No. of Recommendations: 10
Jim recommended no posts on AI
A link below to a Morningstar interview with Joel Tillinghast, one of Peter Lynch's the first protege's and now long time manager of Fidelity Low Priced Stock fund. Tillinghast will soon retire. Morningstar recognized him in 2021 as Outstanding Portfolio Manager and in 2002, as Domestic Stock Fund Manager of the Year. According to Fidelity's website the Low Priced Stock Fund has outperformed its benchmark Russell 2000 index over 3,5,10 yrs, and the 34 yr life of the fund.
https://fundresearch.fidelity.com/mutual-funds/per...Tillinghast is asked about how he and Fidelity are using AI in analyzing company financials. He comments on AI's possible strengths and weaknesses:
"Q: You mentioned a few years ago, we talked about artificial intelligence earlier, but you mentioned that you'd be interested in an artificial intelligence product to: 'Tell you more about potential management and business resilience.' As AI and machine learning have evolved, have you found these tools helpful as you expected? What do you view as the potential for further contribution?
Tillinghast: I'm fascinated by artificial intelligence. Any tool's going to depend on the database used to train it, and mostly these are numerical not qualitative databases and so the answers to qualitative questions have tended to disappoint. Language models don't do original research but try to summarize the conclusions of research that somebody has already done. An example that I can give'while trapped at home, I found that I really liked Perry Mason, and it occurred to me that a lot of the clients, but especially a lot of them murderers, were women. And I asked the artificial intelligence ChatGPT, how many of the murderers on Perry Mason were women? And it said there is at least one, but most murderers are men, which is statistically true.
Q: But not on Perry Mason.
Tillinghast: Yes, but they really couldn't answer that because nobody has gone through and tabulated in how many of the 260 episodes the murderer was a woman , and the Perry client was a woman in however many cases. And there isn't a database that has them. Even IMDB doesn't tell you this is the solution. So, for tech things like that, ChatGPT is not very helpful. I enjoy playing with ChatGPT because it's like talking to a glib Harvard graduate who doesn't care that much about the truth but does say some fascinating things. As long as your crap detector is on, it's great. But I worry about people who don't have the crap detector and I worry about using it to set policy where you're talking about people's health or their life savings, rather than how many female murderers there are on Perry Mason.
Q: It's certainly in the wrong hands. I think that's the biggest fear now and maybe as we get into even perhaps generative AI, where it can start to make its own conclusions that it's drawing, it's going to be really difficult, I think, to be able to tell like you said, what's crap from fact and how much of that might build up over time and start to feed the model itself. It can be a little bit scary to think about.
Tillinghast: And sometimes you have to train it to do the steps, where if you're trying to estimate the future earnings of a company, most human analysts will start with the sales forecast and estimate the costs and arrive at an earnings forecast. But AI model doesn't know that you should subtract cost of goods sold and SG&A from sales. They separately derive them unless you instruct them to. So sometimes you have to order the questions in order to get it to answer your question, and sometimes it still can't."
Joel share some other interesting investment insights from his 40 years of investing.
(audio interview and full transcript)
https://the-long-view.simplecast.com/episodes/joel... 