The AI Transformation: Issue #11

Welcome back to our weekly newsletter where we keep you up-to-date on the AI Wave (see what I did there?) and show you how we’re building the future of Triple Whale around this cutting-edge technology.

The team is working hard towards general accessibility of the next generation of Triple Whale. It shouldn't be long now!

This week's agenda:

🔥 AI News: Move over, Open AI… Safe Superintelligence is here.

📖 What We’re Consuming

🏎️ Under the Hood: Building Moby

🔊Sonar Has Arrived

Let’s get into it.

🔥 AI News: Move over, Open AI. Safe Superintelligence is here.

Ilya Sutskever, one of OpenAI’s Co-founders and former Chief Scientist, announced yesterday the launch of his new AI company: Safe Superintelligence Inc.

Created alongside Daniel Gross and Daniel Levy, the company’s whole purpose is centered around one goal and one product: a safe superintelligence.
From their website: “SSI is our mission, our name, and our entire product roadmap, because it is our sole focus. Our team, investors, and business model are all aligned to achieve SSI.”

Is it just me? Or does this sound very similar to OpenAI’s original mission?

“Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return. Since our research is free from financial obligations, we can better focus on a positive human impact.”

Only insiders truly know why Ilya left OpenAI to start his own company. But I expect many speculations will arise that he was no longer aligned with OpenAI’s approach to building “safe” AI for humanity.

Only time will tell.

“This way, we can scale in peace.” - SSI

📖 What We're Consuming

We have an internal slack channel called #ai-news-pulse. Here’s what our team shared in the channel this past week:

5 wild new AI tools you can try right now: Remember one year ago when that wild AI-generated video of Will Smith eating spaghetti was released, and it was so obviously AI that everyone joked about it? Well.. it’s not so noticeably AI-generated anymore. In this short video, Fireship shows us 5 crazy generative AI tools we can use today to generate videos, images, sound effects, code, and more (and way better than 2023 Will Smith Eating Spaghetti).

Building the Orchestration Layer for AI Agents: While there was initial excitement over AutoGPT and BabyAGI, the buzz eventually died off because real world tasks require more nuanced knowledge and reasoning than Vanilla GPTs could bring. How will AI Agents evolve? In this Training Data Podcast episode, Harrison Chase of LangChain explains what’s changed that’s allowing agents to improve performance and find traction.  

Could AI agentic workflows drive more AI progress than even the next generation of foundation models? When AI models use agentic workflows, we can ask the model to iterate over a document many times. Since this iterative process is critical for most human writers to generate quality text, an iterative workflow for AI will also yield much better results than writing in a single pass. This Tweet from Andrew Ng outlines why he believes agentic workflows are an important trend in AI development, which is further explained in this video.

The Launch of Claude 3.5 Sonnet: Not only is it more intelligent, it’s faster. Claude 3.5 Sonnet sets the new industry benchmarks for graduate-level reasoning (GPQA), undergraduate-level knowledge (MMLU), and coding proficiency (HumanEval). Read more about it here.

New AI Tools on TikTok with Symphony: Businesses of all sizes, creators, and agencies can blend human imagination with AI-powered efficiency to help scale content development, creativity, and productivity on TikTok. Symphony allows brands to use stock avatars (paid, licensed actors) or custom avatars that are crafted to represent a brand with specific likenesses. Sounds cool, right? Read more here.

OpenAI going for-profit? According to a report from The Information, OpenAI could potentially become a for-profit company sometime soon. Apparently this report is based off of a person overhearing Sam Altman discuss plans with shareholders, so we’re not sure how accurate it is. But if OpenAI goes public valued at $86 billion, it’s a great chance to get a slice of that pie…

🏎️ Under the Hood: Building Moby

You’ve heard us talk about Moby. But how is our team working on making Moby generally accessible? Here’s a behind-the-scenes look:

The first thing to know: there's a lot of data involved.

Moby is powered by a flexible system that utilizes multiple language models. Our team is continually testing the latest models and conducting our own internal benchmarks and analyses to determine which perform best for specific use cases.

As mentioned above, Anthropic launched their most recent model yesterday, Claud 3.5 Sonnet.

Our tests showed that Claude 3.5 is slightly faster than GPT-4o (3.0 vs. 3.7 seconds response speed/latency); however, GPT-4o performed more accurately in our internal test suite (0.95 vs 0.9 total score).

Another method our team uses to analyze data is by monitoring how often people interact with Moby and the types of questions they are asking.

Currently, the most popular prompt category is Ad Spend and Performance. Here’s an example of a popular prompt that provides valuable ad performance data:

“Give me my spend, ROAS, NC ROAS, CPA, NC CPA, CTR, CVR, and CPM from Meta Ads over the last 60 days by day in a table”

Here is a breakdown of the most popular prompt categories for this past week:

Lastly, the team is diligently working to improve the quality of the chat output. One way we measure quality is by looking at the % Daily Message Success Rate.

The team uses AI evaluation benchmarks to determine if the message's output was correct, the type of error (if the chat was incorrect), and an explanation of why the output was successful or unsuccessful. This context allows our team to refine the AI evaluation system's accuracy, ensuring ongoing improvements to the quality of Moby. Now that’s meta!

Here is a preview of one of our charts. You can see that the chat success rate has significantly improved in recent months.

🔊 Sonar Has Arrived

Triple Whale’s all-in-one data platform just got even better.

Introducing Sonar 🔊

Now, Triple whale can supercharge your marketing platforms and optimize campaign performance while you sleep.

How does it work?

Sonar uses Triple Whale’s first-party pixel to capture your visitors’ on-site events, enriches them with customer and conversion data from Shopify and Triple Whale, and sends the data back to your marketing platforms via a Conversions API.

What does this mean for you?

📱Your platforms restore previously lost data and accurately identify more potential customers.
📫 Complete data triggers up to +70% more Klaviyo emails and abandonment flows capturing otherwise lost revenue.
📈 Higher data quality = More effective Meta campaign targeting.

The Result?

💸 More revenue and improved ROI. Learn more.

Sonar is now available for all Enterprise customers. If you're not on our Enterprise plan, you can upgrade here.

Set Up Sonar

That’s it for this week! Hope you’ve enjoyed this week’s iteration of the AI Transformation!

Catch you next week!

-Ethan

‍