TV character or historical figure: https://beta.character.ai/

Chat bot using GPT-3.5: https://chat.openai.com/chat

Find out how you can chat with GPT-3: https://lifearchitect.ai/how-do-i-tal...

When GPT-3 accidentally lies: https://www.technologyreview.com/2022...

Microsoft’s chatbot that went racist: https://www.theverge.com/2016/3/24/11...

Is GPT-3 a replacement or tool for journalists: https://contently.net/2022/12/15/tren...

Undark’s interview with GPT-3 on truth & journalism: https://undark.org/2023/01/07/intervi...

Previous Carry the Two episode on statistical language learning with Ben Reuveni: https://podcasts.apple.com/us/podcast...


Follow more of IMSI’s work: www.IMSI.institute, (twitter) @IMSI_institute, (mastodon) https://sciencemastodon.com/@IMSI, (instagram) IMSI.institute
Follow Allyson Ettinger: https://linguistics.uchicago.edu/peop..., @AllysonEttinger
This episode was audio engineered by Tyler Damme.
Music by Blue Dot Sessions.
Sound effects from pixabay.
The Institute for Mathematical and Statistical Innovation (IMSI) is funded by NSF grant DMS-1929348.