Bot Eat Brain
Posts
Meta's metaverse makeover

Meta's metaverse makeover

PLUS: ChatGPT unbanned in Italy

Reece Atkinson & Anthony Castrio
May 01, 2023

Good morning human brain and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

Meta unveils Codec Avatars: realistic, real-time avatars that replace their old “Mii”-style metaverse avatars. 👨🏼
Italy unbans ChatGPT after OpenAI added features to prevent unlawful data collection. 💬
TAM is SAM + video, an open-source tool that can add or remove objects from a video in just one click. 📹️
CatGPT is just like ChatGPT, except it mostly responds with cat-gifs and meows. 😸

Sponsor Bot Eat Brain | New here? Subscribe!

APPETIZER

ChatGPT returns to Italy after ban 🇮🇹

OpenAI has been allowed to bring ChatGPT back to Italy after taking measures to prevent unlawful data collection and ensure underage users cannot access inappropriate content.

The service was banned by the country that invented pizza in early April due to concerns over how they collect and store data which should be protected by the EU’s strict privacy laws.

The message shown to users in Italy

What this means: OpenAI's making nice with Italy is just the start of bigger fights for the company, as it now faces the possibility of investigations in several other countries.

BUZZWORD OF THE DAY

Voice recognition

Voice recognition enables computers to understand and interpret human speech allowing users to dictate commands or input text by speaking.

FROM OUR PARTNERS

Build & Deploy AI Models Quickly

Cerebrium makes deploying ML Models seamless. Deploy & Fine-Tune your model without worrying about infrastructure.

1/ Serverless GPU Deployment - <1s cold start

2/ All Major ML Frameworks Supported (Pytorch, Onnx, XGBoost, Etc).

3/ Don’t have your own model? Access 15+ pre-build models.

4/ Fine-tune FlantT5, GPT-Neo, Stable Diffusion, and more.

MAIN COURSE

Return of the metaverse? Meta reveals realistic human avatars 👀

Remember Zuck’s cringy metaverse demo from last year? The presentation was roasted for featuring avatars that looked like the Mii’s from Wii Sports.

Mark “I used to be human too” Zuckerberg showed off an early version of Meta’s metaverse product in 2022.

Well, those “Mii’s” just got a big update. Meta’s latest update to their avatar-rendering engine Codec Avatars looks much more human. They are not able to render realistic avatars mirroring your appearance and movements in real-time.

The new generation of Codec Avatars

Facebook is working on bringing their avatars to your phone, but to release an app that works for everyone they’ll first need to build a huge dataset of people covering every possible hairstyle, skin tone, and style.

Hot take: The metaverse is still cringe… it’s just slightly less cringe now. Give it another 6 months to a year and you might find this built into messenger video calls.

A LITTLE SOMETHING EXTRA

Track Anything Model: an open-source video editing tool from the future ✂️

Remember SAM? The Segment Anything Model released by Meta early in April makes it possible to crop out any object from a photo in just one click using an AI trained to detect millions of different kinds of objects.

That’s cool, but what about videos? Now meet TAM: the open-source Track Anything Model that extends SAM for video.

SAM + TAM in action: one-click removes the parkour athlete (but leaves his shadow behind).

Just like SAM, all it takes is a couple of clicks to select objects in your video, remove them, replace them with other content, or extract them for use in other videos.

When is this hitting Hollywood? Listen, at the level of big-budget movies traditional techniques are still going to dominate. This isn’t about saving Hollywood money, this is about making your zero-budget special effects 90% as good for 0% of the cost.

What to expect: Snapchat & TikTok filters galore. AI-powered features in the next iOS camera app update.