David Van Dijcke

Incoming AP @UVA Econ. PhD @umichECON. Academic Visitor @bankofengland. Alum @lmhoxford, @LeuvenEconomics. 🇧🇪🇺🇲. Views mine.

[David Van Dijcke @packlesshepherd]

I'm excited to share my job market paper (for the 2025-26 market)!

It introduces a new extension of RDD where outcomes are entire distributions: Regression Discontinuity Design with Distributions (R3D).

Thread below 👇 (1/)

[coarse @CoarseDotInk]

A few updates landed on @CoarseDotInk -- support for the latest models and some bug fixes / new features. Thanks to the contributors on GitHub!

More things coming soon!

[David Van Dijcke @packlesshepherd]

Looking forward to presenting this at the Econometric Society conference @EmoryEconomics tomorrow at 13:45

[David Van Dijcke @packlesshepherd]

I'm excited to share my job market paper (for the 2025-26 market)!

It introduces a new extension of RDD where outcomes are entire distributions: Regression Discontinuity Design with Distributions (R3D).

Thread below 👇 (1/)

[Chenhao Tan @ChenhaoTan]

Following @packlesshepherd we are adding BYOK option to openaireview. Now you can easily review longer papers without setting up the package locally.

[David Van Dijcke @packlesshepherd]

@lopez_prol @CoarseDotInk Great! Any idea what went wrong the first two?

[David Van Dijcke @packlesshepherd]

Big update on @CoarseDotInk: you can now run reviews at ~zero marginal cost through your AI subscriptions!

Just drop your paper into https://t.co/5m1Srky4H8, click "Review with my subscription" and follow the instructions

See https://t.co/ryn463RPeN --> "Use my subscription"

[David Van Dijcke @packlesshepherd]

@lopez_prol @CoarseDotInk Yes, copy paste the command in codex terminal, after launching codex with "codex --yolo"

[David Van Dijcke @packlesshepherd]

@_joaodias @CoarseDotInk Glad you like it and find it useful! Resuming the review is currently not implemented but seems worthwhile to do, I've added a GitHub issue for future versions. For now I'd recommend setting the reasoning effort lower or using the online version with an API key (pricier)

[coarse @CoarseDotInk]

1,000+ reviews served, 1.6K PyPi downloads, in a little over a week.

Thanks for all the love and feedback :)

[David Van Dijcke @packlesshepherd]

@PeiranXiao @guilhermejd1 @CoarseDotInk Just chunk them together into one file! I can add support for multiple files if you think that'd be useful

[David Van Dijcke @packlesshepherd]

@fatihkansoy @CoarseDotInk is not for profit, costs typically less than 2$, and can be run at zero marginal cost on your local subscription though :)

Good prompting can get you a good chunk of the way but doesn't make up for a good harness on top

[David Van Dijcke @packlesshepherd]

@sanjayrajsingh @ChatGPTapp @CoarseDotInk @RefineDotInk @reviewer3com To be clear, for the local version the only API requests are for the PDF OCR the rest runs on your code subscription

[David Van Dijcke @packlesshepherd]

@sanjayrajsingh @ChatGPTapp @CoarseDotInk @RefineDotInk @reviewer3com Thank you! Yes I'd be happy to do that. Could you request this through a GitHub issue?

[David Van Dijcke @packlesshepherd]

@guilhermejd1 @CoarseDotInk Yeah tex files skip the ocr step. The agent might still ask for a key in the way it's currently prompted. Let me fix that!

[David Van Dijcke @packlesshepherd]

Post-script:

Since this allowed me to run more reviews without breaking the bank, I've updated the comparison table on https://t.co/f7BXRigD93 with @ChatGPTapp-5.4 results.

@CoarseDotInk outperforms @RefineDotInk, @reviewer3com, and Stanford Agentic Reviewer across the board!

[David Van Dijcke @packlesshepherd]

I believe this is a big step towards making not-for-profit, open-source AI reviews cheap and available to as many academics as possible. @ChicagoHAI @ChenhaoTan

This is a first release, so please submit any issues here:
https://t.co/wS3xubwsBl