🤖 AI Can Speed Read Harry Potter

PLUS: The EU is Closer To Regulating AI

small-banner

What's up? You're reading Inclined AI. You gotta love that Friday Feeling.

Here's the vibe today:

  • Anthropic drops 100K Context Windows 😲

  • EU pushes the AI Act further

  • More Google I/O takes

  • A look at why AI can’t save Snapchat

CLAUDE IS NOW THE WORLD’S BEST AI SUMMARIZER

Anthropic expanded the token context for their conversational AI, Claude.

They lept from 9K tokens to 100K tokens. For context, a token is a way to measure the number of words a model can consider and recall. For example, 100,000 tokens are ~75,000 words.

Let me ask you this: how fast can you read the first Harry Potter? Keep in mind that the average person can finish it in 5 hours. 

You’re in that ballpark, right? But did you know that in AI terms, Harry Potter is around 100K tokens in length?

Claude can now digest that entire book and respond to you in 22 seconds.

That’s the magic of AI.

This new context window is a huge deal.

Researchers can find key points in long, technical papers without breaking a sweat. Even business analysts can upload an entire 10-K from a corporation and watch Claude break it down for them.

We’ve never seen a context window this large available to the public. There’s the rub because Anthropic will only release this as part of Claude’s API service. 

You can join the waitlist here, but the odds are that only businesses and AI researchers will get to play with this for now.

Token restrictions are now a game of haves and have-nots. Even OpenAI is gating who can access GPT4_32K, the largest context window for their model. 

You love to see these companies make these competitive leaps, but it will be a second before it trickles down to public use.

THE EUROPEAN UNION IS GETTING CLOSER TO ENACTING AI RESTRICTIONS & REGULATIONS

Prompt: robot on Euro trip Style: Selfie

It’s not law yet; the committee proposed this act before ChatGPT launched. So an update was necessary before they brought the legislation to the floor for debate.

We’re getting into the political weeds with this one, but the essential ideas are worth considering:

1/ There are clear, uncrossable boundaries

The EU will not tolerate any AI that uses biometric surveillance to track its citizens in this proposed law.

They also draw the line at predictive policing, but some member states argue that the proposal speaks too generally about that space and risks people’s safety.

Expect that part of the AI Act to receive a ton of debate.

2/ The EU wants AI companies to slow down

Google, OpenAI, Microsoft, and all the others launch models and papers without any regulation to hold them up. This new law would change that.

Instead, every new AI tool must undergo standardized risk assessments and open up some proprietary data. In other words, big tech must disclose any copyrighted material used to train models going forward.

3/ Transparency & Citizen’s Rights above all else

The committees want to create a database of every AI tool in the EU to show people how AI uses their data and what info they collect.

That goes all the way down to general-purpose tools, so any AI that influences your day-to-day is in that database.

Overall, the new edits show that politicians understand the weight of how fast AI is moving. The deadline to pass the AI Act is April 2024.

Fingers crossed, it doesn’t take that long, but this is the bureaucracy we’re talking about.

They don’t run at the speed of a New York coffee shop during the morning rush.

Quick Nuggets

🤔 Another take on how the Google I/O event went

📞 Android phones are about to be awash in AI tools

👻 Snapchat can’t save itself from itself with AI

⭕️ Reshape your customer experience with the help of AI

🚂 An Industrial Revolution-level leap is how this MIT data measures ChatGPT

📍 Acupuncture & AI are closer to each other than you realize

🔍 Google’s Search will never be the same now

👀 Take a look at Raycasts new AI features in the Pro subscription

❓ Should you be worried about robots replacing you at work?

👂 Bots listening to bots is a big problem for Spotify 

🧹 Princeton uses GPT to make a robot clean up after them

📼 Rewind, an AI startup, receives over 170 offers during their raise

🐁 A handsfree gaming mouse is coming around the corner

💵 DeepMind Cofounder wants the government to consider UBI

💤 Boring! Google paints a picture for a tamed generative AI future

🌴 Check out PaLM-2’s full tech report for details on their new model

🔥 Fresh Products

  • GPT4All - run a chatbot locally on your computer (link)

  • DocArray - organizes your data for you for ML projects (link)

  • Fine-Tuner - build an AI agent with no code (link)

  • Metabob - debugs and refactors your code (link)

  • PaLM-2 - is available on Bard right now (link)

  • Steppit - shows you how to plan a course (link)

  • Golem - chat with a beautifully designed LLM (link)

  • HelloAI - mobile chatbot with a prompt library included (link)

  • R.Ai- personalized email responder (link)

  • MyFit AI - creates a unique training program (link)

  • stelvio app - offers unique art styles for consistent work (link)

  • Octavia - your crypto AI assistant that runs on the blockchain (link)

  • ThumbnailMaker - make YouTube thumbnails instantly (link)

  • Inbox Narrator - summarizes your inbox & reads it out to you (link)

  • BooksAI - you can tell a book by its cover (link)

  • Roughly - artistic freedom for creating with AI (link)

Good Content, AI AI

@bushbush492

Google I/O 5/10/2023 #ai #news #aiwars #google #googleio #artificialintelligence

In case you thought I was joking about how often Google said AI yesterday. This is one presenter.

That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.

Check Out Our Sunday Edition

Every week you can dive into more AI news and topics with us by subscribing to our premium edition.

We’ve written about:

If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial so you can read this one. That’s how excited I am to post it.