Wil Yegelwel

[[components/metabind-buttons-embed]]

Quick Info

quick-info-display

= this.notes

Met up with Wil and let him know I’ve asked for a team transfer via internal mobility
Walked through the Xiaofan Yang / Pranav situation and acknowledgements
Explained reasons for the move: H1B concerns, devaluing of my work, and destabilizations after letting Diego go
Told him I loved working with him and want to continue working on evaluations in any form possible
Wil was very supportive and nice, asked how he can make the process as painless as possible

Research sync: Wil asked me to go over the Codeshare experiment and how to do testing
Presented the idea of doing evaluations for users
1:1: Wil indicated that no slot has opened up on labs
I indicated I need to decide what to do next for myself as I believe my days are numbered in MOTO
Spent the end of the meeting discussing how to do Codeshare evals for user preferences

Asked whether I should be working on the feedback tool; he said yes
I agreed and created a design doc for a proposal, asked Xiaofan Yang, Wil & Pranav to review
Wil mentioned he’ll be away for 2 weeks
Said I should work on the LSP research during that time
If we have time, think about how we can systematically measure personal codeshare context similar to what we did with Vasilis

Asked him for guidance on concerns I had with Sasha’s work; he said he was too busy but trusts my intuition
He asked me to send an update to Vasilis about the fix, which I did
Messaged him to set up a sync regarding confusions around role/description
Set a time for 2-3pm on Tuesday next week (2026-03-10)
He replied saying sure and that we can set up a recurring 1:1

Discussed what to do about Vasilis; Wil suggested discussing messaging with Pranav and Xiaofan Yang, which was done
Waiting for Wil to respond on whether the proposed change for Vasilis will work for him
Had a group meeting with David Palaitis, Xiaofan Yang, Pranav, Gabe Mudel, Sasha Caskey, and Claude Greegard
Meeting was about Wil proposing changes to mcontext based on his research
Tense vibes throughout; David Palaitis was particularly combative
David insinuated the work wasn’t innovative and didn’t have much substance
One concession: David noted that improving one area improved another and unlocked “latent” abilities as a small innovative nugget
David raised the productionalization problem and the gap between research and production
Generally a very tense meeting

Discussed my reproduction of Vasilis’s user preference respecting LLM experiment
Wil said to do the minimum to satisfy Vasilis
Conducted experiments showing the static context approach was more brittle than originally imagined
Showed a final experiment with a working recipe + delivery mechanism involving hooks
Waiting on Wil’s feedback on the hooks-based approach
Had a 1:1 where we went back and forth regarding the static analysis experiment
Informed him about page 2 in my research project docs
Agreed we’d pursue the page 2 direction instead of static analysis

Reached out regarding experiment with Vasilis for user preference context integration
Said good job on getting this out quickly
Shared more positive feedback in the team channel
Asked if I was officially on “research”; if so, wants to meet
I said I think so since my tasks are done and remaining work is follow-ups & improvements that Xiaofan Yang has pitched as possibly doing for “research”
Wil wants to know more about what Xiaofan has pitched
Meeting scheduled for tomorrow (2026-03-04) 2:30-3pm, before the mcontext dev sync at 3pm

Met with Wil regarding evals + Miku test cases
Meeting turned into why not make make-loops public and I was essentially like I cant handle the inmail
Agreed to include Wil with the idea of making a tsdeploy command that runs make-loops