🤖 AI Brief: More music drama, ChatGPT unbanned, and what execs fear most about AI

From the firehose of AI news, a few key themes emerged as critically important to understand this week. I narrowed down about 425 saved links and stories into the most impactful items below.

Reader feedback has been so useful since the last issue. From the vast majority of you I’m hearing one thing: it’s hard to keep track of everything, and that’s why a summarized digest matters.

I couldn’t agree more. This is why I write this every weekend, so I go into the next week feeling fully prepared.

In this issue:

🎸 Music is the next legal frontier AI will confront

🇪🇺 Regulatory developments in the EU speed up

😓 Google’s challenges continue

🤖 Elon Musk’s complicated views on AI

🏢 Corporations are unprepared for generative AI, study finds

and more (including the latest research developments)

Did someone forward this to you? Subscribe here!

🎸 Music is the next legal frontier AI will confront

The music industry is confronting a litany of AI-related issues at light speed.

• Since the release and takedown of “Heart on My Sleeve,” which featured AI voices mimicking Drake and The Weeknd, the internet has been flooded with additional AI-made Drake songs.

• Expect each of these to test legal waters around what is fair use, and what is copyrighted

• Josh Constine, a VC at SignalFire, puts it succinctly: “Google is caught between an AI rock and a copyright hard place. Either the AI Drake song trained on copyrighted data is fair use, YouTube floods with this content, and labels panic Or it’s infringement, which means Google’s Bard AI is illegal.”

• Adding to the conversation: musician Grimes has proclaimed anyone can use her voice for AI-generated songs, and she’d split 50% of royalties.

• An ongoing Andy Warhol copyright case could also have implications for generative AI, including AI music.

🇪🇺 Regulatory developments in the EU speed up

The legal landscape impacting AI models is rapidly changing, and this week saw two major developments:

• ChatGPT complied with Italy’s initial demands, resulting in the lifting of the ban. OpenAI added additional information on how it trains ChatGPT, provided EU users with a new form objecting to have their data used for training, and now verifies users’ age when signing up. Investigations into ChatGPT, however, remain ongoing in France and Germany.

• The EU has passed a draft of its AI Act, setting the stage for a finalization phase. The most important provision? A new clause that specifies AI models “would have to be designed in accordance with EU law and fundamental rights,” as well as a requirement that AI tools disclose the use of copyrighted materials. We’ll be watching this closely.

😓 Google’s challenges continue

Poor Google. Since Bard’s tepid launch they can’t seem to catch a break, and new reports highlight exactly how daunting the AI race will be for them:

• Mindshare about Bard remains low relative to OpenAI and Bing. According to Google Trends, ChatGPT is 8.3x more popular than Bing and 33x more popular than Bard.

• Google’s recent merging of Google Brain and Deepmind into a single AI-focused Google Deepmind team will face steep challenges. According to Google insiders, Deepmind has historically functioned very independently, thinking about Nobel prize-worthy problems, while Google Brain has operated with indecisive leadership. For the two teams to merge and move quickly to match OpenAI’s focus and speed will be a daunting task.

🤖 Elon Musk’s complicated views on AI

What exactly are the billionaire’s plans for AI? While no one knows for sure, new details surfaced this week that adds color to the mixed messages he’s been sending:

• The New York Times reported that Elon Musk had ordered Twitter to turn off OpenAI’s access to its historical tweets after ChatGPT surged in popularity

• Despite founding OpenAI in 2015, Musk has had a falling out with OpenAI on its mission and direction. He reportedly grew disillusioned when OpenAI stopped operating as a non-profit and built “woke” AI models.

• At the same time, the billionaire CEO is building his own Large Language Model as part of his new X.AI initiative. TruthGPT, Musk claims, is a “a maximum-truth-seeking A.I. that tries to understand the nature of the universe.”

💰️ AI roils the job landscape

Transformative technology has historically been a net benefit for society and GDP, but not without its intermediate pain. This is playing out at warp speed across multiple professions as AI’s power rapidly forces transformation.

• Dropbox announced a 16% headcount cut, citing AI as one of the reasons behind the significant layoff. What’s notable: this is a profitable, public tech company whose financial metrics have only improved in recent years. For AI initiatives, Dropbox is doubling down — but for mature teams, they’re making cuts. Expect this to be the broad theme of tech as AI surges to the forefront.

• Kenyan ghostwriters, who normally help US college students write essays, are losing jobs to ChatGPT. Rest of World reports that many ghostwriters have seen up to 50% decrease in work as AI has reduced demand for human writers.

• A Stanford/MIT study showed that GPT-3 software helped customer service agents perform as much as 35% better, portending big shifts in knowledge worker jobs as AI makes its way into numerous industries

🏢 Corporations are unprepared for generative AI, study finds

A KPMG study of 225 US executives found that 65% believe generative AI will have a high or extremely high impact on their companies, but nearly the same percentage say generative AI is still a year or two away from having an impact.

• While executives are optimistic, they are also worried it could have a negative impact, especially if risk is not managed

• Almost 4 in 10 executives believe generative AI could decrease social interactions and human connections among employees

• PwC announced a $1B investment in AI over the next 3 years; this is likely to become the norm as corporations pull the trigger on AI investments

🧪 Science Experiments

Amongst the dozens of impactful research papers coming out each week, we feature the most mindblowing examples below. As always, we try to explain anything technical to a non-technical reader.

Text to Video

RunwayML launched its Gen2 text-to-video model and the results are gorgeous. Here are several examples of what users have created in concert with Midjourney. The pace of development in the video space is simply on fire; imagine what could be possible by the end of the year.

• We previously reported on Nvidia’s text-to-video experiments, click here to see their examples.

Segment Anything, but for video

In another sign of how fast open-source tech is quickly improved, Facebook’s Segment Anything AI library was rapidly adapted into a video-tracking tech that beats Adobe’s own rotoscoping features in its professional software. The open source repository can be accessed here.

Robots playing soccer

Straight out of the geniuses at Google’s Deepmind team, they’ve applied a technique called Deep Reinforcement learning to help robots move in a dynamic environment. This is a good reminder that the latest breakthroughs aren’t just limited to generative AI. Check out the full videos here!

Text to 3D Models

A team of researchers generates surprisingly great 3D models out of text prompts. Expect this area of technology to rapidly improve in the next few months. What could that mean for 3D art and the artists who create models? Full research paper here.

Other News

News that didn’t make it into the key themes but is still worth keeping in mind.

WSJ reporter clones her own voice using AI, fooling her family and bank [Link]

Free course for developers on ChatGPT prompt engineering released, taught by OpenAI staff [Link]

Bark, an open source voice cloning tool is released [Link]

ChatGPT finally allows you to turn off conversation history and choose which conversations train their models [Link]

Will AI lead to mass employment? This author argues it won’t and examines how past technology disruptions have played out [Link]

Snapchat users really don’t like the MyAI assistant [Link]

And… that’s it for this week! Remember: your feedback is important and I respond to every email, so hit “reply” any time you have thoughts!

Or just take the poll below — this is massively helpful: