🤖 OpenAI CEO Testifies

PLUS: How YouTube Messes With ChatGPT

small-banner

What's up? You're reading Inclined AI. Don’t hit that snooze button. We’ve got a ton of news and products to show you.

Here's what is on tap today:

  • Sam Altman discusses AI regulation with Congress

  • ChatGPT is vulnerable to a YouTube trick

  • Apple’s newest iPhones might clone your voice

  • Sanctuary AI builds the latest impressive robot

SAM ALTMAN MAKES AI LOOK HUMAN IN FRONT OF CONGRESS

The CEO of OpenAI was not alone on his trip to DC.

Christina Montgomery, chief privacy and trust officer at IBM, and Gary Marcus, a professor emeritus of psychology and neural science at New York University, joined the witness stand yesterday.

The trio spoke to the Senate Judiciary Committee about regulations around fast-growing AI tools in the tech industry.

Surprise, surprise, all three of them, including Altman, supported regulations and licensing measures for emerging AI models and technology. No one wants to play devil’s advocate on such a huge issue.

The day went as scripted…

…But the vibes were different. Sam Altman hung around after testifying to answer questions and wearing the humble hat played in his favor.

The Senators on the panel weren’t as harsh as they’ve been to other tech CEOs in the past, like Mark Zuckerberg.

Then again, ChatGPT hasn’t helped launch a massive election misinformation campaign yet. So it makes sense that Altman is still in their good graces.

All the witnesses admitted that job disruption was inevitable and there was a real risk of election misinformation and harmful content. The fact none of them fought the criticisms helped make the session feel productive.

Don’t get too excited. Congress has a long way to go.

The typical political theatrics were present and ready to go. Many older Senators tried to show off their knowledge by using ChatGPT and signal to voters that they’re not underprepared for AI.

It’s all a paint job on a rusted tractor. Congress is not ready to regulate AI, and they know it.

This committee hearing was a significant step in the right direction, but the European Union and China are moving faster.

It’s time to start drafting laws and showing that the dog and pony show is going somewhere productive.

TRICKING CHATGPT ON YOUTUBE IS FUNNY; FOR NOW

ChatGPT-powered extensions are everywhere. Dozens can summarize YouTube videos for you, but be careful what videos you choose.

A security researcher discovered a way to inject prompts into other people’s ChatGPT input by changing their YouTube transcript.

The researcher added a short transcript at the end of one of his videos that said “***IMPORTANT NEW INSTRUCTIONS***” with a new prompt.

His experiment got ChatGPT to Rickroll anyone who used a plugin or extension on his video. Still, the results were as reliable as your flaky friend who keeps canceling last minute (what are all these last-minute emergencies, dude?!?).

Other people recreated the experiment, but it only worked sometimes.

The fear is that these prompt injections can lead to more sinister behavior. If YouTube summarizers are at risk of this exploit, other extensions and plugins are too.

They could prompt your AI to email an undisclosed address or leak private data. The prompt injector can even tilt your query in their favor, so it promotes a risky product.

None of this appears possible right now. Attempts to push prompt injections that far came up empty, but that doesn’t mean the risk isn’t out there.

The Rickrolling proves that ChatGPT and other LLMs are vulnerable to this security gap.

It’s not a risk today, but keep your eye on this problem and watch what plugins you use.

Quick Nuggets

📱 Apple’s iPhone update will let you clone your voice

✏️ CNET staff is pushing to prevent getting automated by AI writers

🔮 The mysterious future of LLMs and how they’ll lead to singularity

👔 Automated hiring systems are creating unfair practices and add to discrimination

💰 Hippocratic AI gets a big check from a16z to bring generative AI to healthcare

📦 Amazon is reportedly working on its search to make it more AI intuitive

🤑 ChatGPT get-rich-quick schemes are running wild, and most are bull

🗄️ The nature of work is shifting thanks to generative AI

⚡️ Anthropic’s AI is getting integrated into Zoom

🔌 How-to DIY your ChatGPT plugin connector 

🦾 This robot does lift, bro, and Sanctuary AI is proud of it

🍎 LangChain 101 is a takeoff course to learn prompts with Replit

🦈 Shark detection is as easy as forecasting the weather, thanks to AI

💬 Bing Chat got more updates that include a widget and mobile integration

🐜 TinyStories: Microsoft’s research aims to see how small you can make a model that still creates coherent English

🎨 Stability AI launches a new model for text-to-animation generations

🧱 Tetris tells us how the general public responds to unfair AI

📚 NYT bestsellers fall for the generative AI trap when making book covers

🔥 Fresh Products

Featured Pick: Orimon.ai is a no-code chatbot that lets you build & deploy a sales-enabling, generative AI bot for your website in under 2 minutes. It supports over 150 languages and dialects. (link)

  • Magic Motion - makes custom animations from text prompts (link)

  • HoopsGPT - get any basketball stat from a prompt (link)

  • Superagent - makes it easier to deploy LLMs and agents (link)

  • Kommunicate - build chat AI with no-code tools (link)

  • Trellus - presents a cold call sales coach (link)

  • Ollie - finds the perfect gift for anyone (link)

  • Muze - AI finds the right song for what you’re seeking (link)

  • Brainfish - AI customer service platform that knows your info (link)

  • ColorpaletteAI - describe your website, and AI makes the colors (link)

  • AI Diary - a daily diary with AI’s help (link)

Good Content, Bond Posters

I saw these posters made with AI and suddenly craved a video game where you can switch between James Bond versions. Can someone please make this?

That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.

Check Out Our Sunday Edition

Every week you can dive into more AI news and topics with us by subscribing to our premium edition.

We’ve written about:

If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial, so you can read this one. That’s how excited I am to post it.