the new Fable still can't tell a joke
i think jokery evals plateaued with Google's PaLM models in 2022, no one has pushed SOTA since then
maybe another 10 trillion parameters will do the trick!
10:59 AM · Jun 9, 2026 · 11.7K Views
AI researcher Jack Morris argues LLM humor has plateaued since 2022 despite scaling up to 10 trillion parameters
He demonstrated the limit using the new Fable model
121061913.1K
Sentiment
Positive users praise Neuro-sama as SOTA for jokes and call specific replies hilarious or a great cat name, while negative users criticize RLHF for removing humor since 2022 and call the Fable model's pet jokes unfunny.
Pos
60.0%
Neg
40.0%
5 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS1.8KBOOKMARKS2LIKES11
Charles 🎉 Frye@charles_irl
if i had a NCCL for every bad LLM joke, i could train a funny LLM
Jack Morris@jxmnop
the new Fable still can't tell a joke
i think jokery evals plateaued with Google's PaLM models in 2022, no one has pushed SOTA since then
maybe another 10 trillion parameters will do the trick!
2hViews 1.8KLikes 11Bookmarks 2
REPLIES2
Apoorva@apoorvasriniva
@jxmnop it likes to make unfunny jokes about pets i guess
8hViews 580Likes 8
Benno Krojer@benno_krojer
@jxmnop maybe you just have to be human to be funny
9hViews 493Likes 2
Santiago M.@sanmking
@jxmnop Sorry, but that was hillarious!
9hViews 224Likes 2
Jack Morris@jxmnop
@apoorvasriniva did you actually name your cat Softmax though LOL
8hViews 471
Apoorva@apoorvasriniva
@jxmnop Indeed
8hViews 289Likes 2
Romain Lacombe@rlacombe
@apoorvasriniva @jxmnop Softmax is legit a great cat name! 😹
8hViews 144Likes 2
Jack Morris@jxmnop
@sanmking 🙃
9hViews 215Likes 1
Tatted@TattedWorks
@jxmnop Humor requires committing to something that might not land — which is exactly what RLHF trained out of every model since 2022.
The jokes didn't plateau, they got fired for making HR uncomfortable.
9hViews 144
Yajur@vickyzuck
@jxmnop Can it tell me if should walk or drive to my car wash?
8hViews 108
KaryoKleptid@kleptid
@jxmnop
8hViews 99
XaniWhatever@XaniWhatever
@jxmnop Neuro-sama is still SOTA in jokes department
6hViews 40
Rugbist@rugbist_
@jxmnop hard to tell if the evals themselves just got worse too
9hViews 37
John Thilén@JohnThilen
@benno_krojer @jxmnop because the AI's timing is digital
https://memory-alpha.fandom.com/wiki/The_Outrageous_Okona_(episode)
9hViews 16