Berkeley has a YouTube video explaining what the LOR process does. Basically they run the applications through an algorithm that calculates the ‘read score’ of the application. It is an estimate of what an AO would rank the application if they read it themselves. The algorithm only works based on stats though. Of the people who receive a LOR request, 80% were identified by the algorithm and 40% were identified by AOs while the 20% was identified by both. These candidates are all borderline applicants. The LORs help admissions officers admit students who may be great but have lower scores and reject students with high scores who may not be so good. Ideally for the people with lower stats who received an LOR request, submitting LORs can only help them, since they were going to get rejected anyway without that extra information. For people with higher stats, LORs can only hurt them, since they had a higher chance of getting accepted in the first place. If they had a good LOR they would still be accepted, but a bad LOR might get them rejected.
Here’s the link https://youtu.be/M-4pi-zlk8g?t=1668