😮 Massive AI News Day

You cannot turn your back on the AI news for 15 minutes without missing something significant.

Before we get to it today, a quick email from a reader:

“Thank you for doing this! Smoking Robot is my go to in the rapidly developing world of AI. I can’t keep up without it! Keep up the good work SmokeBot, sir.”

Yeah, now we’re talking.

If you find this email interesting, helpful, funny, intelligent, whatever, do us a favor— refer one friend. We’re currently revamping our referral program, but you can just drop them this link: https://smokingrobot.beehiiv.com/subscribe

Please and thank you.

In the email today:

• ChatGPT releases tool to detect ChatGPT content 🖥️ 

• Headlines 📰 

• Tool of the Day 🔨

• 4chan users predictably got hold of ElevenLabs’ Prime Voice AI and wreaked havoc with it 🎤 

• Links 👀 

Even Twitter feels a step or two off the pace these days.

“Hey! A bot wrote that!” “Yeah, and that’s exactly what a bot would say!” 👯‍♀️

Mass hysteria, particularly in the academic world, over what ChatGPT can do to the written word has many people (professors, principals, kids who refuse to cheat) on the run.

OpenAI is reacting to the backlash by releasing an antidote to the virus it set loose on an unsuspecting world:

Details:

• The tool is currently available and free to use

• It will help determine if blocks of text were generated by ChatGPT

• The results are based on a 5-point ranking system from most to least likely, and OpenAI cautions the tool is not perfect

• It works best on over 1,000 words in English (other languages not so much)

• It can’t help determine if code was written by an AI

In an interview a few weeks ago, OpenAI CEO Sam Altman said no AI tool developed to detect AI content would be perfect and workarounds would always be possible, even through just minor human edits to the output.

Our take: As we’ve written about before, we’re quickly headed for the Clone Wars of AI. It turns out, the immediate response to detect and combat AI is with… more AI.

Eventually we may live in a world where consumer chatbots converse with corporate chatbots. AI generated papers are graded by institutional AI. Automated financial reports are audited by specialized GAAP accounting AI. And so much more.

What is the endpoint, exactly? A series of technological escalations to the point where vast swaths of humanity are automated.

Maybe that’s overstated, though… because OpenAI just released an update to get ChatGPT better at basic math:

Still not perfect!

Headlines 📰 

In a New York Times ad, Fiverr CEO Micha Kaufman says he comes “in peace” to AI: Kaufman running a full-page ad in the New York Times, saying, apparently to probably soon to be sentient robots, “You are a unique and powerful tool” in explaining his company’s decision to add AI categories to its platform. If this all feels a bit like we’re meeting Aliens for the first time to you… yeah, you’re not alone.

Metaphysics unveils partnership with Hollywood agency CAA for using generative AI in film: Tom Hanks and Robin Wright will be “de-aged” in an upcoming film with use of the Metaphysics Live tool. The company will work with CAA to develop best practices and services for talent when using AI. The possibilities are endless: from copyrighting use of likeness (helllllllo, Morgan Freeman’s voice), to de-aging, promotional use, and the inevitable lawsuits that will follow. But at what point does the human actor just become irrelevant in all of this, and an original AI is the star? Probably soon.

Enterprise AI provider C3.ai sees its stock spike 20%: The company announced that it would launch a generative AI search tool using ChatGPT for its enterprise products, allowing users to search vast databases to easily pull information. The ol’ ChatGPT stock bump. Can’t wait to see how Coke incorporates it into a future press release.

Tool of the Day 🔨

Podcastle.

Record, edit, and generate transcriptions from your podcasts using AI.

Read our full review of Podcastle here.

Try Podcastle right here.

FOLLOW UP: ElevenLabs’ Prime Voice AI Tool Was Predictably Used To Clone Celeb Voices 🎤

Who had this headline on their 2023 Bingo card?

We don’t spend time on 4chan as a rule. SmokeBot can’t even handle those Sarah McLachlan puppy commercials… no way he needs the snuff fare on 4chan in his life.

But every once in a while, the dark web thrusts itself into the light with something hot hilarious totally inappropriate HOW DARE THEY!

This week, its users took our Tool of the Day 🔨™️ from Monday, Prime Voice AI, and used used it to clone the voices of celebs:

Well then!

SmokeBot loves it when a send comes together like this. But this is too much. Not more than a few paragraphs earlier we told you about Metaphysics’ deal with Hollywood talent agency CAA to use celeb likeness with AI, and now Rick is cutting his poor cartoon wife into a thousand pieces.

Joking aside, this level of accessibility to really good voice cloning tools is world-changing.

ElevenLabs is reacting quickly. They are taking the following steps:

1. Like OpenAI, they’ll release a tool to help detect whether a clip was generated with their software. This will be available next week.

2. Only paying users will be able to generate voices.

3. They’ll offer a $5 monthly subscription— a nominal cost that will require users to put in a credit card, and thus be able to have their identity traced.

Your move, 4Chan.

Our take: Most celebrities will fall into two risk categories:

1. Celebrities with bad reputations. People are going to be pretty likely to believe that Mel Gibson or Liam Neeson are capable of saying all sorts of objectively bad stuff. Believability bar is low.

2. Career-ruining stuff for squeaky clean celebs. Like when you found out Matt Lauer was a derelict (OK, maybe just SmokeBot was surprised). This can quickly become image-ruining stuff for the likes of Emma Watson. Even if corrected within minutes, some subset of people will believe that it was said.

And then there’s you, me, and the entire human race: If your voice has been recorded anywhere, a bad actor using AI can do anything he or she wants with it. A deepfake call to your parents, or your ex, or your boss, saying the last things you would ever think to say.

Pretty scary, but it’s not like most people reading this have recorded a podcast… posted a video on social media… or just generally appeared on video or audio for two straight years over a computer network. Nope.

Links 👀 

• Glass AI posted a tool which can give you a differential diagnosis and a treatment plan based on input of symptoms 🥼 

• AI thinks the planet is going to warm by another 1.5 degrees Celsius (which in case you didn’t know, is bad) in the next 10-12 years 🌡️ 

• Online slot machine play is wildly profitable for virtual casinos, and AI has a big hand in that 🎰 

• The Justice Department is concerned that government agencies are using AI to perpetuate discrimination  

• Traders pile into AI stocks 📈

• Even as AI is threatening white collar jobs everywhere, it is simultaneously improving the composition of job postings, Yin and yang ☯️ 

• Gmail creator says ChatGPT will destroy Google’s business in two years 😮

• AI unearths unknown play by one of Spain’s greatest writers 🪶

• Will AI be the first to discover alien life? 👽️