The Rollercoaster of OpenAI’s last month 🎢 🍿
And my own ups and down as I get back to work
🎁 A Wanderly Gift Card promo: Use RUNNING for 25% off all story gift cards.
In this post:
🎢 The last month of OpenAI news… in emojis
🎉 Wanderly Wins
ℹ️ An action packed info diet
First, thank you so much to everyone for your support on my last post. It was personal and vulnerable, and I was touched by all the support as well as others sharing their own stories. And if I haven’t responded, apologies. There’s been a lot going on. 🙏🙏🙏
I’ve wanted to write a newsletter for the last several weeks, but things didn’t feel steady enough to match my writing speed. For those of you who aren’t following the tech news closely: OpenAI’s Dev Day was a blockbuster event announcing a TON of new changes, then less than two weeks later, Sam Altman, the CEO of OpenAI, was ousted by OpenAI’s non-profit board. In probably the strangest week of corporate Game of Thrones, Sam was back within a week.
As an entrepreneur leveraging OpenAI’s tech stack, the last few weeks have been… strange. Here’s roughly how it felt, in emojis:
🥳 Nov 6 - Dev Day - Wow! So many toys! And the API I use will be faster and cheaper!
🤯 Nov 8 - Whoa! This Assistants API is amazing! I should use it to rewrite my whole backend.
😰 Nov 9 - Oh shit. This Custom GPTs thing is gonna eat Wanderly’s lunch.
😮💨 Nov 10 - Never mind, Custom GPTs can’t support what I’m doing yet.
🤔 Nov 11 - Am I missing an opportunity to leverage Custom GPTs as a marketing channel?
👩💻 - Nov 12 - 16 - Maybe all of this distraction for now; I’ll wait for things to mature a bit since Assistant API and Custom GPTs are missing features I need. Keep building.
🤯 - Nov 17 - Holy shit. The OpenAI board just fired Sam Altman!
🍿😰 - Nov 18-21 - What is going on?? Why did they fire Sam?1 If all of OpenAI resigns, does that set back the industry for at least 6-12 months as everything gets rebuilt?
Late Nov 21 - Sam’s back. Keep calm and carry on.
For the curious, here’s my take on the big drops from Open AI’s Dev Day:
Assistants API: A still nascent API with lots of potential. The idea is that it will handle a lot of conversational threading on behalf of developers + automatic retrieval (you can upload a lot of context, and it will handle embeddings for you). It's very cool / a big engineering timesaver. However, the retrieval tech is still early, and the API doesn’t support streaming yet. So I’m gonna give it some time to bake.
Custom GPTs: I’m still pretty mixed and unsure on this one. Any ChatGPT Pro user can create their own custom agent with their own rules. Eventually, there will be a CustomGPT marketplace, which is now launching in January. There’s a lot to unpack here:
It’s a new feature, marketplace, and social advertising lever all at the same time. It’s a great way to get ChatGPT Pro users to 1) use ChatGPT more, 2) become creators and then advertise to their network, and 3) create another value add to being a ChatGPT Pro user.
Someone on Twitter had an amusing observation that maybe it’s just a ploy to get around copyright / get more training data with all the PDFs people will upload to power their CustomGPTs. 🤔
With an 8K character prompt limit, it restricts what’s possible. I tried replicating Wanderly and couldn’t get it to the quality bar that I think reflects Wanderly. But it is forcing my hand on how to handle Wanderly images. I’m unsure if I should make a CustomGPT to advertise Wanderly. Time will tell. Let me know if you have any opinions! I’d love to hear them.
All this is to say… it’s been a long month since my last update. And on top of all of this, I’ve been dealing with the aftermath of my father’s passing. Since the memorial, it’s been its own rollercoaster: taking art off the walls, finding evidence of little and big ways he loved me, a ton of paperwork, finding some less fun things, and then just having to laugh about it all.
So, I’m going to take a moment to celebrate some of the things I’ve been able to learn and make progress on with Wanderly. Gotta count the wins. 🙂
I rewrote my Wanderly storytelling backend to leverage that GPT-4 is faster and to prepare for some new updates. I think the stories are even better.
I started experimenting with DALLE-3 API for Wanderly images; it’s more steerable than any AI image generator and just might be safe enough to deploy live. I’ll probably do a long-form post on that in early 2024.
I improved my onboarding flow after discovering I was setting the wrong default reading level during setup.
I also hired a React Native dev via Toptal to start porting the app to iOS and Android. I think that’ll unlock some new opportunities and is a top feature request. I’m hoping for a v1 by the end of the year.
I launched Wanderly Gift Cards (and you should buy a gift card for loved ones this year! It’s $10 and $30 for 10 and 30 stories, respectively, and with the RUNNING coupon code, you’ll get 25% off). Although my revenue is small, in November, it was 50% more than in October due to gift card revenue.
I got cold-contacted by a major VC firm, which is fun. Note: I’m not currently raising funding, but I’m contemplating it for 2024.
Thanks again for all your support. I’m looking forward to getting back into building and writing in the coming months and hoping the progress on Wanderly accelerates. 🙂
I was mostly glued to Twitter for the weekend of OpenAI drama, but in addition to that news, here are some fun things that crossed my path:
Biden announced an executive order on the development of AI models and products. (A quick summary and a good Hard Fork discussion). Surprisingly bullish on AI in Education, but introducing regulation for large model creators.
Anthropic launches Claude 2.1 - 200K token context window, improvements on hallucination, and pricing changes. I saw this excellent analysis on the performance of the 200K context window (spoiler: not all context tokens are created equally). But then Anthropic saw the analysis and published some prompting tips.
Sundar/Google tweeted a video of Gemini, a cool video recognition + chat experience that was impressive but failed to demonstrate a practical application.
General AI stuff (notice how most of these are about ChatGPT? It’s just the center of gravity right now).
Some AI-themed fun 🎊
A single creator made a 3-minute movie with AI, and it’s much better than the majority of the AI movies I’ve seen. It does a really interesting job of storytelling and working within the constraints of the current tech.
It's still not entirely clear why the board fired Sam. Some people said it was a disagreement about AI safety. There's been a leak about Q*, which people are debating is the code name for AGI vs. just a way for ChatGPT to do simple math. We'll probably just have to wait and see for leaks or announcements in the coming months.