- Bot Eat Brain
- Posts
- xAI launched a new AI model called Grok
xAI launched a new AI model called Grok
PLUS: generate AI images in 128 different styles
Michael Parrish
November 06, 2023
TOGETHER WITH
Good morning, human brains. Welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
Save $4 a month using Elon’s AI instead of GPT-4 👨⚕️ 🔥
xAI’s new AI model called Grok outperforms many models in its class.
Microsoft’s AI savagely roasts dead peeps 💀 ⚰️
Microsoft disabled its AI-generated polls on news articles. Here’s why.
Unlock more customization in your Midjourney images 👨🎨 🎨
Now, it allows you to generate images in up to 128 different styles.
MAIN COURSE
Elon’s AI is $4 less than ChatGPT 👨⚕️ 🔥
In July, we covered Elon Musk’s new AI company, xAI. It’s stacked with AI engineers from OpenAI, Tesla, DeepMind, and more. The company’s motto is to “understand the universe.”
A few days later, we reported on xAI’s live Twitter event (now 𝕏). The company unveiled a collaboration with Tesla, the use of Twitter’s data for AI training, and more.
More 𝕏 stuff?
You bet.
On Saturday, xAI launched its new AI model called “Grok.” It uses real-time world knowledge to wittily answer questions, suggest questions to ask, and more.
What the Grok?
The name comes from Robert A. Heinlein’s 1961 novel, Stranger in a Strange Land.
In the book, “grok” is a Martian word that means to empathize with others on such a deep level that you merge with them.
What’s the point?
Grok aims to assist humanity in gaining knowledge, supporting people from all backgrounds, and empowering research and innovation.
How is it different from the sea of AI models out there?
After only about four months of training, Grok-1 outperformed many models in its class on several benchmarks like GSM8k, MMLU, HumanEval, and MATH.
Huh?
GSM8k contains math word problems at a middle school level.
MMLU consists of multiple-choice questions in several subjects.
HumanEval includes code completion tasks, primarily Python.
MATH involves math problems at a middle and high school level.
How did xAI make this thing?
xAI built Grok’s infrastructure using a modern, high-performance AI language called Rust.
It also uses Kubernetes, an open-source AI system that allows you to automate, deploy, scale, and manage AI applications.
How do I use it?
Early access to Grok is currently being offered in the United States.
Elon Musk said Grok will be offered as part of the 𝕏 Premium+ subscription for $16 per month.
SPONSORED BY UIZARD
Design has never been easier.
Are you…
Spending hours in complicated design tools for simple mockups?
Waiting for designers or others to make time for your ideas?
Needing to improve your product or ship features faster?
Never leave an idea behind. Ditch the endless back-and-forth. Design UIs easily and in seconds with Uizard! You bring the ideas, AI does the rest.
Uizard is an AI-powered UI design tool for creating wireframes, mockups, and prototypes for digital products. Start from scratch, a template, or generate a multiscreen mock-up from simple text prompts with Uizard Autodesigner, which just got even better…
Introducing Autodesigner 1.5
You can now use the Autodesigner widget (new!!) at the bottom of any Uizard project:
Generate individual screens and images with text prompts
Use text, images, or links to apply new themes
Unleash your creativity and impress your stakeholders with stunning, interactive UIs
Going from idea, to mockup, to clickable prototype has never been so easy.
BUZZWORD OF THE DAY
Benchmark
A set of standards or tests used to compare the performance of different AI systems or models.
It helps to measure how well an AI can understand language, solve problems, or complete tasks against a known set of challenges.
SIDE SALAD
Microsoft’s AI mocks your death 💀 ⚰️
Last month, we reported on Microsoft’s Bug Bounty program. It’s awarding between $2,000 to $15,000 for reported security vulnerabilities in Bing Chat.
Last Monday, Microsoft disabled all AI-generated polls on its news articles. On an article about a woman’s death, it polled people about her cause of death.
Peep those comments:
Didn’t The Guardian write it, though?
The article, yes. The poll was generated by Microsoft’s AI.
Several readers shamed the author of the article and The Guardian in the poll's comments section, whom they assumed was responsible for the blunder.
I see. How did The Guardian respond?
The Guardian criticized Microsoft for the poll, demanding public accountability and the removal of AI-generated content from their journalism.
The Guardian’s CEO, Anna Bateson, contacted Microsoft, emphasizing the inappropriateness of the poll.
She demanded that Microsoft take full responsibility and compensate news publishers for using their content to train AI models.
Did Microsoft say anything?
“We have deactivated Microsoft-generated polls for all news articles and we are investigating the cause of the inappropriate content. A poll should not have appeared alongside an article of this nature, and we are taking steps to help prevent this kind of error from reoccurring in the future.”
A LITTLE SOMETHING EXTRA
Midjourney’s new Style Tuner 👨🎨 🎨
In October, we covered Midjourney’s new upscaling features. They allow you to enhance the resolution of your AI images.
On Wednesday, Midjourney introduced a new “Style Tuner” feature. It allows you to customize and edit your AI images with text prompts.
When will it be available?
It’s available for all paid Midjourney users to use now.
How does it work?
First, you type /tune along with your chosen prompt.
After that, Midjourney allows you to select between 16 and 128 pairs of “Style Directions” that you can refine to achieve your desired look.
Then, it will inform you about the time/cost to generate your styles.
Once submitted, it generates images in different styles.
These images also contain shareable codes that can be used to generate new images.
MEMES FOR DESSERT
YOUR DAILY MUNCH
Think Pieces
Should we stop AI from becoming conscious? How it’s hypothetically possible for AI to have feelings and why we should progress AI to that point.
How could an AI chip speed up genome assembly? It sped up the alignment of protein and DNA molecules 10 times faster than previous methods.
Why is Scarlett Johansson suing an AI app developer? Lisa AI used her name and likeness in an ad without her permission.
Startup News
Cohere released its new LLM, Embed v3. It claims to be a major industry upgrade for querying noisy real-world data.
Browser Brave unveiled Leo, an AI “safe and secure” AI assistant. It discards chats immediately and doesn’t require a login.
Ex-Google CEO, Eric Schmidt, poured $100 million into his girlfriend’s AI startup. He’s married, and its own employees called it a “vanity project.”
Research
MetNet-3 — Google’s new neural weather network that gives accurate 24-hour weather forecasts.
Making music with Diffusion models — how to guide it to produce more desirable, listenable outputs.
Generative AI Paradox — how AI models reproduce experts’ outputs, but don’t have an understanding of what they generate.
Tools
YouTune — allows you to fine-tune SDXL with images from YouTube videos.
Shiny — collection of tutorials to build, create, design, and more with AI.
OpenChat 3.5 — a 7B LLM that performs similarly to ChatGPT.
Beacon 2.0 — AI-powered all-in-one platform for creator-based businesses.
RECOMMENDED READING
Want to go deeper on AI?
Here at Bot Eat Brain we tend to skim the surface of many different topics. Meanwhile, our friends at Practical AI like to go all-in on one topic at a time.
Practical AI is a weekly newsletter written by Thomas Sorheim about Artificial Intelligence and its implications on business and work.
Sound good?
It’s free and you can unsubscribe at any time.
TWEET OF THE DAY
If you’re going to scoop a bagel, don’t do it in front of Grok. It claims it’s an “act of bagel cruelty.”
Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.
AI ART-SHOW
Until next time 🤖😋🧠
What'd you think of today's newsletter? |