Alphabet merges DeepMind and Google Brain in Major AI Restructuring
Big Tech is all in on (Gen)AI. Are you?
Happy Friday all,
I've been all in on GenAI for a few years... and Big Tech is too.
Yet more news today about how all the 'big dogs' are working overtime to get ahead of the GenAI race. Yesterday, Alphabet, Google's parent company, announced the strategic merger of its two prominent AI-focused groups, DeepMind and Google Brain, to consolidate and strengthen their AI development efforts.
"Combining all this talent into one focused team, backed by the computational resources of Google, will significantly accelerate our progress in AI."
- Sundar Pichai, CEO of Google and Alphabet
The new unit will be known as Google DeepMind. Looks like the original DeepMind founder, Demis Hassabis, has had a change of heart then! In January, he was preaching caution, telling Time magazine: "I would advocate not moving fast and breaking things." Now he's the new leader of Google's super-AI unit. (I'm getting flashbacks of Elon Musk signing the AI Moratorium letter before promptly buying thousands of GPUs to set up his GenAI venture.)
I've said it before, and I'll say it again. The release of ChatGPT was a watershed moment in the history of humanity's relationship with machines. The OpenAI Chatbot was the lightning rod that galvanized the current Big Tech GenAI arms race. ⚡️ So what can we expect from these companies next?
First, on the technical side, a race to develop a series of powerful, multimodal AI models. So far, generative models like GPT-3 or Stable Diffusion generate outputs in one medium—in this case, text, or images, respectively. Multimodal models are the next step—they can work across mediums. Imagine a version of ChatGPT that can not only generate a script but also a storyboard to illustrate it as well. That's where we are headed next.
Second, they are lawyering up. They know that big, precedent-setting legal arguments about interpreting and protecting AI-generated content outputs must be made (and won.) AI litigation is going to be hot. 🔥.
Here, I see two strands emerging.
The first is defense. They must respond to litigation that AI-generated output violates IP and copyright protections. Midjourney, Stability AI, and Deviant Art are already subject to a class-action lawsuit on the grounds of copyright infringement. Getty Images has sued Stability AI too. Meanwhile, Microsoft is facing a class-action lawsuit against Co-Pilot, its code-generating model. But as more (and better) AI-generated content emerges 'in the wild,' this fight will get increasingly bitter. This week we got a primer when Universal Music Group forced a takedown of a viral song posted on TikTok featuring AI-generated vocals of Drake and the Weeknd.
But Big Tech (and Generative AI companies) must also go on the offensive. They need to proactively argue in favour of being able to protect AI-generated content as IP and copyrightable. Suppose you conceive of Generative AI as a turbo engine to power all information and knowledge. In that case, there is no way we can operate society without quickly figuring out a way for AI-generated output to be subject to ownership and protection too.
Whatever side you're on— whether arguing that AI-made content is illegal or that it should be protected—lawyers, welcome to the golden age of AI litigation.
See, AI is already creating more jobs! 😜 (I jest….)
Now for the best of the rest:
Stability AI launches their own language model
Stability AI has unleashed its latest creation: StableLM, a set of large language models that they've made open-source to the public.
This comes after their previous success with Stable Diffusion, which made waves in the image generation world.
With current models boasting 3-7 billion parameters, and future models planned to have up to 65 billion parameters, this could potentially be the biggest open-source language model available. And let's not forget their catchy tagline: "Transparent, Accessible, Supportive."
So, the question remains: do you prefer Open AI's black box and API approach or Stability's open source and transparent method? The AI open-source world just got a whole lot more interesting.
Microsoft develops its own AI chips
Microsoft is hunkering down! Now, it’s developing its own in-house chips, codenamed Athena, that could rival Nvidia's offerings. The chip is expected to help Microsoft save money by reducing its reliance on Nvidia's pricey GPUs, which are in high demand and selling for over $40,000.
Though the chips are not direct replacements for Nvidia's, they could significantly cut costs for Microsoft as it rolls out AI-powered features in Bing, Office apps, GitHub, and more. Looks like the competition in the AI chip world is heating up too….
Editing photos with AI just got even better: Adobe introduces Lightroom AI
Adobe Lightroom just got smarter with its latest update. Thanks to AI-powered features like Denoise and Curves in masking, even photo-editing newbies can make highly precise edits to their images.
With adaptive presets and expanded masking categories, Black & White for video, and AI masking and adaptive presets on the web, everyone is now equipped to channel their inner Annie Leibovitz.
Elon Musk has revealed his plans for “TruthGPT”, an AI model that aims to maximise truth-seeking and understand the universe.
Musk sees TruthGPT as a more transparent option than OpenAI's AI models, which he claims might be compromised by profit incentives.
As I commented on media — this is a pretty cynical ploy by Elon (but perhaps a masterstroke for marketing), because, as we know, there is no such thing as an LLM that delivers ‘truth’.
It remains unclear how far along the development of TruthGPT is, but it seems war has started between Musk and OpenAI – a theme I’m sure we’ll return to in EGAI!
AI sings a Drake and The Weeknd duet, gets removed from streaming services
An AI-generated song with Drake and The Weeknd's voices went viral, but streaming services snatched it away. The "heart on my sleeve" singer used AI to replace their voice with Drake's, and listeners were impressed yet worried.
Universal Music Group, representing both artists, claimed a copyright violation and removed the song from all platforms. UMG wants to stop AI programs from training themselves with copyrighted music. (Good luck with that?)
Google Developing GenAI for ads…
Like Meta, Google is rumoured to be working on a platform that allows for AI-generated ads, potentially changing the game in digital marketing.
It would allow companies to reduce ad costs and enjoy more ad personalization.
But what are the ethical implications of AI-generated ads? And how do you control for misleading ads from reaching consumers? It’s going to be a bit of a mess this one.
That’s all for this week… see you next week!
Namaste,
Nina