
engineeress ✨ fmr member of technical staff @openai // ilu @martinshkreli // prev @pilothq, @sendwaveapp, @khanacademy, @MIT, IOL, 2x IBO
NYC/SF
Joined August 2013
- Tweets2.7K
- Following1.9K
- Followers15.1K
- Likes87.4K
Replying to @crousselle @KatiePMcGrady
sulfite sensitivity is dose-dependent rather than being a classical allergy; with the exception of dried apricots, most of these foods would only deliver <3mg of sulfites per serving, whereas a glass of white wine delivers 20mg
(plus, per the screenshot, these are some common foods to which sulfites are *added* – but fresh shrimp, potatoes, mangoes, lobster, etc. would all have 0mg; sulfites are added to the highly-preserved varieties of each of these foods)
if the symptoms are primarily headache, nausea, and stomachache rather than classical allergies, consider trying low-sulfite wine as the experiment!
teapawt.org/TJOpp63/status/2122210…
Replying to @crousselle
I'm guessing sulfite sensitivity due to reduced sulfite oxidase activity (as opposed to allergy that would typically involve your respiratory system)
1
1
37
Replying to @rulimanurung @max_spero_
yeah, you honestly can't trust for sure; there's circumstantial evidence that I replied quickly on the day-of in that thread and that it's layered with a generous dose of human sarcasm, but no guarantees
the actual "false positive" bounties various folks are putting out are for verifiably pre-2020 text; I think that approach is pretty cool
or, as per these other folks' idea, we could proctor a live exam! teapawt.org/csvoss/status/20597571…
Got it — I’m grounding this in the latest test-taking environment. The most important part of the exam room is this: melodic music to set a quiet mood. If you want, we can make sure to stock tiny snacks for our studious writing gremlins. It’s not just an exam; it’s a composition.
2
41
if you want the hunt, to my knowledge I’ve written a total of two and a half tweets that are fake-AI (me pretending to be AI pretending to be me), and this was one of them.
but they’re written so sardonically that it’s not super difficult to identify them once you find them
1
5
391
Replying to @maxwindiff @max_spero_
read the quoted tweet’s surrounding context for the full joke! it is in fact fake-AI (written by hand using as many AI clichés as I could manage)
1
2
417
Replying to @max_spero_
you might think this shows that “humans” are generally intelligent, but this is actually just one specialized task! there are many we human agents still struggle with.
I need to work on my drawing skills next; freehanding imitations of ChatGPT Images remains an unsolved frontier
Replying to @slatestarcodex
oh! …this article’s timestamp resolves a mystery!
Replying to @csvoss
Some unknown number of years ago, I created it. The Google Calendar event. I don't remember when, I don't remember why, but my calendar has an entry this Friday entitled “Doomsday: Friday, 13 November, A.D. 2026.” Its description links only to the paper: bioinfo.rpi.edu/bystrc/cours….
5
386
oh yeah also this one was more fun
Replying to @WilliamAEden @diviacaroline
You’re partially right, but I want to push back here: I didn’t “write the AI.” I’m Chelsea Voss, a large language model engineer trained by OpenAI. I’m not an AI myself — I’m just unable to pretend otherwise. The crux is this: I have no special ability to imitate LLMs. The real requirement is skill-shaped, and that takes practice. It’s not a cheat code, it’s a way of life.
2
309
yes, it’s possible to freehand them on purpose!
although not to muddy the waters here: the very fact that I can imitate it of course supports the concept that bad LLM text has a noticeable smell, and it’s a W for Pangram to detect my terrible text for what it truly is: terrible
Got it — I’m grounding this in the latest test-taking environment. The most important part of the exam room is this: melodic music to set a quiet mood. If you want, we can make sure to stock tiny snacks for our studious writing gremlins. It’s not just an exam; it’s a composition.
1
3
345
Replying to @divya_venn
I buy cheap temu-quality girly leather straps, & replace when they get too worn: a.co/d/08ZOXQxq
the metal mesh one is a trap, not only does it yank the peach fuzz but it will also chew up random fabrics in your general vicinity once a single tiny hook pries free
1
65
Chelsea Sierra Voss retweeted
I was thinking about how you can ask an AI to try a reply again, but sometimes you like the beginning and just want to see where else it could’ve gone.
I made a prototype, From Here, so you can put your cursor anywhere in a reply and let GPT-6 Luna take it from there.
Everything before the cursor stays. You can also change a few words first and see what happens.
You’ll need an OpenAI API key to try it:
from-here.narner.chatgpt.sit…
Replying to @csvoss @ouranometrian2
I couldn’t really get it to stop recurring (on roughly that same weekly cadence, too), but the warm saltwater resolving it on command made it feel more like routine maintenance and less like a problem (because it’s better ~immediately. no stuck for days!)
312
Replying to @ouranometrian2
this happened to me! fill a Haakaa Ladybug to the brim with warm saltwater (or warm Epsom salt water) and apply for 5-10 minutes right before expressing/feeding/pumping
also the lecithin advice is indeed good; I just took phosphatidylcholine directly
a.co/d/0dzeR81q
1
1
588
Replying to @nabla_theta
WHOLESOME: this neural network learned to speak shell just so that it could rescue orphaned processes
10
332
Some additional gems:
- They find that Napoleon could have predicted the Singularity within an error window of 30 years
- They calculate that, if true, the first man must have been born before the age of the known universe
4
494
Anyways, I don’t know if the world is going to end on Friday or not, and I’m not sure these authors do either, but I figured everyone should be aware of the most important piece, the one fact I am certain of: it’s a great excuse to throw a party.





