Justin's AI Newsletter
The most interesting content, news, and insights in AI each week.
Hey friends—happy Sunday. This week, I caught a nasty cold. Is there anything that humbles you more than being sick? It's basically nature's way of reminding you that everything can change fast. Always a good perspective reset towards gratitude for health.
But, let’s get into the AI….
🧠 The Next Frontier: The Promise of a ChatGPT with an Extended Memory
Picture a ChatGPT that remembers every conversation you've ever had with it, complete with in-depth knowledge about your life, work, preferences, and more. One that provides personalized responses tailored specifically to you, each and every time.
Recently, researchers published a groundbreaking study detailing their efforts to develop an AI model capable of handling 1-2M+ tokens. This is the kind of research that could once again be a fundamental disruption in the world of AI.
To put things in perspective, ChatGPT's current memory is limited. When utilizing GPT-4, it has a context window of around 8,000 tokens, which corresponds to approximately 12 pages of text. This means the AI's capacity to remember your past exchanges is bound by this limitation.
Now, envision a ChatGPT with a context window encompassing 1 to 2 million tokens. Such a model could have the capacity to comprehend entire codebases, absorb your full business knowledge base, or even possess an intimate understanding of you on a personal basis. ChatGPT could potentially equal your closest friend in terms of understanding you, or possess as much insight into your company as the CEO.
The key to the success of AI lies in the quality of data and inputs you’re able to feed it. The more context we provide, the better its performance. If a ChatGPT response is off, it's rarely due to a deficiency in reasoning ability, but rather the limitations of the given context and the difficulty in identifying the optimal solution.
This area of AI is moving fast, and this research paper offers a glimpse of what the future might hold. As the context window expands, the output generated by AI models will become increasingly powerful. This is definitely a development to keep a close eye on.
🤝 Advice on Raising Money for a Startup
Right now, the AI market is booming. Thousands of startups are popping up everywhere, and investors are putting millions of dollars into these companies.
Earlier this year, my co-founder and I raised capital for our own AI startup. We talked with many investors before choosing the ones we believed would really help us on our journey. In fact, through our investor, we have a partnership directly with Microsoft and OpenAI (more on that here).
If you’re thinking of raising money for a startup, particularly in AI, here’s my number one piece of advice. Ask yourself this question:
“Do you truly and honestly believe you’re a good investment?”
This was the advice given to me, and as I thought more about it, I realized just how important it was.
Investor meetings shouldn’t come across as sales pitches. Be genuine, share your honest thoughts and feelings, and allow investors to see your vision sincerely.
So, if you’ve got an idea for an AI startup and want to raise money, go for it. Just put yourself in their shoes, would you give yourself the money?
💡 This Week's Recommendations
Apple is reportedly developing an AI-powered health coaching service: I am closely watching this, not because I think an AI-powered health app is big news, but because Apple has been awfully quiet through all this AI excitement. As the biggest tech company in the world, Apple has the most popular "AI" assistant currently available in Siri. There is no doubt that Apple is sitting on some powerful technology, but we have yet to see any of it.
Dropbox is laying off 500 people and pivoting to AI: This is the first tech layoff I’ve seen that directly attributes it to the rise of AI. Dropbox is making a significant pivot into the world of AI and cutting its workforce. This suggests that not only are they developing AI tools, but they also need fewer people to accomplish their goals.
Mini GPT-4 Multi-Modal Demo: GPT-4 is multi-modal, which means that it can receive both text and images as input. This team created an open-source version that you can play around with. Although GPT-4 multi-modal hasn't been officially released yet, this demonstrates the power of what's to come.
The UIs ChatGPT won't replace: For all the product and tech people reading this newsletter, check out this blog post. I spend a lot of my time thinking about what UI/UX changes will come about from AI. However, on the flip side, there's a lot that will remain the same. Here's a breakdown of UIs that are already good enough; AI won't touch them.
Mark Zuckerberg says Meta wants to ‘introduce AI agents to billions of people’: During Meta's earnings call, Zuckerberg stated that the company plans to introduce AI agents to billions of people. It seems that Meta is now fully committed to AI, with AI chat features in WhatsApp and Messenger, visual creation tools for posts in Facebook and Instagram, AI ads, and more.
Thanks for tuning in,