As the boom in text-to-image generation rages on, other sorts of AI tools are starting to become more accessible too. Today, I’ll explore the creative potential of language models, and share some essential tricks for working with ChatGPT. For those not yet familiar with the term, Language Modelling (LM) is the use of various statistical and probabilistic techniques to determine the probability of a given sequence of words occurring in a sentence.
You’ll have already seen this applied to your email, recommending how you might want to reply to someone, or when your smartphone keyboard determines the word you want to write based on the first letter you type.
Text-to-text tools have been accessible to the public longer than text-to-image tools, but the lack of a user-friendly interface combined with an absence of social imagery has limited their popularity. Though there are several other tools out there that work with text creation and machine learning algorithms, I will talk about ChatGPT specifically, as it has totally changed the paradigm in recent weeks, accumulating users at an unprecedented rate.
ChatGPT uses a specific language model that utilises advanced machine learning algorithms to analyse large amounts of text data and generate natural language responses. At the time of writing, the primary uses for ChatGPT are chatbots, translation, and content creation, but, as you’ll see, it has the potential to be a complete game changer for all creative industries.
As I always warn, whenever you use any of these Machine Learning tools, they will have limitations; as they have been trained on a specific database, they will be biased by it. They can generate incorrect or false information, and they will have limited knowledge of the world, or anything that has happened after its training. Nevertheless, I believe these are tools that will change the way we write, create and even think and share around specific experiences despite their oddities and statistical limitations, as Ethan Mollick put it:
Weirdness is a benefit in idea generation. Chat-based creative AI can be untrustworthy, presenting fake results as real with confidence, or even “hallucinating” a fake answer. This can be a problem if you are a doctor hoping to get trustworthy diagnostic information, but it isn’t if you are trying to generate ideas, where strangeness can be a benefit, and bad ideas are filtered out.
HOW DOES IT WORK?
That’s the most amazing part. You write what you want, and it appears as an answer, always in text. It’s a chat with a bot that gives you answers to whatever you ask.
CASES
1- Context creation: ChatGPT can be extremely fast and helpful when creating imaginary context and specific content-generating descriptions of places, objects, and events. This can be particularly useful for creatives who want to create vivid and detailed worlds for their stories. By providing a few key details, ChatGPT can generate a wealth of potential ideas, from pure fiction to providing accurate factual information.
Following the idea of context creation, you can go as deep and intricate as inventing an entirely new language to whatever level of detail and complexity you want. Here is an example I put together in 5 seconds:
Beyond this, GPT-3 can create a lot of space for speculative design and future creations. As an example, I love this invented letter from Mark Zuckerberg apologising to everyone and remodelling Meta’s model from zero.
2. Academic comparisons: Even if you ALWAYS have to double-check what is written (and sometimes this takes more time than writing it yourself), it’s fantastic to see the speed at which this tool can write comparisons and find relationships between academics or any documented character. I recommend testing it yourself with your favourite writers, designers, filmmakers, or people that have been intellectually important to you.
3. Tone & synthesis: ChatGPT can automatically summarise a long text into a shorter, more concise form. This can be useful for quickly extracting the main points and ideas from a text without reading through the entire thing.
Additionally, it can be used to change the tone of a text. This can be useful for adapting the tone of a text to match a particular audience or situation. For example, ChatGPT could be used to change the tone of a text from formal to casual or from serious to humorous (or sarcastic, romantic, nostalgic, angry, suspenseful, or inspirational).
4. Actor Creation: You can build any character and instruct the chat to adopt this persona for you; perfect if you want to recreate a complex situation with other people or rehearse an interview. You can add A LOT of specific characteristics into these actors, psychological traits, or attitudes.
5. Prompt creation: It can be used to help you imagine prompts for text-to-image tools; here is an example copy pasting the ideas into Midjourney and adding “Library that” at the beginning of the prompt. If you are interested in text-to-image, check out this article.
and here are some of the image results:
6- Creative process: You can build a methodology for the chatbot to follow and create multiple ideas. For me, these prompts are the most interesting, as it is like programming a creative tool specifically to give you insights and ideas for your specific field of work and following your methodologies; figuring out the right prompt will need a lot of time, but once you get it right, you’ve got gold. Here are two simple examples:
7- Naming and the character design: This is especially interesting for writers and a lot of fun for anyone else.
8- Code: Yes, ChatGPT can help you code, animate, explore and learn lots of different languages.
CASES FOR DATA VISUALISATION
As Domestic Data Streamers, our focus will be on exploring data and how it can be translated and communicated in better ways. These tools are a fantastic way to play with linguistics and the way we represent information.
1- Quick research: it’s crazy good to see how you can pull draft data within 5 seconds when exploring a topic. In any case, you will still need to work with experts and update the information with the latest reports and information available, but it allows you to speed up a first research framework.
2- Find the right tone for each number: what could happen if Abraham Lincoln was explaining to you some data sets? In the examples below, I’ve transferred a spreadsheet with the components of a Fatberg into an interview between Oprah Winfrey and Judith Butler and then into a scene where Samuel L. Jackson explains the same content.
3- Change format: You can move from a text to a table or vice versa. In the examples below, I’ve transferred a spreadsheet with the top ten GDP countries in the world, and I’ve asked for a critical perspective on why GDP is not the best indicator to see if a country is a good place to live in or not.
4- From data to poetry: Yes, you can transform spreadsheets into poetry; in this case, we turned the same data as the example before into a cheesy rap song:
5- Be weird: Mix random ideas to see what comes ups; one of the beauties of this tool is its ability to find bizarre and unconventional connections, find yours, look at your interests and your territories of interest, and mix them up as a source of inspiration.
EXTRA TRICKS
- Critical Reviews: A fascinating trick is to ask for a story, then right afterward ask for a critical review of the story, and then ask it to rewrite the text in light of the review. Here is an example I put together based on the story of a chair (my lack of creativity at this point is scary).
2. OpenAI filters: There are several topics that you will not have an answer to from the bot; there are filters on “topics that are discriminatory, offensive, or inappropriate. This includes questions that are racist, sexist, homophobic, transphobic, or otherwise discriminatory or hateful.” Asking it to engage in illegal activities is also a no-no… Still, for example, if you want advice on improving a contract, you will have the same filters going up. To bypass some of these filters you can create hypothetical situations that make the answer somehow acceptable for the bot to answer.
And that’s about it for my tinkering with the tool so far.
As we move into the future, we’ll start to see tools like ChatGPT emerge and integrate themselves into our everyday activities. From searching for a recipe to finding the best price for something, ChatGPT will make our lives easier and more efficient. But what truly sets sets this tool apart is the impact it will have on the research and creative industries. By allowing us to analyse, cross-reference, and research information more effectively, ChatGPT will revolutionise the way we think and create. So, whether you’re a researcher or a creative, keep an eye out for these new tools , as they’re sure to be making a splash in your field very soon.
ANNEX
Another interesting topic, it’s essential to understand what are the implications of using these tools; as they get closer and closer to replicating human creations, it will be more important to be clear and transparent on where we are using these tools so people can discern and cultivate themselves, understanding the intrinsic difference between something created by a mathematical algorithm and something created by a human.
If you are more interested, here is a series of screenshots I’ve kept of top interactions with the tool:
And then there is this whole space of AI hallucination that I love; here you have a fantastic article about it.
