Meta's AI copilot

PLUS: AI call centers

Good morning human brains, welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • Apple Bans ChatGPT ❌ 

    Apple is banning its coders from using AI tools like ChatGPT to protect company data. [jump to]

  • Meta’s AI programmer 🧑‍💻 

    Meta just announced a GitHub Copilot competitor exclusively for its internal teams. [jump to]

  • AI debt collectors ☎️ 

    Companies are using digital agents to make millions of phone calls. [jump to]


Apple bans ChatGPT ✖️ 

In a move to protect internal data, Apple is banning its engineers from using tools like ChatGPT and GitHub’s Copilot.

This is because OpenAI still currently trains its models largely on user data. In response, Apple’s playing a game of keep-away with some of the most advanced tech in the world. But they aren’t the first.

Samsung, Goldman Sachs, Wells Fargo, JPMorgan, and Verizon also restricted employees from using ChatGPT after registering multiple incidents of feeding proprietary company data to the chatbot. This is now officially a corporate trend.

One more thing: new job postings show Apple is developing its own generative AI models, but it’s keeping tight-lipped about what they’re going to be used for.

We’ve got our bets on Apple making some AI announcements during its upcoming Worldwide Developer Conference (WWDC) next month.

Our take: tech teams are still desperate as ever for 10x engineers. With Bard and Claude moving up the ranks, OpenAI’s going to need to come out with a privacy-first solution fast if it wants to stay #1 in enterprise.



Think of AI parameters like the building blocks of an AI model. Parameters are small pieces of information that the AI system uses to learn and make predictions.

For example, if an AI system is learning how to recognize pictures of cats, it might use parameters like the shape of the cat's ears, the color of its fur, etc. to make a guess about whether a new picture is of a cat or not.

More parameters can mean better predictions, but this isn’t always the case.


Chat & Create Content With Your Documents

Chat with your documents - Get more insights from your documents by chatting with them. Extract, create, and summarize content.

Create content with AI - Easily create content based on your documents. Create blog posts, reports, and much more.

Automated Workflows - Combine the power of AI with your favorite apps to automate your workflows.

Semantic Search - Search through your documents using natural language. Find the right document and content in seconds.


Meta’s AI engineer 🧑‍💻 

Meta just announced that it’s built its own AI coding copilot, and it’s not sharing. Here’s the scoop:

1/ Meta’s CodeCompose is similar to GitHub's Copilot and is used internally by Meta teams to work in Python and other mainstream languages.

2/ It’ll make suggestions in the form of helpful annotations as you type and can complete multiple lines of code / fill in entire large chunks of the program as it understands the project.

3/ The largest of several CodeCompose models Meta trained has 6.7 billion parameters. This is ~ half the number of parameters Github’s Copilot uses.

4/ CodeCompose was fine-tuned on Meta’s own code, including internal libraries and frameworks written in Hack (Meta’s own programming language).

5/ Meta says that thousands of employees are now already accepting suggestions from CodeCompose every week.

What this means: fragmented corporate AI models, and lots of them. Meta isn’t going to be the last giant to create in-house models to protect company data & avoid overpriced AI deals.

One more thing: GitHub and OpenAI are being sued in a class-action lawsuit that accuses them of allowing Copilot to regurgitate sections of licensed code without providing credit. Meta circumvents this by training on its own code.


AI debt collectors 📞 

The debt collection industry is replacing human collectors with AI callers.

Companies have started to use digital agents powered by GPT4 and text-to-speech tech that can make millions of outbound calls every day.

This makes collecting debts dramatically cheaper and quicker vs paying human callers. Good luck reaching a human agent to dispute their claims.

Using AI in debt collection adds another dystopian element - targeting poor and marginalized groups.

Quick flashback: when Amazon built an AI model to qualify resumes from engineers, it would auto-reject applicants from women’s colleges because it had learned to favor male applicants based on the company’s hiring history.

What this means: sprinkling AI onto the age-old process of pressuring people to pay up means more outstanding debts being collected… but also millions of lost jobs and hundreds more spam calls to your number (now with AI!).

Check out Google’s Duplex demo to see what this might look like soon.



Think Pieces

Market Analysis: will startups have a shot in the enterprise AI race?

Why the risk and benefits of brands using AI is closer than you think.

Startup News

Stability AI open sources its AI-powered design studio. raises $35M for its AI-based approach to app integration.


Can AI help with scientific writing yet?

A systematic review of Green AI research (because the effects of GPUs on the environment aren’t negligible anymore).


PixieBrix: LinkedIn Message Templates. Pre-fill a LinkedIn text input with templates stored in Google Sheets. [Sponsored]

Facia: facial detection and matching that’s 99.99% accurate.

GPT Assistant: bring ChatGPT to Discord.

Dewstack: create, manage, and host intelligent docs.

StealthGPT: create AI content that bypasses detection.


The market briefly dropped after an AI-generated explosion started being picked up as real news by Russian channels with millions of followers.


Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.