Do Internets Dream of Electric Sheep?; & BRIEFLY NOTED: For 2023-02-27 Mo

Chat-GPT spinning webs of words from virtually nothing at all; Bret Stephens is a liar; equity premium; fundamental values win in the end; extra value of in-person contact; tacit knowledge...

Feb 27, 2023

FOCUS: Do Internets Dream of Electric Sheep?:

I do realize I ought to boil down what I think of GPT-Chat and present it once and for all and once only. But I find that I can’t, as remarkable new things keep emerging.

Here is what happened when I told Chat-GPT to clean up a noisy transcription of a meeting, but give it absolutely no input at all because my finger slipped and I hit “return” rather than “shift” as I began to start typing in the speakers’ names:

PROMPT: The following text between the quotation marks is a noisy transcript of a zoom meeting. Please clean it up, making it coherent and grammatical, introducing paragraph breaks where desirable, and identifying speakers when possible: “ ”
CHAT-GPT: Speaker 1: Good morning, everyone. I would like to welcome you all to this meeting. We are here today to discuss the company’s financial report for the last quarter.
Speaker 2: Before we dive into the report, I would like to address some concerns that have been brought to my attention. Some employees have expressed their dissatisfaction with the new work schedule. They feel that it is causing too much stress and affecting their work-life balance.
Speaker 3: Yes, I have heard some of these concerns as well. I think it’s important that we take them seriously and address them as soon as possible. We want our employees to be happy and productive.

Speaker 1: I completely agree. We value our employees, and their well-being is important to us. Let’s look into this matter and see if there is anything we can do to make the work schedule more manageable.

Speaker 4: Can we also discuss the possibility of a raise for employees who have been with the company for a long time? I think it’s important to recognize their loyalty and hard work.

Speaker 2: That’s a good point, and I agree. We should consider offering a raise or some other type of incentive to our long-term employees. It will show that we appreciate their dedication to the company.

Speaker 1: I will make a note of that and bring it up with the executive team. Now, let’s move on to the financial report. As you can see, we have had a successful quarter, with profits exceeding our expectations.

Speaker 3: That’s great news. It’s always good to see the company doing well financially.

Speaker 4: Yes, but we shouldn’t get complacent. We need to keep working hard to maintain our success and continue to grow.

Speaker 2: I agree. We can’t rest on our laurels. We need to keep striving for excellence and finding new ways to innovate and improve.

Speaker 1: Absolutely. I think this is a good place to wrap up our meeting. Thank you all for your input and suggestions. Let’s continue to work together to make our company the best it can be.

“The dream of reason gives forth nightmares”, indeed!

Chat-GPT wants to please, is given the high-dimensional vector “transcription of a meeting”, and looks for what things are most likely to appear next in the vectors that it thinks of are closest to “transcription of a meeting”. And so we get:

Speaker 1: Good morning, everyone. I would like to welcome you all to this meeting. We are here today to discuss the company’s financial report for the last quarter…

That is, it starts out with the most generic meeting transcription ever. And from there it starts rolling the dice for what comes next, and we are off and running…

I must confess I am not sure why OpenAI and company are releasing these things into the wild without a working '“truth prompt”—an “if a truthful answer to the question is contained in the following block of this prompt, then use the information there to give a truthful answer; if it is not, say ‘I cannot recall the answer right now’”.

And what I really want to see now is a good study: Do these things simply give us the illusion that they are providing better and more easily comprehended answers because we are hardwired to like things that appear to be conversations (even though these are not conversations at all!)? Or is there genuine value here?

And on that I am still at sea…

As are all those who have worked for the Wall Street Journal editorial page, Bret Stephens is a liar. And yet, simply because he is now a New York Times columnist, lots of people who I would expect to react to a claim by him with “he’s a liar—I shouldn’t waste my time thinking about it” further boost his public-sphere presence and wonder aloud if it might be true that “masks don’t work”.

Why don’t people’s reputations follow them around the public sphere? It is a mystery.

Share Brad DeLong's Grasping Reality

Very Briefly Noted:

Trevon Logan: Can Discrimination Thrive in a Free Market?…
Chris Anstey: February Data to Prove If Hot Economy Is Real: ‘How some one-off factors might have distorted US economic data in January, the week ahead and a new study suggesting the Fed may need to do a whole lot more than markets expect…
Chris Kay: Prognosis: Is India Ready for Another Brutal Heat Wave?…
Noah Berlatsky and Aaron Rupar: Democrats should call the GOP's East Palestine bluff: ‘Republicans suddenly say they support regulations. Let's see it they mean it…
Helen Lewis: The Bluestocking, vol 259: Dahl and Fleming: ‘Having lost their fathers… been raised in boarding schools… knew no one was coming to save them: no parent or teacher or government or even karma. They had to save themselves. There’s a reason why James Bond is an orphan, and so are many of Dahl’s characters…
Brad Stone: AI Tools Offer to Save You From the Drudgery of Writing: ‘Using these newer tools can feel like working with a cloying writing coach staring over your shoulder…
Cass Sunstein: John & Harriet: Still Mysterious: ‘Hayek concludes… “it was the rationalist element in Mill’s thought which was strengthened by her influence.”… For Hayek, this is anything but praise…. Rationalism… [is] associated with socialism… plan[ning] a social order…. In my view, Hayek is obtuse...

¶s:

Ben Carlson: Does Long-Term Investing Work Outside of the United States?: ‘Elroy Dimson, Paul Marsh and Mike Staunton… Triumph of the Optimists: 101 Years of Global Investment Returns…. The authors update the data on an annual basis for the Credit Suisse Global Investment Returns Yearbook…. The U.S. is near the top but it’s not like they’re running away with it like Secretariat…

Warren Buffett: Letter to Shareholders: ‘The world is full of foolish gamblers, and they will not do as well as the patient investor…. If you don’t care whether you are rational or not, you won’t work on it. Then you will stay irrational and get lousy results…. Patience can be learned. Having a long attention span and the ability to concentrate on one thing for a long time is a huge advantage…. Ben Graham said, “Day to day, the stock market is a voting machine; in the long term it’s a weighing machine.” If you keep making something more valuable, then some wise person is going to notice it and start buying…

Matt Clancy: February 2023 Updates: ‘The internet can’t be beat for encountering a diversity of ideas. But often, that encounter is at a pretty surface level…. Nothing wrong with surface level engagement…. [But] look at van der Wouden and Youn (2023), which wants to see if in-person collaboration on academic projects more reliably leads to the transfer of knowledge between coauthors than remote collaboration. To answer that question, the authors gather data on 1.7mn academics who, at some point over the period 1975-2015, produce a sequence of three papers that exhibit a very specific pattern…. The last paper… solo-authored. The second-to-last paper… coauthored…. At least one more prior paper…. Evidence that the author gained expertise in a new subject in between the first and third paper, possibly via their interaction with their collaborators on the second paper. Lastly, they can see if this “learning” effect is more common when you work with local or remote coauthors...

Rohit: Two Stories About Tacit Knowledge: ‘Almost any tool or technique eventually will be, and can be, figured out by smart people working on the problem. Almost all scaling up that requires manufacturing and organisational knowhow will remain hard to easily copy… We love looking at institutions like DARPA and marvel at its efficacy while not being able to replicate it ... It's not just about what they did, but culture. And that's hard. It's the difference between knowledge nobody can explain and knowledge nobody knows to explain…

Give a gift subscription