Open AI's Own Goal and Women's Servitude in AI Voices. Plus, a summer of super AI from Google and Microsoft

Each week, we bring you Embedded. This will be our news and interview series. We are working on the 10-part podcast season in the rundown. You can support the lean, hardworking and hustler team by subscribing here to help keep the lights on. We appreciate it as always. We have a Bitcoin wallet if that’s easier for you.

Do be in touch if you would like to privately test our Pressmate AI product, built to help communications teams develop more powerful press releases, quickly and better target journalists looking for great and newsworthy stories. More here.

Own Goal

She said no, but Open AI did it anyway. The Washington Post asks, ‘Exactly how stupid was what OpenAI did to Scarlett Johansson?’ This thread captures the mess and the response. Here’s what happened.

• Sam Altman, Open AI, asked Scarlett Johansson if she would be the voice of its GPT chat assistant (Sky)
• According to Johansson, she said no back in September. They asked again. She said no.
• The new GPT was released, and Johansson said she was shocked’ and ‘angered’; it sounded “eerily similar” to her, and they didn’t have her permission. ‘Her', by the way, is a movie she was in where she was the voice of an emotional AI companion.
• Open AI said it was not meant to sound like her.
• Open AI is stopping its use of that voice. An own goal.

Midjourney AI

After riding high last week following the drop of GPT-4o, OpenAI is now in crisis mode and a legal mess of its own doing.

This isn’t as consequential. But consider the context around OpenAI, which is probably the most important technology company in the world right now. OpenAI is already being sued over stealing others’ work, which has called into question the foundation of this new generation of AI. Altman also has a reputation for running roughshod over people who stand in his way - Washington Post

Women’s Voice’s in AI Assistants and Servitude

This whole thing got me thinking about why virtual assistants have a woman’s name and voice anyway. Studies apparently show that attributing female traits to robots makes them ‘feel’ more human.

Midjourney AI

They are warmer and more helpful, and companies cite these studies, arguing that a female voice helps sales. Altman’s case to Johannson was that "it would be “comforting” to people not at ease with the GPT technology.

Here’s a view from El Pais essentially saying the problem lies with the male product guys and stereotypes, and this needs to change.

It is very clear that the problem lies with the people who manufacture and make decisions about this type of technology: “The vast majority of people who have shaped these products are men with great gender and cultural biases”. Although Aranda thinks that it will be very difficult to change these stereotypes, she thinks there is some hope: “The only way to hack the system is for there to be more women in the sector. In the end, the decisions that programmers make when creating these devices are based on their belief system, and if they have always seen women as assistants, they will reflect this in their AI.

Adding to that, the ‘flirty’ personality of the ChatGPT voice is weird. Not cool. Do not like this.

All this adds to the concerns over at Open AI

AI Safety and Responsible AI may be on the line. Key people are leaving, as this Substack reported recently. Jan Leike, a key safety researcher at OpenAI who co-headed the superalignment team ensuring AI systems adhered to human values, resigned just days after the drop of GPT-4o. His resignation follows that of Ilya Sutskever, another senior safety figure, amid growing tensions about the company's priorities. Here’s what Leike said:

• He said OpenAI is prioritizing "shiny products" where “safety culture and processes have taken a backseat.” Not good.
• In a thread, he explained how he’d finally reached a breaking point over disagreements over the company’s priorities:
• Leike stressed the need for OpenAI to focus more on safety, social impact, confidentiality, and security for future models, saying that these critical issues are not receiving enough attention.

An insider source told CNBC that OpenAI dissolved the entire superalignment team that focused on long-term AI risks. When the team was announced, OpenAI has initially pledged 20% of its computing power to this over the next four years.

Here’s what’s new with Google: Veo, Imagen, and Lyria

1. Video

• Google’s new text-to-video generator, called Veo, can now produce content longer than 60 seconds at 1080p resolution.
• This would be their competitor to OpenAI’s Sora, and they’re saying it has the capability to create footage that’s consistent and coherent, so people, animals, and objects move realistically throughout shots.

It’s only available to select creators in a private preview mode, but Google plans to incorporate this into YouTube shorts in the near future. You can check out the unveiling here.

Tools like this have a major impact on the creative industry and Google is looking to preempt any backlash by partnering with artists as spokespeople for their new tools. They did a collaboration video with Donald Glover where they talked about Veo’s impact on storytelling by empowering artists to create quality content quicker and easier than ever before. Worth a watch:

With copyright in mind, they’ve also built their own solution called SynthID, which "embeds imperceptible digital watermarks into AI-generated images, audio, text, and video," so creatives can protect their work.

2. Images

Google also unveiled their new image generator called Imagen 3, which is only available via a private preview, although there is a waitlist.

3. Music

Google also partnered with musician and rapper Wyclef Jean to create a video showing some tracks he made using their new AI music sandbox, called Lyria. The video also focuses on how these new tools help bring ideas to life and make music in ways that have not been possible before. You can listen to it here.

Microsoft is going multimodal

• They dropped the Surface Laptop and a Surface Pro tablet using Microsoft’s CoPilot Plus. Microsoft haven't just focused on new software here, they’ve also switched up their hardware to be more AI compatible.
• Devices will be “58 percent faster” than a MacBook Air with an M3 processor and have battery life that lasts “all day,” according to exec Yusuf Mehdi.
• One new feature of the new line of devices is called “Recall,” which uses AI to create a searchable “photographic memory” of everything you’ve done and seen on your PC. Many people on social media are comparing this to Jarvis, the AI assistant in the Iron Man movies. The AI can watch, hear, and understand what’s going on and respond to your questions in real time.
• Find out more about Recall here.

We’re entering this new era where computers not only understand us, but can actually anticipate what we want and our intent.” If this sounds creepy, Microsoft says there will be an option to filter out what you don’t want tracked to protect users’ privacy - CEO Satya Nadella

• Their new line of laptops and tablets will have an ARM-based chip from Qualcomm that can run some AI tasks without an internet connection. Microsoft is hedging on the energy-efficiency of these chips, that can handle AI models while maintaining a longer battery life.
• Brands like Lenovo, Dell, HP, Asus, Acer and Samsung, which all use Windows, are launching PCs powered by processors so their computers can be AI-ready.
• When will these be available? The new devices will be available from June 18 but can be pre-ordered now. Copilot+ PCs will start at $999.