Google's highest-quality image model yet

Howdy, wizards.

For an easy overview of how good text/image/video/audio generation has become over the last 1.5 years, check out this recent article by Ethan Mollick.

‍

Dario’s Picks

The most important news stories in AI this week

Google quietly ships Imagen 3. It's Google's highest-quality image generator so far, and can generate detailed images based on your prompts as well as simple editing functionality. Contrary to DALL-E and Midjourney, which have significant guardrails for what you can generate, Imagen lets you generate some copyrighted stuff like famous logos and cartoon characters. However, it still has significantly more censorship than the newly launched Grok-2.

US-based users can try Imagen 3 now on ImageFX (Google's AI test kitchen).

‎ Why it matters‎ ‎ With all the storm around AI and copyright, and a federal judge allowing lawsuits against AI image generators for the first time, it may seem surprising that new image generators are getting less censorship. However, it seems that most of the biggest AI companies are taking a firm stance that being first is more important than being cautious.

‍

Anthropic's new prompt caching makes Claude API faster and cheaper. Anthropic just gave developers using the Claude API a big present. Prompt caching means they can now save frequently used context (e.g. uploaded documents, a codebase, any knowledge base) and access it quicker and cheaper. The costs reductions can be up to 90%. It's currently available for Claude 3.5 Sonnet and Claude 3 Haiku, with support for Claude 3 Opus coming soon.

‎ Why it matters‎ ‎ Notion is quoted saying prompt caching is helping them make Notion's AI features faster and cheaper. Features that boost speed and cut costs creates more incentive for developers to implement AI features, which we as users (hopefully) benefit from.

‍

OpenAI introduces structured outputs in its API. OpenAI is, by popular demand, introducing what's known as structured outputs for developers building with their API. Essentially, it ensures that the model output adheres to a specific structure (JSON Schemas) defined by the developer. This can be useful both for calling third-party APIs correctly, as well as responding to the user in a structured way.

‍

‎ Why it matters‎ ‎ Generating structured data based on unstructured inputs is something that has been traditionally hard to do, but is becoming orders of magnitude easier with AI. Let's say you ask for a travel itineary, structured outputs makes it possible (and easy) to give you back a clear organized itineary in a table or list, rather than just a big chunk of text.

‍

‍KPMG survey on GenAI’s impact on businesses, according to what leaders think. KPMG surveyed 225 senior business leaders at billion-dollar companies on AI's impact on how they're shaping their company's strategy.

‍

‍Over the next 3 years:
- 61% plan to expand GenAI applications in their company
- 24% plan to integrate AI deeply into their business
- Over 90% are confident or highly confident about their AI investment's ROI in a 1-3 years perspective
- 55% plan to invest in upskilling their employees for AI

‍

‎ Why it matters‎ ‎ Generative AI is here to stay, and it's not only a startup thing – the most established organizations in the world are currently planning for a future where AI plays a much bigger role in their business processes, and reskilling their workforce accordingly.