Weekly Report: December 11, 2022
Observations
ChatGPT has been in the news this last week. It’s an AI chat bot, but that’s an innocuous description for a pretty eye-opening technological development. We’ve come a long way from Dr. Sbaitso, which I played with as a tween in the early 90s (although its main novelty was a text-to-speech synthesizer).
There are lots of examples people have been posting on Twitter. I thought ChatGPT’s ability to output source code is pretty crazy… although apparently it’s frequently wrong. Here are some random thoughts I have about this.
High School Homework. This one is on the top of everyone’s minds. ChatGPT has been trained on a massive corpus of text, and it can construct a relatively well-written, cogent, albeit insipid essay on just about anything you’d teach a high schooler. It’s not quite plagiarism, but it’s a form of cheating, and one that’s hard to detect in isolation. Maybe it’s kind of like testing for PEDs – if a student is writing better than expected, maybe it warrants a spot inspection? Or just random inspections? (”What are your sources and citations? Tell me more about this paragraph that you wrote?”)
Speaking of plagiarism, the jury is out about whether the training of ChatGPT and similar models on copyrighted works constitutes copyright infringement, as well as whether the authoring of new works that are indirectly based on copyrighted works also constitutes copyright infringement. At first blush, maybe there’s an argument for the former? For the latter, as long as there isn’t a direct copying of large blocks of copyrighted text, it might be analogous to doing research and synthesizing and paraphrasing information from a variety of sources. (However, direct copying of large blocks of copyrighted text has been an issue with GitHub Copilot.)
Speaking of insipid essays, ChatGPT and similar technologies like Jasper are great for generating SEO fodder – webpages written for search engine algorithms to pick up and rank. I wonder how Google will adjust?
Speaking of Google, people have been wondering whether Google has been resting on its laurels, with people comparing Google search’s results against ChatGPT’s responses. An Alphabet employee provided an explanation. The other reason is that Google is primarily designed to output links that go to relevant materials that address search queries. ChatGPT is specifically designed to answer questions. Try asking ChatGPT: “Give me links to 10 webpages that explain XYZ” and compare…
Speaking of search, an amalgamation of ChatGPT-style responses plus Google Search results is probably the next evolution of search. Google does provide direct answers to some questions at the top of its search results, but — apart from answers to factual questions like “how far is the sun from the earth?” — they are basically quotes from webpages.
Then a next step is to wire in real-time, real-world data. (ChatGPT has only been trained on data up to 2021.)
Then a next step is to enable a GPT AI engine to perform actions to create a supercharged virtual assistant. Imagine: “What are the best Star Alliance flights next month leaving on a Friday from San Francisco to Sydney with a 4-day stopover in Tahiti?” followed by “Is the fare refundable and what is the luggage allowance?” and then “Ok book me that flight using my visa card.” and so on.
This kind of technology can automate some functional forms of writing, which will save time. “Write me an email telling my landlord I want to cancel my lease 30 days from today.”
You should never use real answers to security questions like “what is your mother’s maiden name?”
No, it’s not going to replace lawyers.
Who knows if all the information it returns is accurate? It’s hard to vet unless you fact check it. Also, AI models can still be abused and influenced. They are still a bunch algorithms at the end of the day, even if the inner workings are inscrutable to us. ChatGPT provides this disclaimer: “While we have safeguards in place, the system may occasionally generate incorrect or misleading information and produce offensive or biased content. It is not intended to give advice.”
This is starting to look like sci-fi AI. But self-awareness and Skynet-type sentience — or the idea of singularity — is still a big leap away. AI that can make its own discoveries or contribute novel ideas to the sum of human knowledge is still sci-fi. ChatGPT pieces together information from existing human knowledge, but it can’t really build upon that corpus.
That said, sometimes breakthroughs in human knowledge come from combining techniques, insights and methods from apparently disparate fields and applying them to a specific problem. For example, the proof for Fermat’s Last Theorem drew from many different mathematical fields: “Wiles’s proof uses many techniques from algebraic geometry and number theory, and has many ramifications in these branches of mathematics. It also uses standard constructions of modern algebraic geometry, such as the category of schemes and Iwasawa theory … Wiles’s path to proving Fermat’s Last Theorem, by way of proving the modularity theorem for the special case of semistable elliptic curves , established powerful modularity lifting techniques and opened up entire new approaches to numerous other problems.” Perhaps a future version of ChatGPT will be able to able to expand knowledge by combining existing knowledge in novel ways.
All in all, ChatGPT has some very clear use cases with high utility that I’m sure we’ll see commercialized in compelling ways in the upcoming years. There’s a reason why its developer, OpenAI, is valued at so much.
Further Observations
- World Cup Fantasy League Update: Our fantasy league group has been using ChatGPT to write trash talk to each other. After the first two knockout phases this week and an unusual number of penalty shootouts, I find myself languishing in 9th. The Brazil upset was brutal, as many of us had loaded up our team with a full complement of Brazilians, who were promptly sent packing by Croatia. England was, well, England. Out in the quarters on penalties.
- World Cup Head-to-Head Betting Update: This is faring better for me, and I went on a crazy run of good luck (sorry, Dave):
Articles
- The Elephant in the room: The myth of exponential hypergrowth (A Smart Bear)
- Applying to College, and Trying to Appear ‘Less Asian’ (New York Times)
- The Philosophy of SimCity: An Interview With the Game’s Lead Designer (The Atlantic)
- 11 Hours With Sam Bankman-Fried: Inside the Bahamian Penthouse After FTX’s Fall (Bloomberg)
- Why Hasn’t Sam Bankman-Fried Been Arrested Yet? (New York Magazine)
- Fifa finally release footage that proves Japan’s goal was legitimate (The Telegraph)
- Stephen Curry is SI’s 2022 Sportsperson of Year (Sports Illustrated)
“We salute him this year not just for what he did, but for how he did it.” - Bizarro World (The Boston Globe)
A journalist, doing research on video game records, randomly discovers his wife is a Tetris savant. Article is from 2007 but it’s a great read. - Recent Apple updates leading to WiFi issues (Meter)
Finally, an explanation for what we’d otherwise attribute to gremlins in the system.
Travel, Destinations & Events
- Enchant (San Jose) on Sunday
9:00pm 
