10X AI (Issue #37): Bard Upgrades, Midjourney News, Arc Search, and a Rocket Man

PLUS: Amazon's new shopping assistant Rufus, Shopify's upcoming AI image editor, Meta's Code Llama 70B, and more AI voices.

Happy Sunday, friends!

Welcome back to 10X AI: a weekly look at generative AI news, tools, and tips for the average user.

Let’s get to it.

🗞️ AI news

Here are this week’s AI developments.

1. Bard gets better

Google is staying in the AI headlines for the second week in a row.

This week, Google’s Bard chatbot has gotten a few serious upgrades:

Gemini Pro shenanigans

Bard is now powered by Google Gemini Pro not only in the US but in 40+ languages and 230+ countries and territories. (See the full list.)

In related news, Gemini Pro itself soared in the Chatbot Arena rankings to become the second-best-rated model after GPT-4-Turbo.¹

Imagen 2 shenanigans

Bard can now also generate images using Google’s impressive Imagen 2 model.

But funnily enough, Bard didn’t get the memo yet.

Here’s Bard gaslighting me:

If you’re not using Bard, you can test Imagen 2 in Google’s “AI Test Kitchen,” where it joins TextFX and MusicFX under the name ImageFX.

Imagen 2 also joins DALL-E 3, Ideogram, and Midjourney V6 in the growing category of image models that can spell:

2. Midjourney releases Niji V6 and Style Reference

Midjourney is also staying in the news with two updates of its own.

Niji V6 is here!

The V6 version of Niji—Midjourney’s anime-tuned aesthetic—is now available.

You evoke it by adding --niji 6 to your prompt.

Here’s a “family celebrating,” Niji style:

Get consistent styles with “Style References”

Midjourney also added a new --sref parameter:

Source: Midjourney Discord

This works a bit like the classic image prompt or the /blend command, with one crucial difference: It only borrows the style of the reference image, not its subject.

I did a quick test. Here are a few “vanilla” V6 results for the following prompt:

“painting of a weird spaceship”

Now watch what happens when I take the following cartoon image…

…and use it as a Style Reference via the new --sref parameter while keeping the above text prompt the same:

Quite an impact, right?

Midjourney successfully applies the cartoony style of our reference image to the spaceship.

(To see what horrors await when you use the reference image as an “image prompt” instead of just a “style reference,” refer to today’s fail of the week.)

You can add more than one reference image and even define the strength of each image’s influence, giving you plenty of control over your final result.

3. Meta’s new coding model

Meta AI released Code Llama 70B, which rivals some of the top performers on coding tasks, according to several LLM coding benchmarks like HumanEval and MBPP.

It comes in three versions:

• CodeLLama 70B: the foundational model

• CodeLLama 70B - Python: specializes in Python (you’d never guess)

• CodeLLama 70B - Instruct: fine-tuned to follow natural-language instructions

As with all Meta AI releases, Code Llama 70B is open-source and can be freely downloaded.

4. Arc Search…the future of search?

The Browser Company launched a new iOS app called Arc Search.

One of the app’s most impressive features is “Browse for me,” which takes AI-powered search a step beyond Perplexity and SGE. Once you type in your request, “Browse for me” researches relevant websites, consolidates its findings, and then creates a brand-new page centered around the topic:

I admit: This is making me regret being an Android and Windows user. (I’ve been on the waitlist for the Windows version of their Arc browser since I wrote about it last October.)

If you’ve got an iOS device, you should already be able to experience Arc Search for yourself. I’d love to hear what you think about it!

5. Amazon’s Rufus helps you shop

Amazon announced an expert AI shopping assistant called Rufus (yeah).

Rufus is “trained on Amazon’s extensive product catalog, customer reviews, community Q&As, and information from across the web.”

You can ask it any question related to your shopping, and Rufus will help:

If it works as advertised, Rufus should be quite useful, especially for high-complexity, high-cost purchases.

Rufus is currently in beta and available to a subset of US mobile customers (are you one of the lucky ones?). Over the coming weeks, more US shoppers should get access.

6. Shopify teases a smart image editor

After launching Sidekick in the middle of last year, Shopify will soon add an AI media editor to its Shopify Magic suite of tools.

The editor will let sellers easily remove, replace, and generate new backgrounds for their product images, as shown in this video (skip to the 16:58 mark if that’s not where it opens):

The new image editor is expected to come out somewhere in spring.

🗣️ AI voices

I’ve got three more “AI voices” for you this week.

Remember, if you have specific burning questions you’d like the wider Substack community to answer, this is your chance to ask them. Then I’ll try to incorporate the answers into future editions.

7. Nick Potkalitsky uses GPT-4 as a writing assistant

Here’s

Nick Potkalitsky

:

“I'm an enthusiastic ChatGPT-4 user and highly recommend its upgrade for exceptional text editing and refining. My writing is categorized into "assisted" and "unassisted" processes. "Assisted" writing is for routine tasks needing initial thought but not intensive polishing. Using AI here saves me significant time, allowing me to focus on more complex tasks.

In the "unassisted" mode, I ensure my draft is about 85% complete before letting ChatGPT-4 enhance it. This method focuses on the logical flow of ideas rather than perfecting every word. Give it a try and see for yourself! Time is an invaluable asset. AI not only helps us amplify our thoughts but also preserve this most precious resource. ”

Check out Nick’s Substack:

8. Patrick Jordan recommends Perplexity and Poe

Here’s

Patrick Jordan

:

“A year into using AI apps every day, here are the two I use most often:

Perplexity is fast, cites its sources, and lets me organize responses in collections. For research on any cybersecurity topic, its speed and concise answers are golden. Perplexity can access the web, so it's always up to date, which is crucial when e.g. looking up the latest adversary attack techniques.

Poe gives me access to the latest models from OpenAI, Google, Anthropic, Mistral, and more. In just a few clicks, I can switch between GPT-4, Claude 2-100K, Gemini Pro, Palm, Llama, and so on. I generally recommend getting answers from multiple models and keeping up with their new capabilities. Poe lets you do this seamlessly!”

Check out Patrick’s Substack:

9. Tita Costa uses AI for visual art and enjoys Krea.ai

Here’s

Tita Costa

:

“As an artist, I utilize available AI tools to create visual art and films. Currently, I find myself using Krea.ai extensively, especially because it allows me to create my personal models. Using the "AI Training" feature, I can teach Krea by feeding it my traditional paintings. I can then use it to render other paintings in the trained style.

For years, I dedicated my time to traditional painting, eventually developing my own distinctive style. With the integration of AI models, I've found a significant acceleration in my project timelines. This has reignited my interest in the arts, and I'm enthusiastic about exploring this new phase of artistic expression.”

Check out Tita’s Substack:

🤦‍♂️ 10. AI fail of the week

When using a reference image for your spaceship picture goes wrong…

Sunday poll time

(If your answer is “Other,” drop a comment below to let me know which.)

---

Previous issue of 10X AI:

1. For some solid related coverage, check out Lies, damned lies, and benchmarks and Google Is About to Overtake OpenAI