🤝 Google and OpenAI's next steps
Google-Apple AI Partnership ◆ Deepfake Video Generation ◆ Google-Fitbit Health AI ◆ Sam Altman on GPT-5
Dear curious minds,
There was no major news this week, but the field is not standing still and there are many advances in various areas. This week’s issue highlights especially what Google AI is currently working on and gives you on top a glimpse of what happens at OpenAI from an interview with its CEO Sam Altman.
This week’s issue brings you the following topics:
Google's AI Takeover: Apple Partnership In Discussion
Deepfake Alert: Make Anyone Say Anything
Health Insights: Fitbit and Google AI Team Up
Navigating the Future: Sam Altman on the Uncertain Road to GPT-5 (podcast)
If nothing sparks your interest, feel free to move on, otherwise, let us dive in!
📱🤖 Google's AI Takeover: Apple Partnership In Discussion
With the AI industry advancing rapidly, Apple faces pressure to match the pace of competitors like OpenAI, Anthropic and Google.
Apple is developing some AI-powered features for iOS 18, but these will be limited to tasks that can be performed directly on the device.
According to a Bloomberg article, Apple is in active discussions with Google to integrate the cloud-based Gemini AI models into iPhone features.
My take: Apple's job listings indicate work on generative AI tools, but exploring third-party AI technologies suggests its in-house AI development may not be meeting expectations. Google's potential partnership with Apple, combined with existing collaborations with Samsung and their own Pixel line, would position Google to become the dominant force in smartphone AI features.
⚠️🤥 Deepfake Alert: Make Anyone Say Anything
Google researchers have developed VLOGGER, an AI that transforms still photos into realistic talking head videos. The AI takes a photo and an audio clip to produce a video where the person appears to speak the audio's words with corresponding facial and body movements.
Example videos generated with the same audio for various images. [source] VLOGGER uses advanced diffusion models to generate videos from a single image, mimicking speaking, gesturing and movement. A significant dataset with over 800,000 identities and 2,200 hours of video, was used to train the system.
The approach outperforms previous methods by not requiring individual training per person, handling full images beyond just faces, and accommodating diverse scenarios.
Potential uses include language dubbing, video editing, and creating virtual avatars, but there are concerns about misuse in creating convincing deepfakes. Despite its advancements, VLOGGER has limitations, such as generating only short videos with static backgrounds.
My take: The development marks significant progress in generating very specific videos of individuals and highlights the evolving challenges in distinguishing between real and AI-generated content. We have to understand that we are already living in a time in which we cannot blindly trust a video source.
📈❤️ Health Insights: Fitbit and Google AI Team Up
Google shared, in a blog article, a summary of their recent activities at using generative AI for health related topics.
Already last year, Google released Med-PaLM 2, an LLM fine-tuned for healthcare at later in 2023 MedLM building on top of this model.
New AI features in Fitbit Labs will analyze your sleep patterns, activity levels, heart rate, and more. This will help you discover connections in your health data (like how more activity can lead to better sleep).
The Pixel 2 smartwatch brings various health related features. One example is measuring your heart rhythm for atrial fibrillation with the ECG app and easily share results with your doctor. [source] Fitbit, acquired by Google in January 2021, is partnering with Google Research to create a personal health large language model (LLM) that can reason about a user's health data and provide tailored recommendations, similar to a personal coach. This model will be fine-tuned using Fitbit's anonymized health data.
Furthermore, Google shared their work, which intends to make health information easier to find by searching with image inputs and translating educational health information into other languages.
My take: Nice to see Google's advancements in the health area, where a small change can have a positive impact on people’s life. The big upside is shadowed by possible wrong information. Google states that the first step of medical AI is to use the new possibilities as assistance systems for medically trained persons.
🎙️🗺 Navigating the Future: Sam Altman on the Uncertain Road to GPT-5 (podcast)
Lex Fridman interviews for the second time Sam Altman, the CEO of OpenAI.
Sam Altman during the interview by Lex Fridman. Nice mug! [source]
You can listen to the podcast in the classic way. Furthermore, the recording including a video stream of the conversation was published on YouTube and 𝕏. However, to get a fast overview and extract the information on a specific part of the conversation, you might want to take a look at the nicely structured official transcript.
The interview starts with Sam reflecting on his view of the most turbulent week at OpenAI, where he was laid off by the board and five days later returned as CEO.
Sam acknowledges that the performance of Sora in generating videos has exceeded his expectations and that it is a significant milestone in the field. He acknowledges the potential dangers of such a powerful tool, including the risks of even more convincing deepfakes and the spread of misinformation. OpenAI is taking a thoughtful approach about their releases, and there are many possibilities how Sora can have a negative impact.
About the release timeline of GPT-5, Sam expressed that he does not know when it will be released. He stated that OpenAI will release an impressive new model this year, but he was unsure of the name it would carry. He highlighted the complexity of developing such advanced models, emphasizing that it involves a combination of numerous medium-sized innovations rather than a single breakthrough. Sam said that before discussing a GPT-5-like model, OpenAI has many other significant releases planned.
My take: Nothing entirely new or fully unexpected was revealed in the interview from my point of view. However, there is way more discussed in nearly two hours than what is stated above. You should consider listening to it yourself.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.