Andrey Fradkin
@AndreyFradkin
Followers
2K
Following
13K
Media
65
Statuses
1K
Professor @BUQuestrom and @MIT_IDE, climber, immigrant, nerd dinner convener. Co-host of Justified Posteriors Podcast. On leave, AI + markets @ Amazon.
San Francisco / Boston
Joined December 2010
It's great to be able to work with private data sources to understand AI's impact on jobs, but they are not a substitute for high-quality government data. That's why I signed this letter to the US Secretary of Labor.
Today, over forty leading economists and experts are calling on the Department of Labor to gather more data on AI's jobs impact. Signers include former fed chairs, Nobel laureates, and experts on the right and left. Read the full letter: https://t.co/1wesfldRx8
5
13
74
I expect there to be dozens, if not hundreds, of papers on this topic over the next few years. By looking at these questions using multiple approaches and datasets, I hope we get some clarity on this important question.
1
0
1
Magnitudes matter, especially for policy. With a big enough dataset, we can detect all sorts of deviations from trends. It is really important to learn about typical fluctuations in occupation by age groups are over time, due to macro or other factors.
1
0
1
"Canaries in the coal mine" is a valuable contribution, with an awesome new dataset for the study of AI + labor. The evidence is consistent with AI exposure -> reduced youth employment, although it is also consistent with other mechanisms and the timing doesn't quite line up.
1
0
2
There are many research teams looking for the effects of AI on the labor market, and media coverage for finding an effect will be much bigger than for null results. So we should be particularly cautious.
1
1
4
The audience demanded that we cover a recent, sensational paper claiming that AI is already causing job loss for young cohorts. @SBenzell and I dig deep in our latest Justified Posteriors episode. Some thoughts:
1
5
14
We recently did a Justified Posteriors episode on @ide_enrique's interesting paper in case you prefer podcasts to papers (link below). Also, look for an episode about the Canaries paper very soon.
Everyone’s talking about @erikbryn, @econ_b & @RuyuChen’s excellent new paper Canaries in the Coal Mine. It shows entry-level jobs shrinking in AI-exposed occupations. But wait: earlier studies found AI helps novices most. So why are entry-level jobs disappearing? (1/n)
2
2
13
Hard to design a good personalized recommendation system without training on usage data.
NEW: Anthropic will start training its AI models on user data, including new chat transcripts & coding sessions, unless users choose to opt out by 9/28 (it's a pop-up window that will give you the choice). It’s also extending its data retention to 5 years. https://t.co/DCYvFh0Iqu
1
0
3
If you're a coder, you should consider going to this event in SF on Sep 6. I'm optimistic that it will produce useful research, and it needs more participation!
We’re hosting a historic hackathon with @METR_Evals, inspired by their latest paper that measured the real-world impact of AI coding tools. Here's how it works: 🤖 Half of participants will build with AI tools 👩💻 Half of participants will build without AI tools Judging is blind
0
1
4
I'm a bit late to this, but Hal is very generous and gave Scott Baker and I our first seminar invite as grad students, as well as a cool Google Economics shirt.
Hal Varian's retirement from Google is a major event. He was the architect of many aspects of Google's modern business operations and was copied by others I highly recommend this @WIRED article from 2009, which called him the "high priest of Googlenomics" https://t.co/7zg0yPTQvP
0
4
21
Excited to finally have a working paper of this. It’s a cool experiment and model, please check it out.
Using a field experiment, we studied how consumers valued Amazon brands and the extent of self-preferencing by Amazon. We found that Amazon brands bring benefits to consumers, even though they receive a modest prioritization in search rankings.
0
3
27
Did a fun interview with NPR about the business of jobs platforms, a cooling labor market, and the need to give job seekers tools to signal match quality. Link below.
2
6
26
This represents the evolution of consumer LLMs to be more like recommender systems or news feeds. The platform is optimizing the model and prompt to achieve user satisfaction with a cost penalty. There is likely to be A/B testing driving many of these routing decisions.
You are likely going to see a lot of very varied results posted online from GPT-5 because it is actually multiple models, some of which are very good and some of which are meh. Since the underlying model selection isn’t transparent, expect confusion.
0
2
10
Surprised that more people haven't seen this tweet. Jonathan Kanter and friends made the correct call, leading to today's outcome.
Congrats to @Zoink & Figma team on the IPO. 🚀 We need more public companies. It will create more value for everyone — more jobs, innovation, returns, etc.
0
1
3
This of course raises the question of what exposure indices are useful for. I have my thoughts, but what do you all think? @alexolegimas @SBenzell @Afinetheorem @alz_zyd_ @ben_golub
1
0
3
In the latest Justified Posteriors episode, we interview @danielrock, an author of perhaps the most prominent of these indices. Daniel believes that exposure scores are orthogonal to wage loss and unemployment. As he puts it, "exposed and hosed are possibly orthogonal ideas."
1
0
7
One of the key streams of research in the economics of AI is producing exposure or automation indices. A variety of research teams, both academic and industry, have compiled these indices, often grabbing headlines predicting the death of jobs.
2
3
14