TOGETHER WITH

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

• Don’t watch the news. Use GraphCast for weather updates ☀️ ☃️

  DeepMind’s new weather AI outperforms the industry’s gold standard.

• NVIDIA’s new AI chip is a tank 🦾 🤖

  The H200 is almost 2x faster than the H100, its previous GPU.

• Do you make AI vids on YouTube? Got some bad news for you 🤬 🔥

  It updated its guidelines to regulate AI-generated content.

 Sponsor Bot Eat Brain | New here? Subscribe!

MAIN COURSE

AI > Weathermen ☀️ ☃️

A couple of weeks ago, we covered Google DeepMind’s update for AlphaFold. Its advancements in protein folding could directly lead to new drugs.

Last Friday, we reported on DeepMind’s “Levels of AGI” framework. It’s a new method to categorize and measure AGI progress.

Do you have some more DeepMind dirt for me?

Absolutely.

On Tuesday, Google DeepMind unveiled GraphCast. It’s an AI model that generates accurate 10-day global weather forecasts in under a minute.

How good is this thing?

GraphCast only takes a minute for calculations that take previous models several hours.

It outperforms the gold standard, The High-Resolution Forecast produced by the European Centre for Medium-Range Weather Forecasts.

It beats it in both accuracy and efficiency.

‘Murica. 🇺🇸

How does it work?

GraphCast is trained on historical weather data, which helps it understand complex weather patterns.

It predicts weather 6 hours into the future and can extend these predictions up to 10 days by recursively using its own forecasts.

What makes it different from other models? 

It only needs the weather conditions from 6 hours ago and the current state to predict future weather.

It shows proficiency in predicting extreme weather events, and providing early warnings for storms and heat waves.

So, all other models are trash?

Google DeepMind suggests it should supplement, not replace, traditional forecasting methods.

It also lacks the robust uncertainty estimates that current models provide.

How do I check it out?

Here’s the open-source code for GraphCast on GitHub.

SPONSORED BY GAMMA

Ideas are beautiful. Present them that way.

Ditch the design drama.

Instead, transform even the most intricate of your ideas into stunning slide decks in just minutes with Gamma.

😰 Too many ideas, too little time?

Say goodbye to late-night slide struggles. Let Gamma turn your ideas into visual masterpieces.

🚀 Speed up your design process.

Quickly create captivating presentations.

📱 Go beyond static slides.

Every slide is interactive and mobile-responsive by default.

🎨 No design degree? No problem.

Gamma does all the heavy lifting, so you look like a design pro.

And it’s a web app - so no downloads.

The best part? It’s free.

BUZZWORD OF THE DAY

SIDE SALAD

NVIDIA’s new AI chip is a beast 🦾 🤖

In September, we reported on NVIDIA and Anyscale’s partnership. The goal was allegedly to speed up the building, training, and deployment of AI models.

In October, we covered NVIDIA’s Eureka. It’s an open-source AI agent that creates algorithms to teach robots new skills.

More NVIDIA goodness?

You bet.

On Monday, NVIDIA launched its new HGX H200 AI Chip. It offers significant advancements in generative AI and high-performance computing (HPC).

So, it’s bigger and stronger?

Oh, yeah. Compared to the A100, the H200 nearly doubles memory capacity and offers 2.4 times the bandwidth.

It also delivers almost 2 times faster inference speeds on LLMs than the H100, its previous GPU.

Inference speed is the time it takes for a model to make a prediction or decision based on input data.

What’s different about this one?

The H200 integrates HBM3e memory technology, a first for a GPU. This provides faster and larger memory capacity, with 141GB at 4.8 terabytes per second.

💪🤓

Isn’t there a shortage of AI chips?

Very much so. NVIDIA plans to triple its production of its H100 GPUs in 2024.

It didn’t mention how many H200 GPUs it intends to make.

A LITTLE SOMETHING EXTRA

YouTube hates your AI content 🤬 🔥

In August, we covered YouTube’s Music AI Incubator. It partnered with Universal Music Group and announced its new AI music strategy.

Last week, we reported on YouTube’s new AI tools. It’s testing a comment summarization tool and a conversational AI tool.

On Tuesday, YouTube announced updates to its content guidelines. It claims the purpose is to regulate AI-generated deepfakes and synthetic media.

What are the new guidelines?

Creators must now disclose if their content is AI-generated, especially when it significantly alters or synthesizes material.

Videos that fail to disclose AI-generated content may face removal or demonetization.

Is that it?

Nope. 

YouTube will remove videos that realistically simulate identifiable individuals, considering factors like parody, satire, and public figures, in line with fair use and defamation laws.

Huh, what about AI music?

Now, there’s no allowance for parody or satire in AI-generated content mimicking an artist’s unique singing or rapping voice, reflecting the need to maintain good relations with the music industry.

How will I know if I’m watching AI-created content?

YouTube is introducing tools “over the upcoming month” that inform you if the content is synthetic.

It will appear in the description of the video.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Tools

Balodana — Custom-made women’s clothes for work or play. We use AI to guarantee you a perfect fit, every time. [Sponsored]

Insanely Fast Whisper — transcribes 5 hours of audio in a couple of minutes.

Earnbetter — an AI-powered job search tool that writes resumes and cover letters.

Room AI - an AI-powered interior designer who gives you advice to enhance your room’s aesthetic.

Think Pieces

Does AI understand memes? How AI’s processes of creating memes from the ground up differs from how humans do it.

Which LLNs hallucinate the most? Here’s the Hallucination Leaderboard on GitHub.

Who won the AI Startup Draft? Cerebral Valley Podcast host pick their top 5 AI startup picks.

Startup News

Google announced protection measures for Bard users. It protects users from AI scammers, fraudulent claims of copyright law, and more.

Atlas, a 3D generative AI platform, raised $6 million in seed funding. The funding consisted of $4.5 million from 6th Man Ventures, and $1.5 million from Collab+Currency.

Meta broke up its responsible AI team. It moved the key players to its generative AI product team.

Research

Story-to-Motion — generates human motion from text for character animation in gaming and film.

Music ControlNet - Carnegie Mellon University and Adobe Research’s diffusion-based music generator.

ChatAnything — generate personas from text inputs that you can FaceTime with.

TWEET OF THE DAY

OpenAI’s ex-President heartfelt tweet after the firing of Sam Altman and the resignation of 75% of OpenAI staff as a result.

Click here for a play-by-play overview.

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

AI ART-SHOW

Until next time 🤖😋🧠