GPT-4o and the aesthetics of AI; Google I/O is taken over by generative AI; deepfakes blur India's elections; AI thrives in Brexit Britain

AI startups are making their home in New York; AI program breaks barriers for female students; Hugging Face gives away $10m of compute to startups; AI startups target climate change

May 17, 2024

∙ Paid

Two days after Google I/O had ended and three days after OpenAI had livestreamed a Spring Update presentation, Sam Altman posted a message on X comparing the aesthetic of his office to Google’s conference center:

His post took me back to 2022, when OpenAI unveiled DALL-E 2. People were captivated by a different kind of aesthetic: the whimsical, creative images the AI generator could produce from simple text prompts. A corgi wearing a beekeeper suit, a radish walking a dog on Mars—if you named it, DALL-E could generate it.

Similarly, when Google demoed its chatbot Bard at its 2023 I/O conference, audiences were wowed by its ability to explain scientific concepts in simple terms and brainstorm ideas like a human collaborator.

As I saw Mr Altman’s post on X, I asked myself the same question I had when the first DALL-E images came out: “Who is this for?”

Both companies' splashy debut of generative AI tools seemed aimed at capturing the general public's imagination and consumer market. We have a term for this in comms and marketing: gen pop tech. Consumer tech products that achieve widespread mainstream adoption have enormous revenue potential given the sheer size of the addressable market. Social media apps have billions of users globally because they targeted general consumers, rather than just businesses.

There are also advantages in terms of brand mindshare and impact: becoming a household name and capturing mindshare early in an emerging technology space offers sales and marketing advantages. If consumers first experience a new technology through a company's product, it can create long-lasting associations in other parts of their life, including at work. Apple and Microsoft benefited immensely from this with PCs.

So naturally, OpenAI and Google have spent this year converting curious consumers into dedicated enterprise users, trying to sell AI systems to businesses on a hunch that many people are already (and secretly) using generative AI tools at work. And they’re not alone: other startups such as Stability AI, Cohere, Anthropic, Tome and HeyGen have attempted a similar strategy, and in some cases pivoted entirely to the enterprise space where the more lucrative margins can slow down the bleeding from the very high cloud computing costs associated with keeping the AI lights on.

But for the companies pioneering generative AI, the product strategy shift isn't quite so simple. There's a big difference between delighting consumers with fun use cases, and actually delivering enterprise-grade AI models that are reliable and scalable enough for businesses to build products on. One challenge is that these large AI models require immense compute resources just to serve a trickle of traffic from consumers playing around. Scaling that to enterprise usage can not only be prohibitively expensive but also incredibly laggy, as the chart below produced by UC Berkeley shows.

But focusing too much on enterprise customers also presents risks for falling behind on consumer relevance and mindshare—in other words, the consumer space gives you a certain aesthetic that all tech companies value. This might explain why OpenAI marketed GPT-4o primarily on its aesthetics: the model is not necessarily smarter; instead, it’s been touted as faster and more conversational (errr… very conversational), responding to audio inputs in as little as 232 milliseconds, which is comparable to human response time.

As history has shown time and again in tech, when it comes to the consumer versus enterprise product Catch-22, there are no perfect answers. Striking the right balance will be AI's latest existential product challenge.

And now, here are this week’s news.

❤️Computer loves

Our top news picks for the week - your essential reading from the world of AI

MIT Technology Review: OpenAI’s new GPT-4o model lets people interact using voice or video in the same model
Also, from the same OpenAI event on Monday:
- The Verge: OpenAI’s custom GPT Store is now open to all for free
- The Verge: ChatGPT is getting a Mac app
- The Verge: ChatGPT will be able to talk to you like Scarlett Johansson in Her
MIT Technology Review: Google’s Astra is its first AI-for-everything agent
Also, from Google I/O on Tuesday:
WSJ: AI Startups Are Making Their Home in New York. Can They Turn It Into an AI Powerhouse?
The Economist: Today’s AI models are impressive. Teams of them will be formidable
New York Times: A.I. Program Aims to Break Barriers for Female Students
Sifted: ‘Not being subject to the EU’s regulatory fervour is massively working in our favour’ - AI thrives in Brexit Britain
New York Times: The Old-Fashioned Library at the Heart of the A.I. Boom
BBC: AI and deepfakes blur reality in India elections
FT: AI start-ups take aim at climate change
The Verge: Hugging Face is sharing $10 million worth of compute to help beat the big AI companies

⚙️Computer does

AI in the wild: how artificial intelligence is used across industry, from the internet, social media, and retail to transportation, healthcare, banking, and more

🧑‍🎓Computer learns

Interesting trends and developments from various AI fields, companies and people

Keep reading with a 7-day free trial

Subscribe to Computerspeak by Alexandru Voica to keep reading this post and get 7 days of free access to the full post archives.