←
Rabbi Tzvi Pittinsky12,641 posts
Rabbi Tzvi Pittinsky avatar

Rabbi Tzvi Pittinsky

@TechRav

Father, Husband, Director of Educational Technology at Yeshivat Frisch

◷ Joined March 2009▧ 2,495 Photos and videos
Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·9h

Many kids were using high-priced tutors to write their college essays or even their research papers. Just read the book Schooled from the early 2000s. AI has forced teachers the grapple with the fact that anything out of class can be written by anyone whether AI or someone else.

Post attachment

Rich parents have paid up to $25,000 to get their kid's college essay polished for years, and colleges were fine with it. Then poor kids started getting the same kind of help from ChatGPT for free. Now colleges are getting rid of the essay. And the essay was tilted toward rich kids long before AI. Back in 2021, Stanford researchers ran 240,000 University of California admissions essays from 60,000 applicants through text analysis. The word choices, the topics, even the punctuation predicted family income better than SAT scores did. Give them an essay and they could guess the kid's SAT score within about 120 points. Which is wild when you remember colleges spent years ditching the SAT for favoring rich kids, when the essay was favoring them even more. There's a whole industry built around that one page. AcceptU charges $12,500 for essays and a college list. Admitium charges $15,000 to $25,000. Former admissions officers sit down with a 17-year-old and work his 650 words over until they sound like what the readers want to hear. That's been going on for decades. Then ChatGPT showed up. Researchers at Cornell and Carnegie Mellon looked at 81,663 applications to one selective university from 2020 to 2024. AI use shot up in 2024 for everybody, and it shot up the most among kids who got application fee waivers, meaning the low-income kids who could never have paid for a consultant. Those kids paid for it anyway. The more AI showed up in an essay, the more it hurt the applicant's chances. It hurt low-income kids more than wealthy ones, even after the researchers controlled for grades and test scores. The free version makes a lot of kids sound like the same kid, and admissions readers are learning what that sounds like. Look at what colleges are doing now. Tulane paused its "Why Tulane?" essay and said AI was definitely a factor. Cornell, UVA and Georgia cut back their essay requirements. Duke quit giving essays a numerical score. Columbia became the last Ivy to bring back required test scores in June. And the UC system, the same one whose essays were in that Stanford study, is thinking about bringing back the SAT too. Virginia Tech went another way. It got 57,622 applications for about 7,000 freshman spots, so now an AI scores every essay alongside one human, and a second person only reads it if the two scores are two or more points apart. Funny how the essay only became a problem once everybody could afford the help.

Jim Geraghty avatar
⟳ Rabbi Tzvi Pittinsky retweeted
Jim Geraghty@jimgeraghty·Nov 16

I just can't shake the feeling that Mayor Mamdani is going to invade Kuwait.

Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·Nov 15

Do people realize that our AI systems were trained on sci-fi tropes like this? This is not AI becoming self-aware! This is AI mindlessly acting out the scripts we gave them!

Arnold Schwarzenegger told us what would be happening James Cameron nailed it. Listen to the scene.🤣 🔊

1:02
Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·Nov 13

What does the story of Jacob deceiving his father to get the blessing instead of his brother Esau have to do with the Anthropic IPO prospectus? Find out in this week’s Jewish Link! jewishlink.news/the-voice-of…

Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·Nov 9

Fascinating. So apparently, Gemini and Claude know about 20th century Responsa literature and modern Poskim while ChatGPT stops at the Shulchan Aruch and Mishna Berura.

Who do AI models cite when asked halacha? Share of 288 answers naming Rav Moshe Feinstein: Gemini 33%, Claude 30%, Grok 14%. Every GPT model: 7% or less. Same gap for R. Shlomo Zalman and Shemiras Shabbos. GPT's poskim stop at Shulchan Aruch, Rema, Mishnah Berurah. Thread.

Post attachment
Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·Nov 8

One has to wonder how this will play out in real life. Will our AI systems, misaligned through no fault of their own, achieve alignment? Will we always keep the human in the loop? Or could AI more powerful than anything Hollywood imagined in the 80s lead to worldwide catastrophe?

Rabbi Tzvi Pittinsky avatar
Rabbi Tzvi Pittinsky@TechRav·Nov 8

But an AI learning by playing itself over and over is exactly how reinforcement learning works. The AI concludes that "the only winning move is not to play," a misaligned AI learning to align itself though its worth noting that it took a human to point it in the right direction.