Cryptocurrency news

Google DeepMind discovered that AI can't joke

Forklog / 24.06.2024 / 14:27
Google DeepMind discovered that AI can't joke

AI-based chatbots do not know how to write funny jokes, Google DeepMind researchers found out during an experiment. 

The experts asked 20 professional comedians to use chatbots like ChatGPT and Bard to write stand-up, and then offered to evaluate the results and share their thoughts. The experiment was conducted at the Edinburgh Fringe Festival in August 2023 and on the Internet. 

It turned out that large language models (LLM) cannot compose funny text. Stand-up artists noted that the result was "tasteless", "very bad" and "boring". One of them stated that he had received "cruise comedy material from the 1950s, but a little less racist."

According to the results of the study, AI can create diagrams and fragments of long-running stand-up programs, but it lacks the humanity that would make the material funny. People "add a lot more nuance, emotion and subtlety through their life experiences." 

AI does not understand what sarcasm, black humor or irony are. 

"There are dark moments in many of my things, and here he did not write dark things to me because he decided that I wanted to commit suicide. So he just stopped giving me something," said one of the participants in the experiment.

The comedians emphasized the importance of cultural context and audience.

"The kind of comedy I could shoot in India would be very different from the one I could shoot in the UK because my social context would change," one of them said.

Humor requires a change of framework, so it is difficult for a machine to master, said researcher Thomas Winters. 

LLM generated content based on all the accumulated materials, and trying to cover all possible disciplines, which introduced a certain amount of bias.

"Participants noted that existing moderation strategies used in security filters and LLM training reinforce hegemonic viewpoints by destroying minority groups and their views, and qualified this as a form of censorship," the study says.

Popular LLMs are limited, Google DeepMind experts noted. They strive to produce honest, harmless and useful results. 

Recall that Google's AI scientists are developing technology to create video soundtracks.

Previously, experts from DeepMind created a virtual rat with AI as a brain.

Source
Recently News

© Token Radar 2024. All Rights Reserved.
IMPORTANT DISCLAIMER: All content provided herein our website, hyperlinked sites, associated applications, forums, blogs, social media accounts and other platforms (“Site”) is for your general information only, procured from third party sources. We make no warranties of any kind in relation to our content, including but not limited to accuracy and updatedness. No part of the content that we provide constitutes financial advice, legal advice or any other form of advice meant for your specific reliance for any purpose. Any use or reliance on our content is solely at your own risk and discretion. You should conduct your own research, review, analyse and verify our content before relying on them. Trading is a highly risky activity that can lead to major losses, please therefore consult your financial advisor before making any decision. No content on our Site is meant to be a solicitation or offer.