#96 - Summer Update

Welcome to this summer recap edition of Wayfinder, bringing together some of the most important updates that happened in July. While it seems everyone is on holiday, the folks at AI companies are still hard at work. This past month saw an avalanche of new AI releases and updates. However, there's a growing sense that much of this activity is aimed at keeping investor money flowing , as questions about immediate return on investment (ROI) start to surface.

Let's dive into some of the key developments of the past month.

The Shift to Search

A trend that was already emerging accelerated in July: the shift of AI companies towards search. In a bid to monetize their massive AI investments, a wave of new search-related products came out.

• Microsoft launched Bing Generative Search , combining generative AI with search results. They claim early data shows this maintains clicks to websites.
• OpenAI announced a SearchGPT prototype , featuring a simple interface with navigation, external links, and a summary with citations. It heavily relies on external aggregators.
• Perplexity AI launched a Publishers Program for their 'answer-engine' partnering with major publishers like TIME and Der Spiegel, offering revenue sharing, API access, and potentially even bundled subscriptions.

Doubling Down on AI Models

July also saw a flurry of new AI models, each showcasing a different approach. Models are becoming more open, smaller, specialized, mobile, human-like, and customizable.

• Meta released a new powerful version of their open-source AI model Llama (3.1 405B) . By making it free for developers to use and customize. Meta's not just being altruistic - it knows this hurt AI companies that have to monetize their own closed AI models.
• OpenAI introduced a GPT 4o mini-version . This fits in a trend of making AI models smaller and more efficient, which could lead to faster, cheaper AI applications. It also released a long-output version of it’s latest model allowing to output 64k tokens, which is a decent 200 pages of text.
• Google DeepMind has been working on more specialized AI models, focusing on specific tasks like coding or mathematical problem-solving.

Other updates

• Apple released a preview of "Apple Intelligence". Link
• Mistral unveiled their new model Mistral Large 2. Link
• ChatGPT's advanced voice mode has been rolled out (to a limited number of people). Link

AI Content Creation

While AI clearly still struggles with facts, it has made remarkable strides in media creation. A few summer releases showcase improved user control, higher output quality, and are tailored to content creators.

• Runway ML released Gen-3 Alpha. The latest version generates video even faster and offers creators more control over the output.
• Odyssey Systems is pioneering a Hollywood-grade visual AI, where scenery, characters, lighting, and motion can be both generated and directed,
• Google introduced an AI-powered video creation platform called Google Vids. It's made for work and should streamline the process of professional-looking videos with minimal effort.
• Stability AI is expanding its roster of generative AI models, adding a new dimension with the debut of Stable Video 4D . It transforms single object videos into multiple novel-view videos from different angles.
• Creators can now make their own custom AI using Meta's new AI Studio platform .
• AI clones took over a Swiss radio station for one day , demonstrating both the potential and limitations of AI in content creation and broadcasting.
• Related: A new Reuters Institute podcast series looks at AI and the future of audio.

Regulation and Market Developments

While the AI train is still rolling, it might be heading into more challenging terrain. Investors and companies alike are starting to grapple with the realities of turning AI's potential into sustainable business models.

• A recent report from Goldman Sachs raised eyebrows by questioning whether the massive AI investments (over $200 billion this year alone) will yield similar returns.

• Meanwhile the EU AI Act officially entered into force , setting a regulatory standard for AI and potentially complicating the AI gold rush.
• And unbridled access to data is over. Many of the most important online web sources for AI training data have blocked companies from scraping their content. A new MIT study found that 5% of all data, and 25% of data from the highest-quality sources, has been restricted.

Industry Reports and Insights

The McKinsey 2024 Tech Trends Outlook

Pugpig 2024 State of the Mobile Publishing Market Report

Reuters Institute Report on AI Attitudes

Dan Frommer's Mid Year Consumer Trends Report 2024

Digital Startups Monitor by Global Project Oasis

Google Privacy Sandbox Update

Read

Analyst Ben Evans considers the AI Summer. He thinks LLMs might be a trap: they look like products and they look magic, but they aren’t. Maybe we have to go through the slow, boring hunt for product-market fit after all.

Content Strategist Laurence Van Elegem compares the huge amount of reporting about artificial intelligence and the deluge of product updates to the mythical idea of the flood and warns we’re increasingly focused on adaptation rather than prevention.

And Nick Diakopoulos considers the impact of Gen AI on journalistic labor. According to a a new study current LLMs significantly aid 43.3% of tasks for journalists, and 52.4% of tasks for editors. Only 16.7% of common tasks in the occupation can’t be helped by LLMs.

Related: The Pudding tested if AI can make a data-driven visual story. The result - nah - it was helpful in isolated work, but couldn’t handle the full project or overcome complex tasks.

And as a user experience nerd, I was thrilled when I found this collection of new UX patterns . Or how user experience will evolve with the growth of Artificial Intelligence.

Stay tuned for another summer update at the end of August. And don't worry, the regular weekly newsletters will be back in full swing come September.

Enjoy the holiday and feel free to share with friends and colleagues!

Ezra