The EU Wants AI Content Labels

Plus the Influencer Chatbot Goes Rogue!

Author

Nathaniel Whittemore
June 06, 2023

The AI Breakdown First Five - Tuesday June 6, 2023

Today on the First Five:

  • 5. Apple Doesn’t Say “AI” at WWDC

  • 4. The $1-Per-Minute Influencer Chatbot Goes Rogue

  • 3. Meta’s New “Hiera” Model Processes Visual Info Faster

  • 2. ChatGPT Alternative HuggingChat Adds Web Browsing

  • 1. EU Wants AI Content Labels

5. Apple Ignores-Ish AI

Apple has been conspicuously absent from the generative AI wars, so some wondered if we’d get any AI related announcements at yesterday’s WWDC event. The answer was sorta-not-really. Autocorrect and Dictation are now “transformer” powered, but there was no LLM to be seen. Still, many think that the new VisionPro spatial computing experience will inevitably combine with AI in some way.

4. Influencer Chatbots Gone Wild!

A few weeks ago, Caryn Marjorie broke the internet when she released a $1-per-minute chatbot to her millions of fans. During her beta test she was making more than $70,000 a week and estimated that it could bring in $5m a month. Recently, however, the bot has gone rogue, initiating certain…ahem…intimate conversations it wasn’t trained to engage in.

3. Meta Introduces New “Hiera Model”

Adding more and more components to transformer-based visual models has slowed them down. At least, that was the thesis heading into Meta’s new research on Heira, a model that strips out that bloat and uses a visual pretext task called Masked AutoEncoder (MAE) in pretraining. The more efficient model maintains high accuracy and has potential application where rapid image or video recognition are important (like self-driving cars).

2. Web Browsing with HuggingChat ChatGPT Alternative

HuggingChat is an entirely open source alternative to ChatGPT from the folks at HuggingFace. Believers in the importance of open source were thrilled when HuggingChat was announced, and given recent developments, the new browse features should allow it to keep some feature parity with closed source alternatives.

1. EU Wants AI Content Labels

In a recent wide-scale study, 32% of internet users couldn’t tell the difference between a human and an AI bot. This speaks to concerns that some regulars have that AI could be used for disinformation and manipulation purposes. While the EU readies their AI Act, the European Commission wants big platforms like Meta and Google to proactively label AI-generated content.

The Latest AI Breakdown Episodes

Thanks for reading! Let us know how today’s edition hit:

How was today's newsletter?

Login or Subscribe to participate in polls.

Signing off from the future’s past - NLW