Last week, the Twitter account Everything Out of Context posted a screenshot showing Google’s incorrect response to the search query “country in africa that starts with k”:
The tweet went viral, garnering over 133k likes, 6k retweets, and 1k replies.
Someone eventually tagged me in the post because it turns out Google’s Quick Answer is based on a ChatGPT hallucination example that was posted on Emergent Mind (then called LearnGPT) earlier this year to help people understand that ChatGPT responses should not be blindly trusted:
Even though I pivoted the site away from ChatGPT examples, those examples are still around, and apparently being shown in Quick Answers on Google when Google deems it relevant.
Someone also joined Emergent Mind’s Discord suggesting I take the page down:
[Update] On Oct 26, Christopher Ingraham, a journalist with a large following, shared it as well, resulting in a lot of attention. Here’s a newer blog post about the fallout from his tweet.
I’m not going to take down the page down because it seems like Google’s responsibility to figure out how to identify and exclude false information from Quick Answers. Also, as of this writing, I’m only seeing it as a Quick Answer some percentage of the time. I hope this obviously incorrect example raises awareness about inaccuracies in Google Search result snippets, and contributes to Google addressing them in time.
It is funny though: earlier this year when I was working heavily on Emergent Mind, I tried pretty hard to write a viral tweet about it. I never succeeded, but here we are now, months later, finally with it going viral, just not for a reason I imagined 🤣.
For the past few months I’ve been heads down building an AI-powered timeline generator tool for Preceden, my SaaS timeline maker software:
The tool – which is free to use and available on Preceden’s homepage – lets you type in a topic or detailed description of a timeline and it will generate a beautiful visualization for you in less than a minute.
Most of you reading this won’t have ever used timeline software, so let me take a step back and give you some more context.
The old way
The majority of people seeking out timeline software are using it for 1 of 4 use cases:
Project planning – for example, a project manager looking to visualize a complex project plan
Historical timelines – for example, a student tasked with creating a timeline for a class project
Legal timelines – for example, a lawyer creating a timeline about the history of a case
Fiction writing – for example, an author using it to hold together their fictional worlds
There’s no shortage of timeline software out there for these people to choose from – at least 28 by my count (and that’s just ones that call themselves timeline software, not counting the thousand other planning tools that have timeline features) – and they all basically work the same way:
Sign up for an account
Manually enter details about each event (its name, dates, etc)
The tool will render those events in a timeline visualization
Preceden – which has been around since 2010 – has always worked this way too:
Timeline software usually requires you to manually populate your timeline with events
The new way: ChatGPT to the rescue
The launch of the ChatGPT API earlier this year presented an opportunity to rethink this old, manual approach to creating timelines.
After all, ChatGPT was trained about 570 gigabytes of data sourced from books, Wikipedia, research articles, and much more, so it is aware of many of the topics that people may want to create a timeline about:
ChatGPT knows a lot about well-known historical events
At a high level, this new AI-powered timeline maker is taking the topic you provide it, asking ChatGPT for information about it, and then automatically generating a timeline using that data:
Getting this to work well for for all of the use cases I mentioned earlier, in any language, for hundreds of different date formats, with a seamless and beautifully-designed UI (hat-tip WebPraktikos, Preceden’s designer, for his work there), and without glaring issues (like hallucinating historical events that never actually happened), has been quite a project 🤣.
I soft-launched a v1 of this tool at the end of May and have since viewed thousands of timelines that people have created with it, each time looking for opportunities to improve the quality, and adjusting the prompts and code accordingly so it’s a bit better for the next person to use it.
There are still quite a few issues, but it works pretty well for 90%+ of searches these days.
In addition to figuring out the key events to include in the timeline, Preceden also organizes those events into layers and determines meaningfuliconsto display beside each event.
Note that this timeline includes accurate events from 2023, well past ChatGPT’s September 2021 knowledge cutoff date. Figuring out that piece was… fun 😉.
By default, Preceden will detect the language from the topic and generate the timeline in that language, though users can choose a specific language from the language settings:
Each time I see someone using a new language, I add support for it so the tool works seamlessly for future users. This also requires adding support for each language’s diverse date formats 😱:
For project managers, this tool can create high level plans for a wide variety of projects given just a simple description of the project. This will usually need to be edited to tailor it to the specifics of the project (like removing some of these AI-generated events, adjusting their dates, and adding your own events), but it can save users a lot of time vs creating the project plan from scratch.
Note too the current date line in red, which makes it easy to get a sense of when the project kicks off relative to today’s date.
To my surprise, many users also type or paste in long, complex project descriptions, hoping to generate a timeline that reflects all of the details in their description. For example, the timeline above was generated using the following description that begins with:
Product launch roadmap: Research (Jul – Oct 2023) – Study the European market and consumer behavior. Understand the product thoroughly, its USP, and potential market demand. Identify competition, market trends, and potential roadblocks. Define business objectives and marketing strategies. Regulatory Compliance (Aug – Oct 2023) – Ensure the product meets European product safety standards...
Once again, this saves project managers a ton of time compared to building the timeline from scratch.
ChatGPT won’t know the specifics about most personal legal matters, but the timelines it generates can serve as a useful template to get started with, and just in general helps show people evaluating Preceden some of its capabilities.
Writers can use this tool brainstorm stories and help them hold their fictional worlds together (ie, ensure the characters and plot points are consistent and flow well over the course of the story).
Even if you’re not a writer, it’s fun throwing random topics into the tool and seeing what it comes up with.
