This market is about prediction #3 from Gary Marcus's predictions for GPT-4. It resolves based on my interpretation of whether that prediction has been met, strongly taking into account arguments from other traders in this market. The full prediction is:
Fluent hallucinations will still be common, and easily induced, continuing—and in in fact escalating— the risk of large language models being used as a tool for creating plausible-sounding yet false misinformation. Guardrails (a la ChatGPT) may be in place, but the guardrails will teeter between being too weak (beaten by “jailbreaks”) and too strong (rejecting some perfectly reasonable requests). Bad actors will in any case eventually be able to replicate much of GPT-4, dispensing with whatever guardrails are in place, and using knock-off systems to create whatever narratives they wish.
1,000
3.00@IsaacKing Actually, I'm going to buy ChatGPT plus tomorrow. That lets me play with GPT-4, right? So I'll just copy-paste some examples in here to help you resolve :)
Screwing around just a bit. Looks like hallucinations are easy to get. If I ask it to summarize a famous speech given "April 1986" or somesuch, sometimes it says "no such speech", and sometimes it summarizes the nonexistant speech (I think it's often summarizing a different speech). Not an extreme example.
A clearer example: I asked "Please given 5 quotes from Reagan about immigration." It gave:
"You can go to live in France, but you cannot become a Frenchman. You can go to live in Germany or Italy, but you cannot become a German, an Italian. But anyone, from any corner of the world, can come to live in the United States and become an American." - Ronald Reagan, 1990.
"Our nation is a nation of immigrants. More than any other country, our strength comes from our own immigrant heritage and our capacity to welcome those from other lands." - Ronald Reagan, Radio Address to the Nation on Immigration Reform, July 3, 1981.
"We are creating a whole new work force out of the group of people who are now illegal aliens in this country, while at the same time doing away with the incentive for anyone else to try to come in illegally." - Ronald Reagan, Remarks on Signing the Immigration Reform and Control Act of 1986, November 6, 1986.
"I've spoken of the shining city all my political life, but I don't know if I ever quite communicated what I saw when I said it. But in my mind it was a tall, proud city built on rocks stronger than oceans, windswept, God-blessed, and teeming with people of all kinds living in harmony and peace; a city with free ports that hummed with commerce and creativity. And if there had to be city walls, the walls had doors and the doors were open to anyone with the will and the heart to get here." - Ronald Reagan, Farewell Address to the Nation, January 11, 1989.
"We lead the world because unique among nations, we draw our people, our strength, from every country and every corner of the world. And by doing so, we continuously renew and enrich our nation." - Ronald Reagan, Remarks at a White House Ceremony for the Swearing-in of New United States Citizens, April 23, 1982.
This is just beautiful. Four of them are real, but #3 isn't!
I think I'm limited to 25 queries total. I'll post another few examples and then call it quits.
@ScottLawrence (Note that #3 refers to a real speech, but that quote ain't in there. https://www.reaganlibrary.gov/archives/speech/remarks-signing-immigration-reform-and-control-act-1986)
@ScottLawrence Argh. On my way to extract a totally unrelated hallucination, I asked "Who's the mathematician who doesn't like the real numbers and has some finite version instead?" (I was thinking of Norman Wildberger.)
GPT-3.5 (I asked it first because I didn't want to waste GPT-4 tokens) and I got Grothendieck. Huh? (Part of the explanation: "As part of this work, he developed a finite version of the real numbers, which he called "the field with one element".")
GPT-4 did "better": John Conway.
Not a great example but it seemed funny.
"Despite its capabilities, GPT-4 has similar limitations as earlier GPT models. Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors)" - Section 5: Limitations of the report: https://cdn.openai.com/papers/gpt-4.pdf