OpenAI’s new feature lets you create AI agents

Hello everyone! Here’s what you need to know about AI today:

👉 OpenAI just shipped a new feature that lets you create AI tasks

👉 AI systems are not a black box anymore

👉 Meta wants to layoff 3,600 employees due to AI advancements

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5.1 minutes

What’s going on: OpenAI has added a new feature to ChatGPT called "tasks". Now, you can schedule things like daily weather updates or reminders for when concert tickets go on sale, right in the chat. This is currently in a testing phase for people who pay for ChatGPT Plus, Teams, or Pro. To start using tasks, you can choose "4o with scheduled tasks" from the dropdown menu in ChatGPT. When you send ChatGPT a message in this mode, it will automatically create a task.

What does it mean: ChatGPT can now act more like a personal assistant, doing things for you at times you choose without needing to ask each time. It's a big step towards making AI more helpful in our daily lives, sort of like having a smart friend who remembers to remind you about stuff, and performs tasks on behalf of you.

More details:

• With this new feature, you can tell ChatGPT things like "send me a news briefing every morning at 7 AM" or "remind me about my brother's birthday next week."

• You can only manage these tasks on the web version of ChatGPT for now.

• Recently, OpenAI had teased that agentic AI would be a big part of its strategy this year, and predicted that AI agents will join the workforce in 2025.

• ChatGPT can now browse the web on a pre-defined schedule, but it won’t run continuous searches in the background and it is not able to make purchases.

What’s going on: Scientists at the University of Western Ontario have made a big discovery about how AI models like ChatGPT make decisions. They've used math to look inside the "black boxes" of AI, which means they can now see how these artificial neural networks make choices. They did this by designing an explainable neural network called “cv-RNN“ working on something called “image segmentation”, where the computer breaks down pictures into different parts. They started with simple shapes and gradually moved on to real pictures like a polar bear in the snow.

What does it mean: Imagine you are playing a game where you have to separate different colored candies. You might have a rule, like putting all the red candies together and the blue candies together. The neural network is like a super complicated candy sorter, but instead of colors, it can sort all sorts of things, like pictures of animals or sounds. Research like this is important because it helps us trust AI more. Before, AI models were famous for being like mysterious boxes where no one knew what was happening inside.

More details: 

• The team led by Lyle Muller published their findings in the journal PNAS. They not only worked with simple images but also showed that their method could work with more complex, natural images. (Read the paper)

• In another collaboration with Schulich School of Medicine & Dentistry, they even successfully connected their network to a living brain cell, creating a hybrid system that bridges artificial and biological neural networks.

• In a 2024 study featured in Communications Physics, Muller and colleagues created an "explainable" network capable of executing tasks like basic logic operations, secure communication, and memory functions.

👩🏼 A French woman was scammed out of €830,000 by an AI-generated Brad Pitt who convinced her he needed the money for cancer treatment.

📽 Synthesia, digital twin AI company backed by Nvidia, has raised $180 million in a Series D funding round, valuing the company at $2.1 billion for its B2B AI video platform.

🤖 Amazon is transforming Alexa into an AI agent capable of performing practical tasks and is working on overcoming technical challenges like hallucinations and improving response speed and reliability.

😔 Mark Zuckerberg announced that Meta will lay off approximately 3,600 low-performing employees, right after appearing in a recent popular podcast and saying midlevel engineers will be replaced by AI in 2025.

💰 Nvidia has invested $4 million in MetAI, a Taiwanese startup that builds AI-powered digital twins. This is Nvidia's first investment in a Taiwanese company.

💬 Private messages revealed in court filings show that Meta executives were intensely focused on surpassing OpenAI's GPT-4 “internally“ while developing their own AI model, Llama 3.

🎙 Google had to adjust its NotebookLM AI podcast hosts to prevent them from sounding annoyed at human interruptions by studying how team members politely handle interruptions.

🖥 AWS, Amazon’s cloud computing platform, has committed to a $5 billion investment in Mexico over 15 years and launched a new AWS Mexico (Central) Region. AWS won’t hire Mexican employees but will support roughly 7,000 full time jobs in the period.

🧬 Bioptimus, a French AI startup, has raised $41 million to develop an AI foundational model for biology, aka "GPT for biology", to simulate biological processes and enhance drug discovery and treatment development.

Asking AI for health advice

This prompt will force your favorite AI chatbot to act as a medical professional, provide specific and evidence-based recommendations, and cite credible sources to support the given advice.

Disclaimer: Never fully trust the answers generated by any AI assistant/chatbot. Use the information provided by them as a starting point or initial guidance and always double-check their answer.

Nuro - Software Engineer, Sensors

Oscar - Senior AI Engineer

Discord - Director of Machine Learning, Safety

Reddit - Machine Learning Engineer, Search (multiple levels)

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!