1 man, 3 countries, and a $1B company
And, more on Google's MAGI & MiniGPT
Barun Sharma & Barun Pandey
April 21, 2023
GM! Welcome to The Status Code.
We're like stealthy jaguars, quietly stalking the frontier of AI progress for you to discover.
Here’s what we have for today:
✨Google’s new MAGI trick
🖇️Image models: MiniGPT & LLaVA
🧏♂️ 1 man, 3 countries, and a $1B company
(Estimated reading time: 3 minutes 20 seconds)
Not subscribed yet? Stay in the loop with AI weekly by reading the Status Code for five minutes.
Two Headlines
Two main stories of last week. If you have only ~2 minutes 10 seconds to spare
1/ ✨Google’s new MAGI trick
Google is cooking up a fresh project called Magi.
Could India's beloved Maggi noodles inspire it? We'll leave that to your imagination.
This project aims to be a full-fledged app, unlike Google's "experimental" BARD. The New York Times suggests we can expect it to launch in May. So, what makes Magi stand out?
1/ It uses Google Earth's mapping tech
2/ GIFI, an AI image generator, is part of the deal.
3/ It connects to payment systems.
Since November 2022, Google's been feeling the heat from ChatGPT, which led them to kick things up a notch.
Now they have spring rooms and spaces to test Magi's various settings in engineering, design, and other departments.
Many think Magi may have been a planned project that shifted gears.
Bing is the gear stick here.
Samsung announced that it was slowly changing its default search engine to Bing. And Apple is in talks to replace Google as well.
Google pays $15B to Apple annually to keep its search engine by default.
To tackle their rivals, Magi aims to:
1. Be an improved version of BARD
2. Outperform BING.
3. Utilize Google's search data effectively.
If Magi can answer at least two of these questions, it might be more successful than ChatGPT.
However, not everyone on Twitter is convinced.
Some argue that Google has made big promises before but only delivered BARD - which remains experimental.
On the other hand, some optimistic folks believe Magi could change the SEO game by replacing the traditional 10-result format with a better follow-up message function.
Marketers are already prepping SEO hacks for this yet-to-be-released technology. We wonder if Google will survive the challenges every week, only to face new problems the next day.
Yesterday, Sundar Pichai announced that Google's Brain and Deepmind AI teams are Google DeepMind now.
It's highly likely they'll work on Project Magi. This collaboration could be a game-changer for Google.
2/ 🖇️ Image models: MiniGPT & LLaVA
OpenAI's GPT-4 announcement had everyone buzzing about text-to-image models. But before we could even catch our breath, MiniGPT swooped in and grabbed the spotlight.
There are two components of MiniGPT.
BLIP2 and Vicuna
BLIP 2 is a multimodal pre-training method.
And Vicuna runs in the back. It helps you run it on your computer.
But it also means it’s for open-source use only.
People want AI that's easy to use and set up. MiniGPT delivers just that, making it perfect for research.
As experts in AI say, the stronger the research, the better the implementations.
Speaking of bigger, they're already working on a 7B model.
Now let's talk about LLaVA
LLaVA is a language and vision assistant with better OCR skills.
If you’ve seen data entry jobs, you get paid to record sponsor timings in videos by analyzing each frame.
When LLaVA becomes sufficient, it will replace those jobs completely.
For a first attempt with language-only GPT-4, it's pretty impressive. And the best part? It learns without human input!
People are loving these alternatives to GPT-4. It just shows that innovation doesn't always come from the expected source.
And what's up with the names?
One Trend
1 trend you can pounce on. Reading time: ~1 minute 10 seconds
🧏♂️1 man, 3 countries, and a $1B company
A boy born in Jordan in 1983 moved to Bangladesh and later to the UK with his family.
That boy grew up and became a hedge fund manager.
Today, he's the CEO of a top Image-Text Model company.
It is the story of Emad Mostaque, founder of Stable Diffusion.
Emad started Stable Diffusion in 2020. He had a goal of creating realistic and creative images.
They had:
2 employees
Generated 10,000 images/per year
Worked with 1,000 users
Made 0$ in revenue
Fast forward to today, the company now has a valuation of $1 billion.
Big investors, including some from the Y-Combinator board, are showing interest.
This week, they released their new StableLM language model.
It may seem like a step back, going from text-to-video to a more text-focused model.
But let me explain why it's a big step forward.
About 70% of AI companies start with building text models before moving to multimodal models.
Why is Emad doing it?
Emad aims to develop a large model, testing it with 3 trillion tokens.
This gives them a higher chance of making progress than a company that started with a text-only model.
A report from PWC shows that companies using multiple AI models have a 30% higher success rate than those relying on a single model.
The complexity of image and video data for LLMs provides better data than text models.
In the end, it's all about collaboration and growth.
Emad Mostaque's story demonstrates the power of innovative thinking and collaboration.
They have achieved impressive growth and attracted significant investor interest.
Everyone is saying to make AI accessible and open-source.
If this makes you love open-source even more, check out this petition.
🦾What happened this week
MIT released an article about why OpenAI is using data
Reddit demands payments for AI trained on its users
Drones navigate environments using liquid neural network
MRI Brain Images Just Got 64 Million Times Sharper
ARTEMIS, a soccer-playing humanoid robot, is ready for the pitch
9 Useful resources to get more out of Generative AI
Use ClassifyAI to sort data with a personalized classification model
Humane gave a demo of their AI-powered wearable projector
Jira and Confluence get Atlassian Intelligence
Microsoft is designing a new chip called Athena
💰Funding Roundup
Oso-AI raised €10M In Series A Funding
Blockchain company CryptoGPT raised $10M in funding
FairlyAI raised $1.7M in funding
AI-based vision inspection startup SwitchOn raises $4.2 million
Singaporean start-up Groundup.AI raised $1.8M seed funding round
Anthropic calls for $15 million in funding to boost risk assessment
Responsible AI governance firm Trustible comes out of stealth with $1.6m
SoundHound Raises $100M in Strategic Financing
🐤Tweet of the week
engineers when PMs say “we can build this using chatGPT”
— gaut (@0xgaut)
Apr 19, 2023
😂 Meme of the week
That’s it for this week, folks! If you want more, be sure to follow our Twitter (@CallMeAIGuy)
🤝 Share The Status Code with your friends!
Did you like today's issue? |
