👯 Allocation Multiplicity: Evaluating the Promises of the Rashomon Set by Jain et al. (incl. @kathleencreel.bsky.social) argues that allocation (vs. model) multiplicity should be seen as a pathway for reducing discrimination, homogenization, and arbitrariness in decision-making problems.

.. just as (to recall another Schliesser post) pragmatist philosophy diffused over decades into neighboring fields, influencing the very Operations Research whose tools fed #4 above. (7/7) digressionsnimpressions.typepad.com/digressionsi...

But also, technologies just take a long time to diffuse, even obviously useful ones like tractors: www.economist.com/christmas-sp... (6/7)

I would bet, however, that what Eric describes has come not from the latest generative AI but instead from tools that have been around in some form for 10+ years.

What's changed in that time? Libraries made it easier for non-technology companies to stitch well-tested packages together. (5/n)

Their stats/ml wizardry is in collecting the right data, stitching all the tools together together correctly, and creating dashboards to display this information and push recommendations to warehouse workers, managers, and drivers. These are guesses -- supply chain management is not my field! (4/n)

2. Time series forecasting to predict seasonal and holiday trends: facebook.github.io/prophet/
3. Graph-based inventory modeling to keep track of where everything is and should be: www.dgl.ai
4. Optimization tools to route vehicles efficiently: developers.google.com/optimization (3/n)

My guess (although Eric can confirm) is that there is no conflict here. The retailer's data team is using *non-generative* ML models to work their magic:

1. Deep reinforcement learning for dynamic inventory allocation based on tabular data of past sales. www.tandfonline.com/doi/full/10....

Thank you!!

Oh nice, thank you!!

This is a great lead — thank you!!

Oooh yes! That is exactly the kind of thing I was hoping existed, thank you!!!!

That’s great - thank you!