Sasha
@_shydrie
Followers
67
Following
98
Statuses
65
Does anyone have a compelling Zotero + LLM workflow? Thinking I might throw together an MCP server for accessing local artifacts.
1
0
4
0
0
2
@cheese_mishra Interesting, I bet this is because they share the Milbrae station. I noticed some odd behavior traveling from my house where a short walk saves 90 minutes.
1
0
2
day 16: this experiment has gone on for slightly longer than the initial plan of two weeks. I think I shall retire it but try to be more proactive about posting project updates. frantic housing search has begun.
day 15: - travel day—returned to SF! - worked on blog backlog and CRDT project - housing plans for feb onwards have combusted. will probably be investing a lot of time into rectifying this
0
0
5
day 15: - travel day—returned to SF! - worked on blog backlog and CRDT project - housing plans for feb onwards have combusted. will probably be investing a lot of time into rectifying this
day 13 & 14: working on a crdt project at nysrg, adding collaborative editing to obsidian. I am surprised that there don't seem to be light-weight solutions in this space, though there are more comprehensive ones like Relay.
0
0
6
day 13 & 14: working on a crdt project at nysrg, adding collaborative editing to obsidian. I am surprised that there don't seem to be light-weight solutions in this space, though there are more comprehensive ones like Relay.
day 12: being unemployed is funny, weekends turn from optimal grind time into opportunities to catch up with employed friends. did nothing safety-pilled today, haven't even talked to claude
0
0
6
day 12: being unemployed is funny, weekends turn from optimal grind time into opportunities to catch up with employed friends. did nothing safety-pilled today, haven't even talked to claude
day 11, scheming (me, not the models): - planning out how to train my own transcoders and ideally eventually cross-coders - thinking I will train a NanoGPT model (124M params) following methods from the speedrun leaderboard—might be fun to fine tune for model diffing
0
0
6
day 11, scheming (me, not the models): - planning out how to train my own transcoders and ideally eventually cross-coders - thinking I will train a NanoGPT model (124M params) following methods from the speedrun leaderboard—might be fun to fine tune for model diffing
day 10: - exciting developments, minimal grinding
0
0
6
day 9: - had (and prepared for) an onsite. I am cautiously optimistic and mentally fried - I want to train a crosscoder
day 8: - caught up on work that some friends are doing at Tilde—it's exciting to see steering increase reliability in the real world - dove deeper into ARC's agenda. I am glad that people are working on theory but am not motivated to do so myself - big on-site tomorrow!
1
0
6
Some of my questions about theory: - are there theoretical results proving "interesting" properties of "real" models? - is imitative generalization still alive? - how could we identify if theory is neglected?
0
0
5