Sunday Rundown #52: Microsoft Build & Golden Gate Baby Names
Sunday Showdown #12: Claude 3 Sonnet vs. Google Gemini 1.5 Pro: Who's better at summarizing docs?
Happy Sunday, friends!
Welcome back to the weekly look at generative AI that covers the following:
Sunday Rundown + AI Fail (free): I share this week’s AI news and a fail for your entertainment.
Sunday Showdown + AI Tip (paid): I pit AI tools against each other and share a hands-on tip for working with AI.
On today’s “Sunday Showdown,” I’ll see which model is better at summarizing PDF documents.
Let’s get to it.
🗞️ AI news
Here are this week’s AI developments.
Microsoft had its annual Microsoft Build conference. Watch the recap by CNET. I’ll separate the announcements into “available now” and “coming soon” as always.
👩💻 AI releases
New stuff you can try right now:
Microsoft releases announced during Microsoft Build:
New business features in Copilot, including a Team Copilot, agentic custom copilots, and extensions.
GitHub Copilot now lets developers plug into third-party tools via Extensions.
The Phi-3 family of small language models (first announced in late April) is now generally available.
ElevenLabs launched a tool called “Audio Native,” which lets paid users add lifelike narration to their website or blog.
Perplexity now offers interactive, visual knowledge cards for certain queries, powered by Tako.
Khan Academy made its AI-teaching assistant Khanmigo available for free to all US teachers.
Midjourney now lets users with over 100 generated images create private rooms for collaboration with friends.
The Arc Browser introduced a feature called “Call Arc” that lets you get answera via voice.
Adobe added “Generative Remove” to its Lightroom product, letting you easily erase unwanted objects from images.
🔬 AI research
Cool stuff you might get to try one day:
Microsoft’s upcoming launches announced during Microsoft Build:
A range of Copilot+ PCs built for AI with some controversial features like “Recall," which takes screenshots of your work.
Microsoft Paint running on Copilot+ PCs will have an AI “Cocreator” that turns text prompts and doodles into new images. (Kind of like the very first post I’ve ever published on this newsletter.)
Microsoft Edge will be able to translate and dub YouTube videos in real time.
Amazon plans to release an AI-powered version of Alexa with a paid subscription.
Similarly, Meta is working on a paid version of its AI Assistant.
📖 AI resources
Helpful stuff that teaches you about AI:
“Mapping the Mind of a Large Language Model” - a fascinating deep-dive by Anthropic into how LLMs think. [Full paper here.] [Silly “Golden Gate Claude” experiment here.]
“Google CEO Sundar Pichai says AI search will actually help the web” - an interview by The Verge’s Nilay Patel [VIDEO].
🔀 AI random
Other notable AI stories of the week:
OpenAI and News Corp signed a multi-year agreement to help improve OpenAI’s offerings via integration with news content.
🤦♂️ 10. AI fail of the week
More a “for fun” than a fail. Golden Gate Claude suggests baby names:
Anything to share?
Sadly, Substack doesn’t allow free subscribers to comment on posts with paid sections, but I am always open to your feedback. You can message me here:
⚔️ Sunday Showdown #12 - Claude 3 Sonnet vs. Gemini 1.5 Pro: Which one can best summarize a document?
Last week, Claude finally came to Europe.
Naturally, I’m now eager to put Claude through its paces. Claude is touted as being particularly great at document summarization, so that’s what I’ll be testing.
The second participant is Google’s Gemini 1.5 Pro, which scores similarly to Claude 3 Sonnet on LLM benchmarks. They’re also both known for their long context windows.
They’ll be summarizing the DALL-E 3 research paper I once wrote a whole post about.
Let’s get rolling…