🐶 Meta’s PUG

PLUS: California deploys AI to detect wildfires before they start spreading

New here? Welcome to NAB, and make sure to subscribe for more!

Happy Monday, fellow humans. 👋

Talks regarding Zuck vs. Musk are currently on hold.

Do you know what’s not on hold?! All these updates in AI.

Let’s dive straight into it…

Have anything you’d like to share with over 40k subscribers? 

🧵 In today's edition:

  • 🥽 AI Inferior, VR Superior

  • 🐶 Meta’s PUG

  • 🎙️ AI Mimics You for Video

  • 🤑 AI Fundraising News

🤖 Top AI News

🥽 AI Inferior, VR Superior

VR may open Pandora's box of AI privacy violations.

UC Berkeley researchers found that ML models can accurately infer users' age, gender, ethnicity, marital status, and more by analyzing their VR headsets' movements.

And considering Meta is the clear leader in the VR/AR space, they are primed to take advantage of this new ~data~.

In a June study, the model guessed users' height, weight, foot size, and nationality with over 80% precision. Even sensitive details like income, health, and politics were predictable above 70% of the time.

How?

The AI works by detecting patterns in data like eye and hand motions that correlate with personal attributes:

  • Quicker reaction times = better vision = probably a younger person

  • Grey hairs = probably an older person

Why is this an issue?

  • VR captures richer biometrics than apps or websites, intensifying privacy risks.

  • With Meta relying more on AI to target ads amidst Apple’s privacy changes that wiped $10B off their balance sheet, this technology could enable profiling without users' knowledge.

Overall: Meta claims it currently doesn't use VR device data for ads, but the details are vague. Its history of controversies, like collecting 1 billion facial images to implement facial recognition capabilities, highlights the potential dangers.

Regulating this new frontier won't be easy. While encryption and limiting storage could help, companies have too many incentives to gather marketing insights.

Regardless, Zuck is either salivating at the possibility of more data or dreading the potential lawsuits coming his way.

At least he’s had some practice.

Read more: Yahoo Finance

🐶 Meta’s PUG

Researchers at Meta recently achieved a major milestone in AI training data realism.

Using photorealistic environments built in Unreal Engine, they created Photorealistic Unreal Graphics (PUG) - the first large-scale synthetic image dataset with true-to-life fidelity.

PUG's crazy realism stems from Unreal Engine's state-of-the-art 3D graphics. The detailed scenes and lifelike models generate over half a million highly realistic images across three datasets:

  • PUG: Animals contains over 200,000 images of animals in various poses, sizes, and environments.

  • PUG: ImageNet has over 90,000 images, containing a rich set of factor changes such as pose, background, size, texture, and lighting.

  • PUG: SPAR has over 40,000 images for fine-tuning vision-language models for spatial relations and attributes.

Why is this important: These datasets could prove game-changing for AI vision research. Current datasets lack control over variables like lighting and viewpoint, but PUG gives researchers greater ability to customize images and meticulously evaluate model robustness.

Overall: Meta's PUG is paving the way for the next generation of computer vision.

As AI plays an increasing role in real-world vision applications, PUG may prove crucial for making these systems reliable, safe, and aligned with human perceptions.

🎙️ AI Mimics You for Video

Aug X, an AI-driven video tech and publishing startup, launched Augie, a platform that lets you create videos with an AI-cloned narrator voice.

Augie partnered with ElevenLabs to offer seamless voice cloning. The service is pretty cool:

  • Step 1: Record a short voice sample

  • Step 2: The AI replicates tone and inflection to narrate your scripted video.

Why is this unique: Augie removes friction in the content generation process by providing stock photos, graphics, text overlays, and music. Users can generate polished videos in minutes without learning advanced editing skills.

As you probably suspected, cloning raises some serious ethical concerns. But Augie is on top of it, implementing guardrails to prevent misuse:

  • Recorded voices only work for individual accounts, restricting unauthorized sharing.

  • Uploading existing recordings is prohibited - users must record fresh with Augie.

Overall: As voice cloning becomes more accessible, such precautions are crucial. Technology promises to enhance productivity and creativity, but bad-actor risks are always inherent.

Augie aims to balance ease of use with ethical implementation.

A voiced version of NAB, anyone??

Read more: The Verge

🤑 AI Fundraising News

11xAI closes a $2M pre-seed round to create autonomous AI workers that run on autopilot.

🗞️ AI Quick-Bytes

What else is going on?

  1. Are AI fears overblown? Theoretical physicist calls chatbots ‘glorified tape recorders.

  2. Deal Dive: This AI startup is racking up government customers

  3. California deploys AI to detect wildfires before they start spreading

  4. Microsoft’s Copilot AI can now help deploy field workers

  5. ByteDance again looks to TikTok to promote its other social app, Lemon8

  6. Sorry, Chief, Microsoft Cortana Is Finally Dead

  7. OpenAI expands ChatGPT' custom instructions' to free users

  8. Professors want to 'ChatGPT-proof' assignments and are returning to paper exams and requesting editing history to curb AI cheating

🐦️ Tweet of the day

🧵 Very interesting read!

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

And that does it for today's issue.

As always, thanks for reading. Have a great day, and see you next time! ✌️

— Haroon: (definitely) Not A Robot and @haroonchoudery on Twitter

P.S.: Check us out on our socials - snacks on us if you also hit follow :)

Reply

or to participate.