Next was a jam-packed panel on the competition risks of AI acquihires with @florianederer.bsky.social, @lugaricano.bsky.social, @igorletina.bsky.social, @imarinescu.bsky.social, Hans Zenger, Jonathan Kanter, and @cristinacaffarra.bsky.social at the @cepr.org www.youtube.com/watch?v=5zrS... (4/7)

How about this: whoever has an H1B and an employment contract in the US gets an automatic work permit in Europe. No bureaucracy, no nothing. US firms can simply tell their stranded workers they can work from one of their offices in Europe.

Link: assets.publishing.service.gov.uk/media/68c3e8...

I've heard from a few people that they use the same approach as you. To me, it appears not to be too much different from just sending the bullet points in the first place.

There are two issues with this. One is that I don't know if I am reading the person's opinion or the chatgpt's opinion. The other is that the prompt is usually much shorter than the output, so that this practice leads to longer e-mails.

I think I also feel like it is not really polite to send AI text to someone and then expect them to read it, especially since AI tends to write too much. I think I'd rather if the person sent me the prompt they used...

I used to love using em-dashes in my prose, but now that it has turned into a chatgtp "tell" I actively avoid it.

Additionally, theory can be a first, cheap test of whether a novel design idea is good and worthwhile to be tested empirically.

But abstract *design* theorizing can be useful.

Empirical work mostly examines institutions, policies and mechanisms that already exist and normally does not generate novel designs. Theory does.

Sources:
- www.reuters.com/technology/m...
- www.reuters.com/technology/a...
- www.reuters.com/technology/a...
- www.reuters.com/sustainabili...
- www.reuters.com/business/goo...

Regarding economic consequences and possible policy implications of acquihires, I have a recent paper (doi.org/10.1016/j.eu...) as do Bar-Isaac, Johnson and Nocke (pubsonline.informs.org/doi/10.1287/...).

But there is much that we still don't know and I hope to see more work on the topic!

Compensating equity holders is important for a number of reasons, from eliminating future litigation risk to maintaining good reputation with VCs.

If the buyers were not worried about antitrust scrutiny then they would probably go for a straight acquisition (as big tech did many times in the past before the authorities started scrutinizing startup acquisitions). This is simpler and cleaner for all parties involved.

I think the deals are structured in this way for two reasons: 1) avoiding antitrust scrutiny and 2) compensating startup's equity holders.

In the Meta/Scale AI deal, Meta acquired a minority stake in the startup. Again, this can make economic sense for Meta even if, after key talent has moved to Meta, what remains of the startup is worth very little.

A "reverse acquihire" only lacks 2), which is irrelevant as the only thing left is a husk of the original startup anyway.

In all but the Meta/Scale AI deal, the equity holders were compensated via a tech licensing deal. Whether the tech is worth anything is beside the point.

"Reverse acquihire" is really a misnomer -- a "faux acquihire" would be better. A traditional acquihire has 3 elements: 1) transfer of talent from the startup to the buyer, 2) acquisition and shutdown of the startup, and 3) compensation of startup's equity holders.

If you are interested in the economics of talent hoarding, @jmbenkert.bsky.social, Shuo Liu and I have a paper about it that was just accepted at the European Economic Review (and which is open access): doi.org/10.1016/j.eu...

2/2