AI and celebrities: A winning combination?

GM! Welcome to The Status Code.

We're pure magicians; brewing AI news with a spell.

Here’s what we have for today:

1. ✍ Novel writer GPT and AudioGPT

2. 😲HuggingChat; the open-source ChatGPT

3. 😎 Celebrities in AI

(Estimated reading time: 3 minutes 30 seconds)

Not subscribed yet? Stay in the loop with AI weekly by reading the Status Code for five minutes.

Two main stories of last week. If you have only ~2 minutes 10 seconds to spare

1/ ✍ Novel writer GPT & AudioGPT

Scaling Transformer to 1M tokens

Guess what? Stephen King's worst nightmare is coming true: an AI that can write novels!

Remember BERT?

It's a natural language processing framework made by Google AI Language in 2018. It’s one of the most effective Transformer-based models out there.

Some folks made a way to increase BERT's memory and context length. They have increased it to 2 million tokens.

How did they do that?

New storage system and recurrent memory.

You see, models like GPT-4 use a self-attention mechanism.

Suppose you type in:

“The Phoenix Suns are interested in trading for Deandre Ayton”

GPT-4 analyzes tries to find relationships between:

• Phoenix Suns

• Trading

So, the quadratic complexity increases after you add different words and sentences to the mix. The computation mechanism grows and becomes hard to process.

Introducing the RMT framework!

RMT divides the input into smaller segments instead of processing them at once.

And they are executed in sequence. This makes RMT efficient in handling large amounts of input.

AudioGPT

AudioGPT is an I/O multimodal system that can generate and analyze spoken dialogues.

It is the next level for assistants like Siri and Google Assistant.

AudioGPT works in several steps:

1. Modality transformation

2. Task analysis

3. Model assignment

4. Response generation

What sets AudioGPT apart from past models?

First, it uses ChatGPT to understand human intentions and assign audio models. This means AudioGPT uses language models as an interface.

Second, it brings variety in data, including audio tasks, speech, music, sound, and talking head.

It can single out your crush’s part in the song in a noisy environment.

This is faster and covers more area than AssemblyAI’s speech models.

So, what do people think?

Researchers and teachers are excited about the recurrent memory model. ChatGPT-created curriculums were cool, but research papers are the next step.

Imagine feeding the model your entire life and asking questions based on that. Or upload a virtual version of yourself somewhere. The possibilities are endless!

2/ 😲HuggingChat; the open-source ChatGPT

Bard, Jasper chat, ChatSonic… did we really need another chatbot?

HugginFace is the #1 host of the newest LLMs and models in town.

But they wanted to shake things up.

So, what did they do?

They created HuggingChat, an open-source version of ChatGPT based on OpenAssistant.

What makes it stand out?

1. It uses a pre-trained domain expert model like ChatGPT.

2. It connects more than 400 task-specific models (and 24 tasks).

3. It integrates seamlessly with different models.

4. It can use APIs and do dynamic research.

5. It has code syntax highlighting.

HuggingChat is a bit edgy. CEO Clem thinks open-source alternatives to ChatGPT are crucial.

LLaMA's license prevents its use in enterprise applications on HuggingFace. But that's not a big deal, as new licenses can be added.

For coders, HC writes decent Python code. A user tested it on "Code a Snake game," and the results are impressive.

But there's a catch…

The code syntax highlighting is great, but the coding capacity is limited.

HC also faces the usual chatbot hallucinations.

What's next?

Creating HuggingChat apps.

OpenAI announced plugins a month ago, aiming to be app killers. HuggingChat is taking a similar approach.

Are they targeting the app store?

Probably not. But they're making the most of their core business. They are putting models and datasets to good use.

1 trend you can pounce on. Reading time: ~1 minute 20 seconds

😎 Celebrities in AI

Let's take a trip back to 2012, shall we?

For me, 2012 was the year for AI. Here's what happened:

• Deepmind (now Google Deepmind) was founded

• Watson (IBM's AI) went commercial

• Siri appeared on iOS devices

• Google showed off a neural network

Remember all this? Maybe not.

What you might recall is the world-ending scare(2012).

The hype was unreal, even bigger than the "Earth is flat" debate.

Celebrities fueled the belief.

Back then, AI was in its infancy. Only a few celebs knew about it, like Ashton Kutcher.

Spotted….

He invested in companies like Vicarious, now valued at $1B and backed by Elon Musk and Mark Zuckerberg.

Fast forward to 2023, and the landscape has changed. We've gone from robots like Sophia to virtual AI like Genesis.

The global pop culture market is worth $2.3 trillion in 2023. Celebrities can contribute to AI in three ways:

1. Investments

   We're talking about celebs with deep pockets here. Think The Weeknd or Joe Biden. Kutcher and his peers fit this category.

2. Influence Culture

   Celebrity marketing is powerful, so that they can raise AI awareness. Grimes, Elon Musk's ex, offered 50% royalties on any AI-generated song.
   
   She's the third to do this, following Holly Herndon and YACHT. Herndon made her voice available in 2021, but only to her organization members.

3. Contribution to industry

   Celebrities can create systems for healthcare, tech, movies, sports, or entertainment. Movie stars might promote AI for filmmaking, while sports stars could invest in AR/VR systems for training.

What lies ahead?

More celebs are adopting AI systems, mainly for content creation. They use it to manage careers, save time, and connect with fans.

Expect a boost in VR/AR experiences. Apple, quiet on AI so far, will reveal their AR headset at WWDC.

We'll see celebrity chatbots once they upload their personalities to a model.

Celebrities hold significant potential to influence the AI world as we look ahead. Now's a great time to support them as they embrace this technology.

🦾What happened this week

• OpenAI adds options to disable chat history

• Packy released a Positioning problem article

• A paper on the cookbook of self-supervised learning was released

• Siri founders predicted the year 2034

• Tweethunter for Instagram, AutoGPT-social was released

• Greg Brockman, OpenAI founder, gave previews of OpenAI plugins

• Accenture article on how generative AI will transform work

• Bloomberg reported Apple’s Quartz program

• Bard learned how to code

• Musicians like Grimes choose to make money from AI

• Yelp adds AI to its platform

• Superblocks released their OpenAI enterprise

💰Funding Roundup

• Sequoia is leading a Series A funding round

• Pinecone raises $100M in Series B funding

• Malaysian AI startup VOX raised $240K in funding

• AI workspace, Ctrl secures $9M in funding

• Cardiac Solutions Tricog secures $9M in funding

• Amazon funds $29M to Veo Robotics for human safety

• NIH received $6.2M in funding and plans to use it for cardio diseases

• Former SpaceX engineer raised $40 million to build portable reactors

🐤Tweet of the week

😂 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!