“Hey, that’s my data”

Welcome to another edition of what we’re determined to make the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today: 

• "Content without consent" heats up

• Ads shouldn't be hard, and Omneky can help

• Check out this insane demo on AI agents within AR glasses

Let’s get to it! 👇

TODAY'S PERSPECTIVE

"Hey, that’s my data!"

Data ownership is already a hot topic in the ChatGPT discourse, and it's definitely getting hotter.

And "content without consent" is the next big alarm bell.

Yesterday, the Washington Post released a deep dive on a Google web crawl dataset that was known to be used in training Google's T5, Meta's LLaMA and likely many more. In it they found hundreds of shocking examples: most news websites, personal blogs including Medium, creator platforms including Patreon and Kickstarter and more. 

All used to train large language models without consent.

We're already seeing platforms react (note: even before the WP article).

Reddit, whose forums data has been cited as a training data source for both OpenAI’s ChatGPT and Google’s Bard, recently announced a plan to charge for access to its data.

This sounds interestingly familiar to Twitter’s big API changes a few months back which took the platform from being a free research resource for many, to a paid and walled garden… smart move Elon.

And as platforms tighten up their controls on data access, it shouldn’t be a surprise that governments, companies and individuals will as well.

We're already seeing this in government:

Checking back in on Italy’s ChatGPT ban - they have now given OpenAI a to-do list of data privacy-related items including publishing a notice describing its training data, and tightening up their legal justification to use our data to train future models.

While our foundation model providers like OpenAI and Google may face hot water - this is yet another reason to pay attention to the data privacy and ownership terms of the tools you or your company uses. And if you're implementing AI, expect that as competition and possibly controversy heats up - you'll want to be able to easily switch between model providers.

USE CASE FROM OUR PARTNERS

Use AI to make better ads

As former marketers ourselves - we know the pain of creating, testing and iterating on a huge amount of creatives just to see what sticks. Not to mention - understanding the insights and using that to inform your next campaign. That’s why when we first heard of Omneky last year, we were excited.

AI should not only make it easier to understand ad performance but also iterate and create new campaigns with ease. That’s exactly what Omneky delivers on — help with generating and personalizing ads, across every platform - at scale.

On average, people are seeing a 2.7x boost in ad performance. We’ll be hosting a workshop with them soon - but in the meantime, check them out to see for yourself!

LINKS

For your reading list 📚

Check out...

• AI agents + AR glasses = recognizing faces, understanding what you're looking at, and responding via text-to-speech; and this demo is pretty insane.

• This next-gen AI code school that helps people learn to build with the latest & greatest with AI

And if you're really nerdy...

• Follow ArXivGPT on Twitter for daily summaries by GPT-4 of new AI papers published daily to arXiv.org

• Compare gpt-4 with Claude and Google Flan-T5 all directly within Vercel AI Playground

Premium subscriber links (more info on joining here!)

• If you missed it, check out the trending playbook chapter the State of AI Agents: your guide to AutoGPT, BabyAGI, and other advancements in AI agents!

That's all!

We'll see you again on Tuesday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected]. 

Cheers,

🪄 The AI Exchange Team