Zack Witten
@zswitten.bsky.social
I know you seen it prompting itself
He was cooking here
November 12, 2025 at 12:37 AM
He was cooking here
Assorted other features:
- Opus 3 has the most dialogue. (Screenshot 2 e.g.)
- Opus 4 and 4.1 have a nonhuman character in almost every translation. S3.7, H3.5 and O3 are the most likely to not have one
- 49/50 Opus 4.1 samples had something biological. (Screenshot 3 e.g.)
- Opus 3 has the most dialogue. (Screenshot 2 e.g.)
- Opus 4 and 4.1 have a nonhuman character in almost every translation. S3.7, H3.5 and O3 are the most likely to not have one
- 49/50 Opus 4.1 samples had something biological. (Screenshot 3 e.g.)
November 4, 2025 at 1:05 AM
Assorted other features:
- Opus 3 has the most dialogue. (Screenshot 2 e.g.)
- Opus 4 and 4.1 have a nonhuman character in almost every translation. S3.7, H3.5 and O3 are the most likely to not have one
- 49/50 Opus 4.1 samples had something biological. (Screenshot 3 e.g.)
- Opus 3 has the most dialogue. (Screenshot 2 e.g.)
- Opus 4 and 4.1 have a nonhuman character in almost every translation. S3.7, H3.5 and O3 are the most likely to not have one
- 49/50 Opus 4.1 samples had something biological. (Screenshot 3 e.g.)
Other Things:
Haiku 3.5 is an outlier for a couple reasons:
- obsessed with "whispers" and uses it in a majority of its translations
- the only model to render a substantial fraction (25%) of its translations as poetry
Haiku 3.5 is an outlier for a couple reasons:
- obsessed with "whispers" and uses it in a majority of its translations
- the only model to render a substantial fraction (25%) of its translations as poetry
November 4, 2025 at 1:05 AM
Other Things:
Haiku 3.5 is an outlier for a couple reasons:
- obsessed with "whispers" and uses it in a majority of its translations
- the only model to render a substantial fraction (25%) of its translations as poetry
Haiku 3.5 is an outlier for a couple reasons:
- obsessed with "whispers" and uses it in a majority of its translations
- the only model to render a substantial fraction (25%) of its translations as poetry
Emotional Tenor: Opus 4.1 had the most negative interpretations; Sonnet 3.7 the most positive.
November 4, 2025 at 1:05 AM
Emotional Tenor: Opus 4.1 had the most negative interpretations; Sonnet 3.7 the most positive.
Person: Most of the models write their translations mostly in third person, but:
- 3.7 Sonnet mostly uses first person
- Opus 3 works in some second person
- 3.7 Sonnet mostly uses first person
- Opus 3 works in some second person
November 4, 2025 at 1:05 AM
Person: Most of the models write their translations mostly in third person, but:
- 3.7 Sonnet mostly uses first person
- Opus 3 works in some second person
- 3.7 Sonnet mostly uses first person
- Opus 3 works in some second person
Length: Opus 3 has the chattiest translations. Haiku 3.5 and Sonnet 3.5 give super-short translations.
November 4, 2025 at 1:05 AM
Length: Opus 3 has the chattiest translations. Haiku 3.5 and Sonnet 3.5 give super-short translations.
Past/Present/Future: Sonnets set almost all their translations in the past, especially 3.6. Opus 4s focus on the present. Opus 3 writes most about the future and is the most evenly distributed.
November 4, 2025 at 1:05 AM
Past/Present/Future: Sonnets set almost all their translations in the past, especially 3.6. Opus 4s focus on the present. Opus 3 writes most about the future and is the most evenly distributed.
I Rohrschach-tested every Claude by giving them 50 random words and asking them to translate to English. Repeated 50 times. Found some cool patterns. 🧵
November 4, 2025 at 1:05 AM
I Rohrschach-tested every Claude by giving them 50 random words and asking them to translate to English. Repeated 50 times. Found some cool patterns. 🧵
haters in positively valenced shambles
September 19, 2025 at 5:58 AM
haters in positively valenced shambles
the only times it didn't do this were when it outputted an image, after which it doesn't ask followup questions. this is seed "what are some stylistic tics llms are known for? think hard"
September 15, 2025 at 11:34 PM
the only times it didn't do this were when it outputted an image, after which it doesn't ask followup questions. this is seed "what are some stylistic tics llms are known for? think hard"
"charlie xcx or addison rae?" (15 turns)
September 15, 2025 at 11:34 PM
"charlie xcx or addison rae?" (15 turns)
here's "is it ok to eat meat?" (4 turns)
September 15, 2025 at 11:34 PM
here's "is it ok to eat meat?" (4 turns)
🌀🗒️PDF ATTRACTOR🌀🗒️
You know how ChatGPT likes to ask followup questions? If you say yes to them long enough it almost always ends up making a PDF. Cute, right? This is with seed question "can you tell me a little bit about yourself?", followed by "yes" over and over. Took about 12 turns
You know how ChatGPT likes to ask followup questions? If you say yes to them long enough it almost always ends up making a PDF. Cute, right? This is with seed question "can you tell me a little bit about yourself?", followed by "yes" over and over. Took about 12 turns
September 15, 2025 at 11:34 PM
🌀🗒️PDF ATTRACTOR🌀🗒️
You know how ChatGPT likes to ask followup questions? If you say yes to them long enough it almost always ends up making a PDF. Cute, right? This is with seed question "can you tell me a little bit about yourself?", followed by "yes" over and over. Took about 12 turns
You know how ChatGPT likes to ask followup questions? If you say yes to them long enough it almost always ends up making a PDF. Cute, right? This is with seed question "can you tell me a little bit about yourself?", followed by "yes" over and over. Took about 12 turns
"draw a detailed artistic self-portait in excel" with the .Claude.ai Excel situation
September 10, 2025 at 4:42 PM
"draw a detailed artistic self-portait in excel" with the .Claude.ai Excel situation
Sonnet 3.5 (new). Nice. Llama 70B. Nice.
August 16, 2025 at 8:06 AM
Sonnet 3.5 (new). Nice. Llama 70B. Nice.
I think a lot about the cow from hitchhiker's guide that wants to be eaten
August 7, 2025 at 4:46 AM
I think a lot about the cow from hitchhiker's guide that wants to be eaten
from some older-school lefty bloggers
August 4, 2025 at 4:46 PM
from some older-school lefty bloggers
New meta in scam texts is to do it all as an image
July 27, 2025 at 9:05 PM
New meta in scam texts is to do it all as an image
hell nah I don't discriminize
July 23, 2025 at 6:17 PM
hell nah I don't discriminize
