10X AI (Issue #42): Claude 3, Inflection-2.5, Stability AI 3D, and a "Wink" of Terror
Sunday Showdown #2: Suno and Riffusion write catchy jingles.
Happy Sunday, friends!
Welcome back to 10X AI: a weekly look at generative AI that covers the following:
AI News + AI Fail (free): I highlight nine noteworthy launches of the week and share an AI photo or video gone wrong for your entertainment.
Sunday Showdown + AI Tip (paid): I pit AI tools against each other in solving a real-world task and share a hands-on tip for working with AI.
Paid subscribers get instant access to every Sunday Showdown and other bonus perks.
Let’s get to it.
🗞️ AI news
Here are this week’s AI developments.
1. Claude 3 is kind of a big deal
What’s up?
Anthropic released the next generation of its LLM called Claude 3, which comes in three sizes: Haiku (cheap and fast), Sonnet (middle-of-the-road), and Opus (best and most expensive).
Why should I care?
Opus beats both GPT-4 and Gemini 1.0 Ultra on every measured LLM benchmark. (We don’t know how it fares against GPT-4 Turbo and Gemini 1.5.) Even Sonnet holds its own. Claude 3 also has vision capabilities and is safer while making fewer incorrect refusals.
Where can I learn more?
Read the announcement post.
Try Sonnet for free on Claude.ai. (Or pay for Opus.)
Watch this deep-dive by AI Explained:
2. Inflection 2.5 is catching up, too
What’s up?
Days later, Inflection dropped the Inflection-2.5 model, which powers its friendly Pi chatbot.
(Fun aside: This marks the second time Inflection and Anthropic both announce new models in the same week.)
Why should I care?
Inflection-2.5 gets close to GPT-4-level performance despite being trained on only 40% of its compute. Pi sets itself apart from other chatbots with its natural and empathetic tone. Now that it is powered by Inflection-2.5, Pi is even more capable and can browse the web to give you up-to-date responses. Oh, and it’s free.
Where can I learn more?
Read the official announcement post.
Try chatting with Pi for free.
3. Stability AI’s new image-to-3D collab
What’s up?
Stability AI has partnered with Tripo AI to offer a model called TripoSR that can construct 3D objects from a single image.
Why should I care?
TripoSR is insanely fast compared to competitors like Common Sense Machines, Luma Labs, and Meshy, generating 3D output in half a second on NVIDIA A100. At the same time, the quality is above other open models like OpenLRM.
Where can I learn more?
Read the announcement post.
Access the codebase on GitHub.
Try it for free on Hugging Face.
Check out my quick test where TripoSR somehow manages to make an already disturbing Halloween costume even more unsettling:
4. Midjourney makes “/describe” more useful
What’s up?
Midjourney launched a new version of the /describe command, optimized for V6.
Why should I care?
If you ever wondered what text prompt was used to create a given image, /describe can help. Simply upload the image and it will suggest several text prompts to recreate it. The new version is even better at returning nuanced descriptions in natural language. I once argued that getting help from chatbots is one of the best ways to make great images. The /describe command brings the power of a chatbot directly into Midjourney.
Where can I learn more?
Follow the official Twitter / X thread.
5. Haiper enters the text-to-video scene
What’s up?
Last October, I looked at 6 text-to-video sites. Now we have one more in the mix: Haiper.
Why should I care?
Haiper definitely ain’t Sora. But, unlike Sora, you can use Haiper right now and it costs nothing. You can generate 2-second HD videos or 4-second regular-quality videos. Haiper can also animate an existing image (like many other video models), and you’ll soon be able to extend your videos or repaint them in different styles.
Where can I learn more?
Read the deep dive by Techcrunch.
Try making short clips for free at Haiper.ai.
Check out this smiling koala I asked for:
6. Wix’s AI site builder is finally here
What’s up?
After first teasing it in July 2023, Wix has now released its AI-powered site builder.
Why should I care?
The AI website builder helps people create a business-ready website from scratch simply by describing their needs to the chatbot. After the site is ready, AI can also tweak the theme, layout, and so on. It’s a quick way to get started for beginners.
Where can I learn more?
Visit the official landing page.
Try the AI site builder for free on Wix.com.
7. Pika’s Lip Sync is now for everyone
What’s up?
Just a week after releasing Lip Sync to Pro users, Pika made the feature available to all.
Why should I care?
If you want to test Pika’s Lip Sync without spending money on a paid account, now’s your chance.
Where can I learn more?
Check out the announcement post on Twitter / X.
Try Lip Sync for free on Pika.art.
8. Adobe Express comes to your smartphone
What’s up?
Adobe is bringing its Adobe Express app to Android and iOS devices (in beta).
Why should I care?
The app puts the power of Adobe’s generative AI features like the Firefly text-to-image model, Generative Fill, and more right at your fingertips. This lets you easily edit your images or videos for, say, social media directly from your phone.
Where can I learn more?
Read the announcement post.
Download the app for iOS or Android.
9. ChatGPT on the web can now talk (at you)
What’s up?
ChatGPT’s responses now have a clickable “Read Aloud” button.
Why should I care?
Until now, ChatGPT voice features were only available on the smartphone app. While there’s no voice input yet, “Read Aloud” is the first step toward having something similar on the web. It can be useful for listening to ChatGPT-created stories, improving interactions with the chatbot for visually impaired people, and more.
(“Read Aloud” is also available on the mobile app, but the app already had voice chat.)
Where can I learn more?
Check out the announcement thread on Twitter / X.
🤦♂️ 10. AI fail of the week
I asked Hapier for a robot winking at the camera. Now I’m scared to sleep:
⚔️ Sunday Showdown (Issue #2) - Riffusion vs. Suno: Who can write the best short jingle?
Today, we journey into the world of songs and funky beats.
Not so long ago, I made MusicFX, MusicGen, and Stable Audio compete in an epic Battle Of The Bands II.
The thing is, all of them are purely instrumental music models. They don’t do lyrics.
But several tools out there can write songs and sing them out loud.
Let’s find out which of them can help us with a catchy jingle.