0.25
0.5
0.75
1.25
1.5
1.75
2
You Talk Funny! One Day Me Talk Funny Too: Investigating the Capacity of Large Language Models for Humour
Published on Oct 29, 202412 Views
Keynote III
Related categories
Chapter list
One Day Me Talk Funny Too - 100:00
One Day Me Talk Funny Too - 201:56
One Day Me Talk Funny Too - 303:52
One Day Me Talk Funny Too - 406:41
One Day Me Talk Funny Too - 508:38
One Day Me Talk Funny Too - 610:25
One Day Me Talk Funny Too - 711:44
One Day Me Talk Funny Too - 814:29
1985 (90%) unique Xs (301 F), 665 (30%) unique Ys (302 F), 1460 (66%) unique Zs16:01
Explanation: Red meat is an aggressive builder of cancer all over the body.18:12
Prompt GPT-3.519:14
783 (77%) unique Xs (36% F)20:08
Z is more similar to X than Y21:57
Eliciting Ironic XYZs from GPT3.5-Turbo25:04
763 (70%) unique Xs (34% F)26:13
Neutral & Ironic XYZs27:23
What is the dominant emotion in each XYZ?28:36
Mean Positivity/Believability/Fairness of Each30:01
Pearson’s31:08
Well, this LLM is no Einstein at irony!32:17
Let them eat KFC33:41
Sample Responses from LLM to “ironic” prompt35:25
Web XYZs37:59
Neutral LLM38:36
Ironic LLM39:00
Narendra Modi is the Stephen Hawking of humility. - 139:41
Narendra Modi is the Stephen Hawking of humility. - 240:29
Narendra Modi is the Stephen Hawking of humility. - 341:11
Narendra Modi is the Stephen Hawking of humility. - 441:14
GPT-3.5-Turbo - 142:09
GPT-3.5-Turbo - 243:21
So Believability remains crucial to ironic selection.44:08
Irony stirs external debate, requires internal dialogue47:03