Admittedly, the output is far from perfect. Some of his ChatGPT answers contain bias, circular logic, and inaccuracies, often disguised in overly confident prose.
However, the range and speed of topics ChatGPT can spit out first drafts is jarring.
legal document? check. financial analysis? check. Cold sales? check. corporate strategy? check. coding? check. comedy? No (as someone who writes silly jokes on Twitter all day long, my inability to decipher ChatGPT’s current humor gives me a sliver of hope).
Ethan Mollick, Professor of Innovation at the Wharton School of the University of Pennsylvania, has shown that he can apply ChatGPT to his work to create authoritative course syllabi and lecture notes.
“I think people underestimate what they can get from ChatGPT,” says Mollick. “If you’re a white-collar worker, this is a productivity game changer.”
That is the current OpenAI LLM. The organization plans to release a much stronger LLM in 2023, and Google has been working on it for years (full disclosure: I co-developed a research app built on top of his LLM. created).
Mollick says the key to understanding ChatGPT’s potential is recognizing its true strengths. Current chat AI may be inadequate for fact-and-prediction tasks, but it is a powerful tool for corrections and ideas.
Of course, mileage varies from role to role and depends on how many errors you can tolerate on the job. Many ideas need to be generated and mistakes can be fixed quickly without harm. Conversely, when it comes to managing a nuclear power plant, you want more factual certainty and fewer corrections.
In a recent article, Mollick shows four ways to interact with ChatGPT to show promise as creative assistance (designing games and saying, “I tend to lie, but I want to make you happy.”) (including joking as “a magical intern with great aspirations”). ).
Across the white-collar industry, Morrick believes that “using AI is better than AI alone.” The question is, in what percentage of each industry the combination of AI and humans is better than AI alone. 10%? 20%? 30%?
Former Bloomberg Opinion columnist Noah Smith and the famous pseudonymous AI researcher Loon have also laid out a future path for human-AI collaboration called the “sandwich model.”
• Human gives prompt (pan) to AI
• AI generates a menu of options (hearty fillings)
• Human selects options, edits and adds as desired (panning)
Smith and Roon point out that workflows are for all types of generative AI (text, visual, etc.) and give some very relevant examples.
Attorneys probably write legal summaries this way, and administrative assistants use this technique to draft notes and emails. Marketers come up with campaign ideas, generate copy in bulk, and finalize it. Based on your short vision, the consultant will create an entire PowerPoint presentation with a coherent narrative and provide the details. The financial analyst asks for the type of financial model and provides an Excel template with auto-populated data sources.
In fact, Roon says everyone should be “on top” of AI development in their respective fields. Some examples: Harvey for law, his Github Co-Pilot for coding.
“People who know how to use AI tools get a raise,” says roon, who is also a great source for funny AI-related tweets.
Another feather in the “ChatGPT won’t replace you yet” cap is humans’ persistent desire to let other humans into the loop. Roderick Kramer, a social psychologist at Stanford University, said: It has become an advantage in our struggle for survival. From the moment they first meet their mother’s eyes and begin mimicking her facial expressions, they crave and nurture the sense of security that comes with human contact. Mollick points to two of his studies that show pushback against recommendations for AI in HR and healthcare settings.
However, attitudes adapt. Based on the embarrassing photos I’ve been floating around online, I suspect that the general willingness to put personal information online is higher today than it was 20 years ago. And the idea of hailing a stranger’s car or sleeping in a stranger’s spare bedroom didn’t sound like a $50 billion concept 20 years ago.
So, do you think ChatGPT can do the job for me? Its inventiveness and first draft are terrifyingly good. As a reminder, I have hours of interpretative stand-up comedy workshops.
Bloomberg Opinion Details:
• Is ChatGPT the beginning of the AI revolution: an editorial
• Google faces serious threat from ChatGPT: Parmy Olson
• ChatGPT could further disrupt democracy: Tyler Cowen
This column does not necessarily reflect the opinions of the editorial board or Bloomberg LP and its owners.
Trung Phan is co-host of the Not Investment Advice podcast and author of the SatPost newsletter. He previously served as his lead writer for Hustle, a technology newsletter.
More articles like this can be found at bloomberg.com/opinion.