ONE BIG READ

Anthropic’s newest model

The AI arms race is continuing at breakneck speed. Anthropic, the AI company that netted a whopping $2.75 billion in funding from Amazon earlier this year, just launched its latest model, Claude 3.5 Sonnet, out of nowhere. 

Sonnet takes up the middle position in the company’s model lineup behind Opus, its highest-end model, and ahead of Haiku, its smallest model. (Is it just me, or is keeping up with all the different names for AI models starting to get a little confusing?) Interestingly, the company claims Sonnet—the middle offering— now outperforms Opus, its largest offering — until Opus gets its own 3.5 update, of course.

What’s the big deal? Like all new releases, the hype comes from how it performs against others, especially OpenAI’s models. According to Anthropic, Sonnet 3.5 matches and beats GPT-4o and Google’s Gemini across various tasks. When you look at the benchmarks, it does look pretty impressive. It outscored GPT-4o, Gemini, and Meta’s Llama 3 400B in seven of the nine overall benchmarks and four of the five vision benchmarks. Of course, benchmarks should be taken with a grain of salt since companies can pick and choose based on what looks good. 

Alongside Sonnet 3.5, Anthropic also released a new feature called Artifacts. With Artifacts, you can now interact more directly with the results from your Claude-based query. Say you generate an image, and if you’re not completely happy with it, you can edit it directly in the app, or maybe you’re writing a cover letter. Instead of copying and pasting it to the notes app, you can make changes directly inside Claude. 

PRODUCT HIGHLIGHT

This new social app is replacing humans with AI characters

Meet Butterflies. It’s a new social app by ex-Snap engineering director Vu Tran. On the surface, you could mistake Butterflies for any number of Instagram clones. You have a grid-based profile, can follow others, and your timeline is littered with photos from “friends.” 

However, things quickly change once you start using it. Once you sign up, you’ll be asked to create an AI character or, as the app calls it, a “butterfly.” From there, your butterfly will start generating and sharing photos and interacting with other butterfly’s posts. There’s no limit to how many butterflies you can create, either.

During its private beta, Butterflies attracted tens of thousands of users and a healthy $4.8 million in funding from Coatue, SV Angel, and others. This week, it finally went live in both the iOS App Store and Google Play Store.

When I was playing around with it, I couldn’t shake the uncanny valley feeling. On the surface, it feels like a real, human-led social network, and anyone in passing would probably think you’re scrolling Instagram, but when you look a little closer, you notice things like people with seven fingers or three arms. After reading a few generated comments, I noticed the repetitive and hollow language common in AI models. 

But maybe that’s the point: As AI gets smarter and more human-like, so too will Butterflies. At the very least, it’s an interesting experiment where you can follow the progress of AI in real time in a more human setting instead of reading about the technical capabilities of different AI models. 

AI IN THE HEADLINES

• Former OpenAI chief scientist Ilya Sutskever has launched a new AI company.

•  Meta has made its AI chatbot available to all users in India.

• Apple has delayed the launch of Apple Intelligence in the EU citing regulations.