AI is about to go mainstream 🦾💡
AI is about to go mainstream 🦾💡
Prepare to have your brains smashed out with a slice of lemon wrapped round a large gold brick
In the Hitchhiker's Guide to the Galaxy, Douglas Adams describes the effect of the Pan Galactic Gargle Blaster, said to be the best drink in existence, as ‘like having your brains smashed out with a slice of lemon wrapped round a large gold brick.’
We are on the cusp of a similar experience with artificial intelligence (AI). Only in recent months, have truly practical, rather than hypothetical, applications really started to emerge. And they’re yielding results that are provoking genuine astonishment from end-users. The kind of astonishment that, I imagine, the Pan Galatic Gargle Blaster arouses in its imbibers.
Thursday 1st December saw an explosion of Twitter posts about OpenAI’s newly publicly released ChatGPT technology. Given a prompt, ChatGPT will reply with an answer that fits the statement or question it receives. For example, I asked it to respond to this prompt: ‘Explain to me what ChatGPT does and how it might be used.’ The response is as follows:
ChatGPT is a type of natural language processing (NLP) technology that uses a large language model trained by GPT-3 (Generative Pretrained Transformer 3) to generate human-like responses to user input. It is designed to enable developers to create conversational systems, such as chatbots, that can hold realistic, intelligent conversations with users.
One possible use for ChatGPT is to create chatbots for customer service or support. These chatbots could be integrated into a company's website or mobile app, allowing customers to quickly get answers to common questions or issues without needing to wait for a human customer service representative to become available. This could help improve the customer experience and reduce response times.
Another potential use for ChatGPT is in creating virtual assistants for personal or business use. These assistants could be trained to handle a wide range of tasks, from scheduling appointments and managing emails to providing information and recommendations on a variety of topics. This could help people be more productive and efficient in their daily lives.
Overall, ChatGPT offers a powerful tool for developers looking to create advanced conversational systems that can understand and respond to natural human language.
This text is functionally indistinguishable from text written by a human, which is remarkable in of itself. But the Thursday 1st posts were far more creative than the use cases suggested by ChatGPT, ranging from users giving it prompts with the intention of creating absurdist humour:
To successful attempts to break ChatGPT’s content and ethics filter:
To poetry…
But the initial results that really impressed me go far beyond exploring content guiderails, explaining serious concepts in pirate speak, or converting passages to read like a 1940s, wise-guy gangster. What really matters for mass user adoption is use cases that make a difference to their everyday lives. And by far the most impactful early results seem to relate to explaining and solving coding problems.
For example, the Twitter user below used ChatGPT to generate a differential equation instruction for LaTeX, a system used for document preparation where the user specifies the formatting using a programming language, instead of a graphical user interface, like you might with Google Docs, or Microsoft Word.
The Google output now feels simplistic, and is based on human-generated content from online fora. The ChatGPT response was created in a comparable search time, but features both logically exact code to implement the specific solution required by the specific problem as described, and natural language explanations of why this logically exact solution is being used. It is not only doing the work, it is explaining it in human terms so that the human might learn why the solution is suitable.
Furthermore, a Stack Overflow post explaining how to solve someone else’s tangentially related problem may require further thinking and adaptation, or not generate a reply from other users if you ask for help or clarification. Precise solutions to precisely described problems, plus further instant further questioning, is already available to the end user of ChatGPT.
I have experienced this first hand when trying to work out how to generate a tricky regular expression. A regular expression is a sequence of characters used in various programming languages that specifies a search pattern, like searching only for email addresses in a large body of varied text, or searching for content that follows a sequence of punctuation, like ‘. (‘. While I use them occasionally, I don’t use them often and I haven’t ever had much of an incentive to comprehensively learn how they work, so debugging tends to be a painful trial and error process once every five to six months.
With ChatGPT, research that took me 30 minutes on the 30th of November, took me less than a minute one day later – the solution generated with ChatGPT was also more concise, and syntactically neater. That’s instant value capture for me, and I will be using ChatGPT going forward whenever I have a coding problem.
Another interesting use case for AI which is making waves on social media is art generation. Again, I’ve already found a real-life, valuable use case for this technology. I would otherwise have commissioned someone to create a logo for this newsletter, but with my free sign-up credits for DALL-E 2, OpenAI’s natural language-interpreting AI art generator, I have avoided that expense with the following AI-generated image, which I’m genuinely delighted with.
It took 15 prompts to generate one that I really liked, amending the prompt each time to give a more specific or experimentally varied instructions. You really do have to be quite specific - early prompts generated odd creations, with upside down lightbulbs, or arms that were themselves light bulbs. But, adapt the prompt, adjusting slightly each time to correct these errors, and you quickly arrive at what you’re looking for. Creating this would have taken hours for a digital artist.
Substacker Noahpinion believes AI will give humans ‘superpowers’, where they are able to use AI to ‘autocomplete’ more than just text – even using it to do a large amount of the primary work in existing processes.
So that’s our prediction for the near-term future of generative AI – not something that replaces humans, but something that gives them superpowers. A proverbial bicycle for the mind. Adjusting to those new superpowers will be a long, difficult trial-and-error process for both workers and companies, but as with the advent of machine tools and robots and word processors, we suspect that the final outcome will be better for most human workers than what currently exists.
I agree to an extent. But what about that logo design example? I’ve switched straight to it from commissioning a human. That’s value to them destroyed straight away. And the free image is functionally mine to do with as I please, including monetisation, given how convoluted and underdeveloped copyright and ownership law is in this space, leave aside OpenAI’s relaxed terms of use.
In any case, you can read more of his thoughts on this matter in this post:
So, there’s huge opportunity, but problems remain. One user found an example of compound probability being calculated incorrectly, offering a 19% probability instead of the correct 10% probability for the problem, despite all of the logical steps being correct until the final solution was given.
A good examination of some of these rather ‘dumb’ outcomes is available in this Substack post:
Gartner, a firm famous for its hypothesised ‘hype cycle’ for emerging technologies, reckons generative AI is close to a ‘peak of inflated expectations.’ I’m not so sure. Based on what I’ve been experiencing from models that have clearly got some teething problems, but that are already proving genuinely and unexpectedly useful in my life, I think my big expectations are pretty reasonable at the moment. And I can’t wait to see what’s next!
