doctor brames
brames.bsky.social
doctor brames
@brames.bsky.social
pretraining the bad machines
the discourse around AI here is completely bonkers. this is a highly technical field and we measure stuff and report on it. it's kind of dull! maybe there is some kind of evil cabal trying to kill art but I haven't found them yet
December 20, 2025 at 12:34 AM
setting up a google alert for your real name is a great way to find out you've died in several places
December 18, 2025 at 10:28 PM
one of the neat things about being a father is how bedtime is creeping to 8:30pm
December 18, 2025 at 2:33 PM
twitter really just feels like 4chan if /pol/ built a recommendation system for it

not good
December 18, 2025 at 1:57 AM
seeing more and more instances of everyone loving AI
December 17, 2025 at 11:46 PM
the recognizable 'AI tone' is a consequence of post training right? I don't really see it in pretrain only models

I think eliminating it is actually wickedly hard because people are much better at recognizing it than machines
December 17, 2025 at 10:08 PM
if anyone uses the wretched Abacus to aid their Mathematics one must assume they have Humours of the Cranium
December 13, 2025 at 5:07 PM
I have had it up to here! at the discourse
December 12, 2025 at 7:53 PM
all forms of understanding are essentially plagiarism. read a textbook chapter and understand it? plagiarism. use that knowledge on a test? buddy you better believe that's plagiarism
December 12, 2025 at 7:53 PM
if you figure out how I can play video games on a peloton screen while still getting reasonable exercise I will give you one gorillion dollars
December 12, 2025 at 7:51 PM
rare garf
December 12, 2025 at 5:40 PM
I keep having lucid dreams grounded in my academic experiences and at some point each time I realize I have not attended any classes so I am either going to flunk out or get fired depending on whether I was supposed to be attending or teaching. total vibe killer
December 12, 2025 at 4:03 PM
when AI drinks the water it also does plagiarism by stealing cold from it
December 12, 2025 at 3:08 PM
all this for "you are doing good" when my expectation was that I was doing good seems like a good use of time. good!
"just summarize your accomplishments in a few hundred words, in and out, an hour max" yeah sure. week absolutely obliterated
December 10, 2025 at 10:48 PM
ohhhhhh this is what has been bothering me about all of this
Ceding techno optimism to the right is a generational scale mistake
December 10, 2025 at 9:33 PM
how many parameters does a model need to have before using it is a moral failing
December 9, 2025 at 7:50 PM
ya gotta get the butter tray folks. it's just nice!
there are people who let their butter soften on the kitchen counter before using it and then there are people who don’t love themselves
December 9, 2025 at 4:58 PM
it has become clear to me that people who use AI have the brain pan of a stage coach tilter
December 8, 2025 at 1:19 AM
kiddo dropped "six seven nine eleven" on us this morning
December 6, 2025 at 5:41 PM
each time the train pulls into Penn Station I feel a podcast forming around me
December 5, 2025 at 11:53 PM
one of the interesting things about Bitcoin is that it is tracking black market activity in a way that is public and cannot be deleted even in theory, which seems like a liability for participants
December 5, 2025 at 4:04 PM
new incredibly disturbing safety scenario unlocked
this is possible... with the power of AI
December 5, 2025 at 9:56 AM
I wish the central limit theorem were taught early and intuitively. if you are going into any kind of technical field you will benefit from understanding it. but I didn't get to it until graduate statistics, which seems nuts
December 4, 2025 at 11:00 PM
I have been Yelled At for working in AI maybe twice?
December 4, 2025 at 6:43 PM
wait is it NeurIPS? thaaats why no one is at work
December 4, 2025 at 5:19 PM