Emma Wiles

How many times is too many times for me to bring up the new Taylor Swift album at today’s faculty meeting?

Identify a tick, like the bug! Nothing newfangled!

Asked my teenage babysitter who just started vet school to ID a tick. Without blinking, he took a photo and uploaded it to ChatGPT.

The Youths are using AI with the same reflex my generation had for googling or watching youtube tutorials.

There are about 100 dynamic constraints that need to be satisfied for my newborn to sleep.

If this is representative, I do not see how the human race is still alive

bless you. But also, trust no one, you know?

Yep

Rough day for those of who work with large companies to study the effects of AI

Huge day for me, frankly

Lots of people dunk on this, but replace dolls with medical equipment or any number of food items and it's less funny.

Broke it

I’ve thought about the Maggie Smith poem Good Bones every day since she was born

Happy to announce that @edwardwiles.bsky.social and I have a new addition to the family! Welcome to the world, Penelope!

Trump has perfected the art of Pareto Worsening, a change in allocation making at least one person worse off without making a single person better off.

Trade war, Medicaid cuts, science cuts, there are no winners here, only losers.

New office decorations just dropped

Your tariff rate is your star sign divided by the number of boys you've kissed, multiplied by one

👏 come on universities, let’s see more of this. Conceding has gotten Columbia jack s***, let’s stand up to the bully together

In Bloomberg interview, Eisgruber signals that Princeton will not make concessions University President Christopher Eisgruber ’83 signaled that Princeton would not make concessions to the federal government after news broke that the Trump administration had suspended dozens of the U...

Relatedly: University President Christopher Eisgruber signaled that Princeton would not make concessions to the federal government in an interview with Bloomberg after news broke that the Trump administration had suspended dozens of the University’s research grants.

📌

10/10

Thank you ☺️

Thank you!!

TLDR: AI caused a lot more job posts, but no more jobs!

While the intervention did save employers time, the welfare loss to jobseekers from time wasted on applications is 6x the size of the welfare gain to employers. Suggests widespread use of LLMs can harm market efficiency.

After the experiment, the market rolled out the tool to 95% of new employers. A difference in difference analysis suggests results hold or get worse as more of market is exposed to AI.

Here, you can see that in the control group, job posts where the employer spent almost no time writing receive fewer applications. However, in the treatment group, job posts where employers spent 0-2 minutes writing the job post get *more applications*!

The AI appeared to dilute signals of employer seriousness: here we show that in the control group, the time an employer spent writing a job post is correlated with how long the job post is. In the treatment group, this relationship is much weaker.

Why? First, the treatment appeared to induce job posts from employers who were less inclined to make a hire (marginal jobs.) Second, even the inframarginal jobs were affected by the AI treatment---the job posts in the treatment group look significantly more generic than those in the control group.

However, employers in the treatment group were much less likely to make a hire. Despite the ~20% more job posts, and the increase in applications, there were *no more* jobs formed in the treatment group than in the control group.

And the first order effects look good! Employers in the treatment group were 19% more likely to post a job than those in the control group, those who submitted job posts spent 44% less time writing them, and the treated job posts received more applications!

Take up of the service was widespread- 75% of employers opt-ed in to receive the AI-generated first draft. However, despite encouragement from the interface to edit the post, most employers did not make any edits.

We randomly assign newly registered employers to an online labor market to a control group or treatment group after they have begun the process of posting a job. We offer treated employers access to a GPT3.5 written job post, based on a short description of their needs.