hpr-knowledge-base/hpr_transcripts/hpr3803.txt

Episode: 3803
Title: HPR3803: Chatbot hallucination
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr3803/hpr3803.mp3
Transcribed: 2025-10-25 05:33:00

---

This is Hacker Public Radio Episode 3803 for Wednesday 1 March 2023.
Today's show is entitled Chatbit hallucination.
It is hosted by DNT and is about seven minutes long.
It carries an explicit flag.
The summary is, the inevitable show featuring a segment written by the chat bit chat.
Hello and welcome to another exciting episode of Hacker Public Radio.
This is your host, DNT.
Today we're talking about ChatGPT a little.
In late 2022 something called ChatGPT came out.
It was an AI-powered chatbot made by an organization called OpenAI that could answer questions
and write stuff for you.
It could even write code.
It was all the rage in the news with the usual deluge of online articles expressing either
extreme praise or extreme alarm.
From OpenAI's frequently asked questions section about ChatGPT.
Can I use Output from ChatGPT for commercial use?
Subject to the content policy in terms.
You own the output you create with ChatGPT, including the right to reprint, sell and merchandise,
regardless of whether output was generated through a free or paid plan.
Turns out ChatGPT can also write episodes of Hacker Public Radio.
That's coming shortly.
From Wikipedia, OpenAI acknowledged that, quote, ChatGPT sometimes writes, plausible sounding,
but incorrect or nonsensical answers.
This behavior is common to large language models and is called hallucination.
Here is the script.
Greetings and salutations.
This is the inverted rabbit here to bring you another episode of Hacker Public Radio.
Coming to you from the depths of an LSD trip.
Today we're exploring the fascinating intersection of Mandarin-speaking pythons, Python-speaking
mandarins, Xi Jinping, and the great people of Iowa.
Let's start with Mandarin-speaking pythons.
That's right.
You heard it here first, folks.
Pythons are not only masters of constriction, but they can also speak Mandarin.
This is made possible through the application of advanced language algorithms and machine
learning techniques.
And now let's turn our attention to Python-speaking mandarins and other citrus fruits.
Yes, you heard that right.
These fruits are not just delicious and juicy.
They're also able to communicate with pythons using the Python programming language.
It's a fascinating combination of technology and nature.
Now let's talk about Xi Jinping.
Xi Jinping is a well-known leader and advocate of the Chinese people who has a deep
appreciation for technology and innovation.
He's also known for his love of good food, which is why we believe that Mandarin-speaking
pythons could play a crucial role in enhancing the culinary experience of his guests.
And finally, let's turn our attention to the great people of Iowa.
Iowa is a state in the United States that's known for its friendly and hardworking citizens,
its fertile farmland, and its rich cultural heritage.
By combining the Mandarin-speaking pythons with the Python-speaking mandarins and the
people of Iowa, we can create a culinary experience that's unique and showcases the best
of both worlds.
So you see, by bringing together Mandarin-speaking pythons, Python-speaking mandarins, Xi Jinping,
and the great people of Iowa, we can create a unique and innovative culinary experience
that's unlike anything else.
It's a testament to the power of technology and the ingenuity of humanity, and that's
all for today, folks.
Join us next time for more trippy adventures with the inverted rabbit over and out.
Now it's a D&T again.
I have to say it was pretty fun to make chat GPT hallucinate, and I think chat GPT had
a good time too.
I didn't write down the prompt I used to get this output, and I actually cheated a little
bit, and I merged in here one paragraph from another script that it wrote.
There has actually been some debate about the matter of whether anyone can have rights
over the output of chat GPT.
If chat GPT doesn't actually have rights over the training data it used, we of course don't
care because we want free culture anyway, so let's not waste time on that matter.
Recently there was some controversy in the chess world when a chess player accused
another of cheating, reading about the subject I came across an article in which the accuser
Magnus Carlson said he and other top players will sometimes play a second best move, just
to force their opponent off book, that is to force them away from a well-known game
progression and force them to be creative.
The idea is that in that type of game a player like Magnus Carlson stands to win.
I think making chat GPT hallucinate shows how these AI chat bots, at least today, are
simply imitators, and just as with human imitators, they fall apart under pressure, and quickly
show how they don't actually know what they're saying if that makes sense.
The problem is, for much longer than open AI has been training chat GPT, we have been
training ourselves and each other to be imitators, we're using words, patterns, and styles
that others see as smart or fashionable, and maybe we use them without even understanding
why they're smart or fashionable.
At this game I think computers will overcome us, just like before them, machines over
came us at our manual tasks, so perhaps like Carlson the chess player, we too should start
playing second best moves every once in a while, just to force the people around us to
be more creative, not only when they speak, but even when they listen.
So if you're like me and you see things around you and you can't help but think about it,
maybe make connections, maybe even forms and theories, if you're just fascinated with
that new type of joiner you've been learning about, or if you're just so sure you have
discovered the ideal X window manager for you, maybe you'll live with your partner, and
here she just can't stand your observations anymore, record an episode of Hacker Public
Radio, we'll listen to it.
To do that go to Hacker Public Radio dot org, click give shows, have a read, and when
you're ready, click upload.
Thanks for tuning in, and come back tomorrow for another exciting episode of Hacker Public
Radio.
You have been listening to Hacker Public Radio, and Hacker Public Radio does work.
Today's show was contributed by a HBR listener like yourself, if you ever thought of recording
a podcast, and click on our contribute link to find out how easy it really is.
Hosting for HBR has been kindly provided by an honesthost.com, the Internet Archive
and our Sync.net.
On the Sadois status, today's show is released under Creative Commons, Attribution 4.0 International
License.