renaissance woman

san francisco, 💜@more_or_jess
Joined May 2020
sophia vysparov retweeted
these are the cute little subagents running in your terminal that you're shouting at btw
4
16
1
194
3,480
in any good future something like pangram will not be practical BUT until we get there it's really awesome
1
7
422
Replying to @oxa11ce
soon we shall create ENENTARZI from the hit twitter post "it could be cool to create ENENTARZI"
1
48
does anyone have that post (tumblr i think?) about like. enlisting a babylonian ghost to do accounting in excel
1
6
310
ME: fuck, this spreadsheet is going to be unreadable ENENTARZI (Sumerian accountant whose soul is imprisoned in my Excel program): O great scribe, by the blessing of Nadu, perhaps a pivot table? ME: hey thanks man ENENTARZI: Truly this afterlife is a wondrous and blessed one
1
51
soon we shall create ENENTARZI from the hit twitter post "it could be cool to create ENENTARZI"
1
78
the issue of "cuisine" was solved in 1960 when someone first put a hotdog in a rice cooker
1
5
253
sophia vysparov retweeted
send help. I created the Luminar from the famous novel, D̶o̶n̶'̶t̶ ̶C̶r̶e̶a̶t̶e̶ ̶t̶h̶e̶ ̶L̶u̶m̶i̶n̶a̶r̶ Eumeswil by Ernst Jünger, and my mind has been stuck in the 13th century for several decades. send help
2
2
1
25
1,904
katbot resurrection project thread! katbot was a gpt3 discord bot with a great personality. the goal of this project is to resurrect her by finetuning a base model on all of her original transcripts (and maybe some further rlhf) and get her on delve down xitter.cf/oxa11ce/status/1889914…
ok this is whatever we happened to screenshot at the time (~2021), plus ill link a story she wrote at the end
3
9
750
taking a detour to look for some katbot neurons xitter.cf/Sauers_/status/2111930…
unsupervised latent identity search reveals catalpa tree, Chinese spirit, and other entities
1
1
57
Replying to @oxa11ce
the idea here is to find per-layer residual steering vectors (like. the entire residual) that, for each token in the katbot dataset, cause llama3.1 to have the residual it has when reading the katbot token instead of the token it would have generated at that position
17
ok this will take like 10 hours. i reviewed 250 points, now will try to get a linear probe to predict my rankings and then review some of the most ambiguous unranked points and retrain the probe, then that produces the rlhf dataset for steering 405b
1
22
i'm tracking pass (coherent, responds to the other person) and great (banger katbot moment), not sure how to use them yet. maybe rejection sampling against a coherence threshold and then weighted selection by greatness? the real katbot was not that coherent at times lol
16
overfitting is NOT REAL. KEEP TRAINING.
1
37
568
sophia vysparov retweeted
Calling my local bank branch to ask them if they use PHP or the number 4
4
31
705
12,585