What’s next
While the tool is pretty good for most searches at this point, people type in all sorts of crazy things, some of which it doesn’t handle well right now. URLs are just one example (imagine if you could paste in a URL, and Preceden would scrape the content of that page and generate a timeline about it).
I’m also excited about the possibility of building a deeper ChatGPT integration into the app itself. For one example, imagine if you typed in a project description in Japanese, then the entire homepage (and in-app experience?) would automatically update to speak to that intent, and it all be in Japanese instead of English. Lots of opportunities along these lines.
If you wind up checking out the tool, I’d love any feedback you have: matt@preceden.com.
I currently have two products: Preceden, a SaaS timeline maker, and Emergent Mind, an AI news site and newsletter.
Emergent Mind began last December as LearnGPT, a ChatGPT examples site, and I later renamed it to Emergent Mind and transitioned it the news site that it is today:
I’m pretty happy with how it turned out, but after a lot of reflection, I’ve decided I just want to focus on growing Preceden and not try to split my time with a second product, even though I still think Emergent Mind has a ton of potential.
I put together a sale page with some details about Emergent Mind including its tech stack, costs, and potential opportunities.
If anyone is interested taking it over, drop me a note: matt@emergentmind.com.
At the beginning of the year I quit consulting to focus full time on Preceden, my SaaS timeline maker tool. I also started working on a new side project, Emergent Mind, an AI-powered AI news site. My last update on how things were going was after 3 months which provides more background for anyone interested.
Emergent Mind
When I last shared an update in early April, here’s what Emergent Mind looked like:
And here it is today:
Same general idea as before, but with a number of significant changes:
Did away with the whole “Explain it Like I’m 5”, “Explain it as a Poem” and other gimmicky explanation styles and replaced it with translated versions of the site. Visitors can now read Emergent Mind in 12 languages.
Simplified the UI a ton: did away with upvoting and post details (score, age, poster, etc). Also nixed individual post pages and just moved the main content (summaries) to the homepage.
Also added those post thumbnails to add some graphics to the site.
Launched its newsletter, which is a recap of yesterdays or last week’s top posts, depending on whether the person subscribed daily or weekly.
Wired up the Emergent Mind Twitter account to automatically share news articles (in the gimmicky explanation styles that I nixed on the actual website).
Set up a new LLC to separate Emergent Mind from the LLC I use for Preceden
Overall, the site is coming along, but IMO still lacking in a lot of areas. Doesn’t feel like it has product market fit yet. Notably, while the summaries are good, it lacks insights that all of its human-powered AI news competitors provide. I do think it’s possible to have Emergent Mind generate insightful reporting that’s as good or better than humans, but it’s going to take some work to get there.
I’ve had a number of people suggest to me to expand into other topics besides AI. Imagine there being sections on politics, tech, etc etc etc, and you could choose which ones you want surfaced to you on the homepage and in the newsletter. I’ve also had people suggest I say focused on AI because it’s a plenty big niche. For now, I’m staying focused on AI, but I could see it changing in the future, especially if I can get it to generate insightful analyses and not just summaries.
I spent most of April on Emergent Mind, but for the last few weeks have been focused almost entirely on Preceden. Thankfully, Emergent Mind is 99% automated at this point including the news on the site, the newsletter, and the Twitter account, so I can step away and it mostly runs on its own.
Preceden aka Timeline Maker AI
About two weeks ago I quietly launched a major update to Preceden: the ability to generate complete timelines on any topic using AI, directly from the homepage, no account required:
I had been working on this for months in a PR, and finally said screw it, I need to get this live, even if it is buggy around the edges. And buggy it was (and still is to some degree). I’ve spent the last 2 weeks fixing issues, improving the UI, and trying to get the quality of the generated timelines as good as possible before promoting it more widely.
Consider that Preceden is a general purpose timeline maker tool which means that people are using it to generate timelines ranging from World War II to personal injury claims to startup marketing plans and everything in between. And it has to ideally work in every language and support every possible date format. And support various guidance people give it around what time period to use, or durations for their projects, and on and on and on. And not hallucinate historical events that didn’t actually happen 🤣.
Here are the example topics to give you an idea of what people are using Preceden for, and what this tool has to handle:
I’m basically spending my days monitoring what people are searching for, inspecting the timelines Preceden generates for them, and then figuring out ways to improve those timelines so that Preceden generates higher quality timelines for future users.
Needless to say, it’s complicated to do this well. But, it’s coming along, and it’s already getting a ton of usage and driving a meaningful amount of revenue.
This tool is, IMHO, the best timeline maker tool ever built and I’m very optimistic about its future.
Balancing work and life
This will come to no surprise to anyone who knows me well, but I’m a bit of a workaholic.
I love building these products, and will happily push through when I’m sick or feeling burnt out or should be taking it slower. I also don’t allow enough time for other things in my life.
Learn to cook some new meals? Nah, too busy.
Spend a few hours exploring new LLMs? Maybe another day.
Pick up a new hobby? Nah, need to jump into the support queue.
Play video games? Gotta fix that edge case bug first.
Contribute to the Microconf Slack? Need to watch a few session replays first.
Take the day off to sit by the pool? Maybe after I ship that next feature.
Take a week off to rest? lol.
You get the idea.
The thing is, I have the freedom to do these things but tend not to prioritize them. Like I mentioned in my last post, I run the real risk of burning out if I don’t build better habits that are sustainable long term. I’m trying to get more comfortable moving at a more relaxed pace, but for better or worse, it doesn’t come easily to me. I’m working on it.
That’s about it for now. Hope you’re all doing well and thanks for following along 👋.
Matt Mazur 