What's Victor doing in this image? 🤔 You might already have figured it out - but get the full context at the end of this email.
This post is from our e-mail newsletter. Sign up here to receiver regular updates on AI from Applai.
Here's what you need to know for 23 October, 2023 in 2 minutes and 18 seconds.
In this edition, we'll cover:
After reading this, you'll have a grasp of OpenAI's recent achievements, the mechanics behind Large Language Models (LLMs), and the advancements in prompt engineering for ChatGPT. Plus, a sneak peek into our latest activities at Applai. Let’s get the ball rolling:
How Did They Do It? 🚀
As many of you know generative AI and LLMs are on the rise. One of the way to see this is to follow the progress of OpenAI, the company behind ChatGPT. It has remarkably escalated its revenue, clocking in an impressive $1.3 billion, up from a modest $28 million last year. That's more than just a little bump.
More on this story: OpenAI's Financial Ascent
A lot of people have of course tried ChatGPT and noticed how well it is working for many tasks, but if you want to know more about why and how large language models work, we’ve stumbled across a really good article explaining it in simple terms.
Why Are They So Awesome? 🤔
These sophisticated models, like the GPT-3.5/4, have witnessed unparalleled growth in recent years. To put it in perspective, imagine if flight speed magically increased by 1,500x within two short years. Mind-blowing, right? We often discuss LLMs, but we seldom dive into their inner workings—the mechanics behind the magic. Ever wondered how ChatGPT takes in a prompt and crafts a concise, engaging narrative?
Answering that is the goal of this article: LLMs for Dummies
And once you have a good understanding of how LLMs work, you’ll quickly see how well good prompts work to enable e.g. ChatGPT to perform much better. A recent development in the field of “prompt engineering” is not only ask ChatGPT to perform a task - but explicitly ask ChatGPT to think step by step, when solving the task. This gives much better results:
Want Smarter AI Conversations? 🧠
Researchers have found a method to boost ChatGPT's problem-solving precision. By leveraging an "chain-of-thought-prompting" approach, ChatGPT can think progressively, producing a structured reasoning chain. The outcome? An AI that thinks and solves complex problems more efficiently. Try it for yourself, by e.g. asking ChatGPT to "think step by step" for more difficult problems.
Logical question in ChatGPT (GPT-3.5) gives wrong result
Question + Chain of though (asking ChatGPT to think step by step)
Explore the research and learn the new prompting technique here: Chain of Thought Prompting Improves ChatGPT's Reasoning
Or read a good blog post about chain-of-thought-prompting: How to Make ChatGPT Smarter, Josh Wade
That was all the external news from the past week. Here's what we have been up to in Applai.
As always, there's plenty cooking in our Applai kitchen. Oh yes, the image from the beginning of the email? That's from our recent "learning day" at Gl. Brydegaard. Together with the team in Gl. Brydegaard we discussed leveraging ChatGPT and AI in daily hotel, restaurant, and event center operations. One of the tips was how to write structured prompts in ChatGPT as you probably figured out from the image, if you understand Danish. Otherwise, here is ChatGPT's summary of what is going on in the image:
In the image, a man is giving a presentation in a room. He stands next to a screen that displays information in Danish. The content on the screen appears to be about obtaining desired output tailored to specific tasks and roles, with various instructions and details provided. The man seems to be explaining the content, and there's a watermark or logo at the bottom right corner labeled "Applai".
We also explored how ChatGPT and AI can be used e.g. for writing their email newsletters, and to develop a new version of their menu in the restaurant, only by taking a picture of the menu card and utilizing GPT-4V in ChatGPT - the new feature we described in our last newsletter.
This week, we have also played around with GPT-4V in ChatGPT and its visual input capabilities on our own. See for example our LinkedIn post about more or less confusing parking signs here.
That was all from the second edition of AI unhyped.
How did we do, Victor? Reach out to mathias@applai.io if you have any feedback. And if there's any news you believe deserved a spot in this newsletter but didn't get one, we'd love to hear about it!
Help us amplify the reach of AI unhyped. Share the goodness with your buddies. Here's the sign-up link to share the love.