Justin's AI Newsletter

The most interesting content, news, and insights in AI each week.

--

Hey friends 👋,

Greetings from Sunriver, Oregon!

My team and I will be working out here for the next couple weeks. Through our investor Neo, we're here co-working and collaborating with other startups in the mountains.

Working alongside other brilliant startup founders is unbeatable - everyone is attacking their own unique challenges, allowing us to share ideas and collaborate. This atmosphere really gets the creative juices flowing.

---

AI INDUSTRY

🖼️ Multimodal Models are Here.

This week, both Google’s Bard and Microsoft’s Bing Chat released multimodal functionality. This means that, in addition to text, you can input images into the AI model, opening up a world of possibilities.

Here are some of my favorite examples:

• Calorie Counter: Upload a picture of your meal and the model will estimate how many calories it contains.
• Landing Page Critique: Provide a screenshot of your landing page and the model will suggest ideas to increase your conversion rate.
• Recipe Generation: Give the model a picture of a meal you want to make and it will generate a custom recipe for you.
• Design-to-Code: Provide screenshots of a design file and the model will write the front-end HTML and CSS code for your ideal design.

I made a video demonstrating some of these use cases here.

While the functionality is far from perfect at this point, it provides a glimpse into the power of multimodal AI models. Essentially, we are giving AI models eyeballs, a way to visually interpret the information we provide.

This is only the beginning. These models will become more intelligent, as well as incorporate more modalities such as audio and video.

I highly recommend that you play around with them. And, as always, let me know if you discover any interesting use cases!

---

AI INDUSTRY

📉 ChatGPT Getting Worse?

For months, there has been an ongoing debate about the declining quality of OpenAI models.

This week, an interesting study was released that actually supports this claim. A research team gave 500 problems that GPT-4 was able to solve in March, and the score of those tasks dropped from 98% to 2.4%.

The reason for this decline is not entirely clear.

One side of the argument suggests that what is happening is simply "behavior drift." This means that as models change over time, the way they respond to certain prompts also changes. Prompts that used to be effective may no longer work.

Other arguments propose that the issue could be model degradation, limitations in GPU, or fine-tuning.

Whatever the cause, this trend highlights the risks of relying on a single model. Over time, these models will inevitably change. If you depend too heavily on one prompt or one model, you put yourself at risk.

---

💡 This Week's Recommendations

🔗 Apple is testing a ChatGPT-like AI chatbot: Apple is reportedly developing a ChatGPT-like AI chatbot, internally referred to as "Apple GPT," to compete with OpenAI and Google. I'm very excited to see what Apple does in this space. An LLM-powered Siri, at the power of GPT-4, would be game-changing.

🔗 Google is testing an AI tool that can write news articles: Google is testing an AI tool called "Genesis" that can generate news articles. The tool is being pitched to publications like The New York Times and The Washington Post as a way to automate certain tasks for journalists. However, there are concerns about the potential spread of misinformation if AI-generated articles are not fact-checked or thoroughly edited.

🔗 The $1 billion gamble to ensure AI doesn’t destroy humanity: An interesting article on Anthropic and their approach to AI safety. Literally every single one of the company’s seven co-founders was previously employed at OpenAI. They're taking a different approach to AI development...

---

Thanks for tuning in,

Justin

​Justin's Full Content Library | Previous Newsletters​

113 Cherry St #92768, Seattle, WA 98104-2205

If you'd like to update your email settings, choose one of the options below.

1) Unsubscribe from my weekly newsletter.

2) Unsubscribe to be removed from all future mailings, including occasional and highly important AI updates. That'll make me sad. But hey... I get it. You're busy. Just know that once you click this link you won't receive any more emails from me.

Other Preferences​