🦾 Nvidia creates an AI to train AI

ALSO: The Olympics are using AI… and DeepMind can creates sounds

Estimated Read Time: 5 minutes

One of the biggest obstacles for creating AI is the training. You need massive volumes of data and tons of resources and energy. Many companies are actively running out of new training data to improve their models (such as Adobe). Nvidia thinks the solution is… to create an AI which creates training data.

• 🦾 Nvidia announced their open AI model.

• 🫠 Microsoft recalls their “Recall” feature… (yes I made that joke).

• 🏃‍♂️ The Olympics are using AI to protect their athletes.

• 📢 DeepMind is developing sound generating AI.

• 🎥 TikTok adds AI avatars for advertisers.

🦾 Nvidia announced their open AI model

What happened: Nvidia just announced their new AI model, the Nemotron-4 350B. It’s open, free, and available now.

The details: 

• The Nemotron is more than a single LLM, it’s an entire family of LLMs designed for commercial and industrial use (medical, finance, etc).

• The Nemotron is not a normal LLM, it’s designed to create synthetic training data to train other LLMs.

• The instruct model creates diverse synthetic data mimicking real-world scenarios. The reward model filters and grades synthetic data on various quality attributes. (We’ll have to wait for real world tests to find out how effective it really is.)

• This is massive because training data is one of the most expensive parts of creating an AI. And it has a limit… you can run out of training data. The Nemotron hopes to solve both of these problems: cost and data quantity.

• It uses an open license which you can check out here. This means Nemotron is free and accessible. You can also download Nemotron here.

  

Why it matters: Nvidia clearly wants to cement their position outside of just providing AI chips. And a model which creates data for training other models is definitely unique and useful. Hopefully this solves some real problems… but I’m a bit skeptical how realistic or useful the training data will be. It needs to be accurate or else we’ll get more Gemini/Google AI problems.

🫠 Microsoft delayed “Recall” after tons of security and privacy concerns. In case you missed it, Recall was already hacked before its release and has been the topic of privacy concerns. It was supposed to launch this week… but that’s on hold for now. The feature will NOT be shipping with Copilot PCs as Microsoft is revising Recall to make sure it “meets our high standards for quality and security.” Sure dude.

👮 OpenAI hired an former top US cyberwarrior (expert) and intelligence officer and Retired Army Gen. Paul Nakasone to their Board of Directors. His job is to protect them from “bad actors”, AKA hackers.

✉️ Yahoo’s been adding AI features and will soon add AI summaries to the homepage. Among the features they’ve added are AI in their email client (in beta for the US) which highlights what it believes is the most important information and emails you’ve received. This includes an AI created inbox holding your most important emails. All that being said, I don’t think Yahoo’s going to get more users. The brand is just… to outdated for younger generations.

🏃‍♂️ The Olympics are using AI to protect their athletes

What happened: The International Olympic Committee just announced they will use AI in order to block any abusive comments on social media directed at the thousands of athletes and staff for the Olympics in Paris next month.

The details:

• AI will monitor and erase abusive social media posts directed at 15,000 athletes and officials. (So it’s censoring to create a safe space.)

• A big reason for this is the war in Ukraine. Yes, athletes from Ukraine and Russia will both show up. And the Olympic Committee is expecting a lot of online harassment for these groups.

• Russian and Belarusian athletes will compete as neutral athletes, not under their national flags. (Geez.)

• They're expecting over 500M social media engagements involving the Olympics this year.

• The specific access athletes need to provide for this monitoring was not disclosed. (Some might refuse.)

  

Why it matters: I’m actually concerned if they can censor effectiviely using AI. That’s a dangerous technology. Of course, it sounds great when we’re talking about protecting athletes and public faces from trolls/assholes. But this kind of tech could easily be abused for all sorts of purposes. Let’s see how effective this AI is.

📢 DeepMind is developing sound generating AI

What happened: DeepMind announced the development of a V2A AI model which generates audio from a video input.

The details: 

• The model DeepMind’s V2A tech takes the description of a soundtrack paired with a video to create music, sound effects and dialogue. 🫢

• DeepMind even claims the audio will match the tone of the video scene. But right now there’s problems with realism and quality.

• They insist the V2A model won’t be released to the public until it undergoes safety tests. (I don’t think they’ll ever stop deepfakes though).

Why this matters: If done well, this could be the missing piece in AI generation and allow AI generated videos to be fully complete from the start. As of now, AI generated videos create silent videos, they need sound effects to be added separately. V2A tech is the logical next step.

🎥 TikTok adds AI avatars for advertisers

What happened: TikTok announced they’ll be offering generative AI avatars and creators for branded content and ads on the platform.

The details:

• The AI avatars will be able to scale the user’s likeness to create multilingual AI avatars to reach a global audience.

• There are also “Stock Avatars” built off the appearance of paid actors and will be available for commercial use.

• Creators who make an AI avatar will be able to control the rates and determine who uses their avatar. So you could make money from this.

• There’s also an AI Dubbing tool which allows you to translate your content into multiple languages.

• These features are part of TikTok’s Symphony, their suite of AI-powered advertising tools.

• Yes, TikTok still might get banned if parent company ByteDance doesn’t sell them.

Takeaway: I’m a bit skeptical if this is a good thing. I can already see someone creating an AI avatar of someone else without their consent. Also it could misrepresent products depending on what’s allowed. I foresee a crap-ton of drama coming from this. Maybe TikTok will serve as a good testbed for these issues. (Hollywood recently suffered through strikes over AI taking away jobs and using their likenesses.)

🤖 Killer Robots!

Source: Midjourney

Give us your feedback!

Got feedback? Rate today’s newsletter by clicking below! 

Got more personal feedback? Fill in the box when you rate us and give us your criticism and feedback.

Thank you for reading! 

❤️Share the (ARTIFICIAL) Love!

Got a friend who wants to appear as smart as you? An AI fanatic or someone shaking in their boots of getting replaced? Share us by clicking the link below and you’ll get our love ;)