We created our own GPT: TruthGPT
The Whole Truth of food and fitness powered by AI
The most trustworthy source of food and
fitness journalism in the country.
Editors note : Hello there. We’ve created a new product — a GPT that allows easy, contextual access to all TBT knowledge. Presenting: TruthGPT!
We’re operating at the intersection of cutting-edge technology and journalism here, and I’m so excited that all TBT readers — you — get an exclusive, two-day access to TruthGPT. Try it , and give us feedback. Please. Help us perfect it before we release it to the world.
Oh, and if you’re wondering why we made this, how it works, aand why the hell a food brand is tinkering with LLMs, Shashank Mehta explains it in today’s edition of Truth Be Told.
Tell me the dangers of AI for journalism in 50 words,” I asked ChatGPT.
It said:
“The dangers of AI for journalism lie in potential bias amplification, ethical concerns, and job displacement. AI algorithms can reinforce existing biases, compromise privacy, and spread misinformation. Additionally, the automation of news production may lead to a loss of human judgement, creativity, and the essential role of investigative journalism in holding power accountable.”
Then, I asked this: “Tell me the benefits of AI for journalism in 50 words.”
Here’s the response:
“AI can benefit journalism by automating labor-intensive tasks like data analysis, fact-checking, and content generation. It enables efficient news gathering and personalized content delivery to audiences. AI-powered algorithms can help identify patterns and trends in large datasets, aiding investigative journalism. It also enables innovative storytelling formats and improves newsroom efficiency and productivity.”
Do you notice a subtle difference?
AI thinks AI is beneficial when used as a tool, as an aid, for humans doing journalism. And it’s dangerous when it starts becoming the journalist itself. A very important distinction.
But wait, why is your clean-food brand talking about AI?
Because we’re on a mission to rebuild the world’s trust in its food. And on this mission, solving food and fitness journalism is as important as solving food itself.
I wrote about it when we launched Truth Be Told, our weekly food and fitness newsletter. As TBT is about to complete one year, my conviction has only grown stronger: what you read and believe shapes what you eat and repeat.
Which is why the AI chat. Because AI will shape the future of journalism. It already is.
And depending on whose hands you put them in, large language models (which power conversational bots like ChatGPT) can be a force for massive good or evil in journalism. Sadly, the latter has already started.
Many pseudo-news outlets are using AI to write their content, which is obviously problematic. Especially when AI itself warns us it tends to amplify bias and lacks judgement.
🤝 Warn your techie friend
The thing is, the mechanics of our content ecosystem incentivise biased writing. More polarised and shallow content gets more eyeballs. This means the incentive to continue using LLMs as journalists is high.
We can’t stop this. But we can create examples to the contrary. We can use AI for good. As a tool. And leverage its impressive abilities to make food and fitness journalism more accessible.
Hence…
🥁*drumroll*
Presenting, TruthGPT. A GPT that’s trained on all of The Whole Truth’s original, fact-checked and deeply-researched knowledge base.
What does that mean? And how is that different from ChatGPT?
Let me explain. Think of how you use ChatGPT. You ask it the same questions you used to ask Google. But unlike Google, it answers in sentences rather than links. And with context.
The concerning bit is ChatGPT uses the whole world wide web to source knowledge. That’s what it’s trained on. That’s why its answers are usually generic. And because the internet has a mix of truth and falsehoods, trusting ChatGPT’s responses is hard. One should be cautious.
That’s the problem TruthGPT aims to solve. We believe the real power of this new tech lies in training them on sharp, small, specific, and limited datasets. Why?
Let’s ask Chat GPT.
“Training a GPT on a sharp, small, specific, and limited dataset offers advantages such as reduced computational requirements, focused knowledge domain expertise, improved accuracy within that domain, minimised bias, and decreased risk of misinformation. However, it may limit the model’s generalisation and effectiveness outside of the specified domain.”
Perfect. This is what we want: a bot that answers only queries related to health and fitness. And from sources we trust are reliable. And does so specifically, accurately, and without bias.
That’s TruthGPT—our attempt to show how AI can serve good journalism.
And also solve a specific problem you have repeatedly told us about. Let’s talk about that.
👩💻 Take a crack at it
Since TWT began, people have loved our content. But they’ve had a constant grouse. Our stuff is too long. Be it TBT newsletters or TWT Academy videos. They are information-dense but take time to read and consume.
Especially so when you’re looking for the answer to a specific query like ‘Is olive oil good for frying?’ But to get there, you have to read a 2000-word article about all oils.
We know. We hear you.
But look, we chose the long-form route consciously.
The biggest problem with listicles and 30-sec reels is the loss of context. And devoid of context, journalism devolves into clickbait.
Who can grab your attention longer? Who can monetize you better? Information and education, the whole point of the damn thing, is lost.
So we had to choose: long-form writing with context and deep research or quick-to-consume information that’s mostly worthless and occasionally harmful.
The thing is, we don’t have to choose anymore. Thanks to LLMs.
Now, if you have a specific question like ‘Should I have whey protein during pregnancy’ or ‘Is it cool to have black coffee first thing in the morning’ — you can ask TruthGPT. There is no need to read a 2000-word article (which was written with a lot of effort and love, so do read it later. It’ll answer questions you didn’t know you had.) Voila!
Of course, TruthGPT, like all GPTs, is an experiment. We’ve tested it a bit, but we’re sure you’ll find a lot of cases where it doesn’t do a great job. And that’s expected. This isn’t a protein bar we are shipping. It’s an experiment, and we are all in it together.
So we need your help.
Just like all things AI, TruthGPT is experimental. No matter how much we test it, I’m sure there are broken parts we’ve not yet discovered. Please help us discover them.
Help us train TruthGPT. Help us make TruthGPT useful. Help us set an example for journalists and journalism across the world. An example of using AI for good. As an aid for those seeking the truth. Not as a weapon for manufacturing half-truths that serve as clickbait or, worse, serve vested interests.
So here’s a contest. And our call for help.
Play around with TruthGPT. Ask it as many questions. Take it for a ride and give us feedback.
What was the most useful answer? What was the most useless? Did it hallucinate? Did it show bias? Did it go off on some complete tangent?
The three most useful, most helpful, or most funny pieces of testing and feedback get a surprise TWT hamper. Yay. And thank you in advance.
PS: if you’ve been wondering, the cover image of this article is our imagination of TruthGPT – rendered using Midjourney, an incredible AI tool for images.
For the curious, here’s the exact prompt we used: “Show a classroom full of adults, no kids, with an AI robot teaching them about health, food, and fitness. The board should have illustrations of food which are clearly visible. We want to see the back of all the adults sitting in the class so that we can see the robot and board from the front. Make the robot look cool, fun, and approachable.”
🤖 GPT = Go Please Tell (we tried)
