Day 95 MIT Sloan Fellows Class 2023, CSAIL Thesis defence - context for AI
Everyone becomes mad about ChatGPT.
There are plenty of articles explaining how innovative it is and why it is disruptive.
I just summarized my thoughts about ChatGPT technology in NLP context and picked up some insightful articles worth reading.
Adaptive NLP model and ChatGPT
Before diving into ChatGPT innovation, we need to mention about one paradigm shift in NLP modeling. As you can see my diagram, the traditional NLP model was task-based. For example, if you want to resolve name entity recognition problem(NER), you build specific model for this task.
However, the attention model and transformer changed the paradigm and make NLP model more versatile.
Then, ChatGPT unleashes this remarkable progress with user-friendly interface. You can check how powerful the user interface is. It achieved 1M users in five days.
Future differentiator
So what is the future differentiator?
- Exclusive dataset ( private chat, not open data)
- More data formats: image, video, sound
- More integration in deployment field
- Democratising NLP model with low infrastructure requirements
Changed human behaviour
Deep learning has one significant disadvantage. It can not be interprettable for humans. Then, no one can stop crazy AI like Skynet. Also, many AI chatbots failed to realize unbiased conversation.
Trolls turned Tay, Microsoft’s fun millennial AI bot, into a genocidal maniac - The Washington Post
So many data science people have tried to develop a transparent model. However, what chatGPT realized is to make it more black-box but make the result more reasonable for humans.
Now, very few people consider why and how the result comes out. Rather than that, everyone tries to get more reasonable answers with optimized prompts.
Useful links
For begginers
Prompts are fundamental techniques you need to learn in order to manipulate ChatGPT more productively and innovatively.
As the first step of integration between ChatGPT and other deployment tool, you should try GPT for Sheets and Docs. It makes a lot of mistakes, but you should focus on how to correct mistakes rather than researching countless data by accessing bunch of websites.
GPT for Sheets™ and Docs™ - Google Workspace Marketplace
Also, Chrome extension is another useful integration with ChatGPT.
ChatGPT for Google - Chrome ウェブストア
The impact of ChatGPT, there are several interesting articles, but in my opinion, you need to follow "industrial impacts" in education and search ads.
Education
ChatGPT, Chatbots and Artificial Intelligence in Education - Ditch That Textbook
ChatGPT banned in NYC schools over learning impact concerns
Don’t Ban ChatGPT in Schools. Teach With It. - The New York Times
Search
Google plans to demo AI chatbot search as it panics about ChatGPT - The Verge
ChatGPT AI Threat Pulls Google Co-Founders Back Into Action, Report Says - CNET
Google has everything it needs to counter ChatGPT