Italy bans ChatGPT

PLUS GPT-4 is a reasoning engine

Author

The Alignment
April 03, 2023

Welcome to The Alignment. ChatGPT receives it's first taste of government regulation as Italy bans the platform. Also, according to OpenAI, users need to think of GPT-4 as a reasoning engine not a knowledge base !

As always if you enjoy reading our posts be sure to spread the word !

Here’s what we have lined up for you today -

  1. Italy bans ChatGPT

  2. GPT-4 is a reasoning engine, not just a knowledge base

  3. A sci-fi shot film create by AI

The ban on ChatGPT in Italy might come as a surprise to many, thinking that the country is being wary of AI doom. However, the reason for the ban is something more technical. Italy is concerned about privacy violations that ChatGPT might be committing by potentially accessing users’ personal data which is not allowed under GDPR (the EU’s data protection law).

ChatGPT is now geoblocked in the country and if Italy’s position is proven to be correct there is a real possibility that other European countries might start their own investigations soon. That would mean potential fines and restrictions for how OpenAI operates its services in the region.

But as always, creative humans have launched similar bots built on top of OpenAI’s API which don’t collect user data. Try out Pizza GPT for yourself !

If a small startup like Open AI ( ~ 400 people) is already gaining the attention of regulators, one can imagine the pressure on big tech as they make their bets on AI.

Most people think of GPT being a knowledge base over which semantic search is performed when users prompt it for information. Dan Shipper, who’s a prominent writer and founder of writer collective Every explains that it operates more like a reasoning engine.

Having attended a private Sequoia AI event where Sam Altman explicitly stated that a lot of the secret sauce on improving GPT-4 comes from understanding how users can make it “reason” better. This can primarily be done by providing as much of the right knowledge as possible with the correct “context”. And so if you want to achieve better reasoning from ChatGPT-like systems you need to feed it with the right knowledge.

Hence not too different from incumbent search systems, storing, tracking and retrieving information we consume, watch and read will be important. But now instead of keyword matching which leads to poor quality answers, we’ll be able to deploy semantic search over data. And because these systems are context aware, they will be able to draw out semantic relationships between our query and the knowledge that they are aware of and have access to.

This is one of the reasons why we have been seeing companies spin up domain specific GPT like experiences relatively fast. Be it Snap for social networking or Quizlet and Khan Academy for Education. Having amassed years of high context data these companies are now using it to their advantage to build their knowledge specific bots.

Around the industry -

Product Corner -

There are way too many cool products to showcase just one !

  • Enhance AI : Add AI Autocomplete to any website in 2 minutes

  • Wiseone : Enhance your online reading experience

  • Til AI : Ai generated podcast network

  • Littlecook : Ai generated recipes based on ingredients you want

  • Quilt : Accelerated financial modelling with AI

  • Refact : Another coding assistant hits the market

'Last Stand' is a sci-fi short film made by AI

All in 2 hours of work !

Here’s what the director, Hashem Al-Ghaili had to say about AI’s role in the movie making process -

“ None of it is real. It’s just a movie, made mostly with AI, which took care of writing the script, creating the concept art, generating all the voices, and participating in some creative decisions. …… This short film was created as a demonstration to showcase the potential of AI in filmmaking.”