AI Breakfast
Posts
OpenAI Launches "Tasks"

OpenAI Launches "Tasks"

AI Breakfast
January 15, 2025

In partnership with

Good morning. It’s Wednesday, January 15th.

Did you know: On this day in 2001, the first edit was made to Wikipedia?

In today’s email:

OpenAI’s “Tasks”
Security Flaw in Prompting Exposed
OpenAI Adjusts Position on Political Neutrality
3 New AI Tools
Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with GAMMA}

The future of presentations, powered by AI

Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.

Today’s trending AI news stories

'Tasks' pushes ChatGPT toward true assistant status

OpenAI has rolled out Tasks, a beta feature that edges ChatGPT closer to a true AI assistant by enabling scheduled actions, recurring reminders, and cross-platform notifications. Tailored for Plus, Team, and Pro users, Tasks integrates seamlessly into ChatGPT’s workflow, offering contextual suggestions drawn from previous chats. While notifications are accessible across devices, the task manager remains limited to the web interface.

Today we’re rolling out a beta version of tasks—a new way to ask ChatGPT to do things for you at a future time.
Whether it's one-time reminders or recurring actions, tell ChatGPT what you need and when, and it will automatically take care of it.
— OpenAI (@OpenAI)
8:40 PM • Jan 14, 2025

This move has reignited speculation about Operator, a rumored AI agent poised to expand ChatGPT’s operational range into document retrieval, data synthesis, and advanced problem-solving. OpenAI positions ChatGPT as a compelling alternative to tools like Asana and Google Calendar, with a focus on streamlined functionality and adaptive intelligence. Read more.

AI security flaws exposed by simple prompt manipulation

Image: Microsoft

Microsoft’s security team has tested over 100 AI tools since 2021, revealing that the most effective attacks are often the simplest. Forget gradient calculations—prompt engineering is where the real disruption happens. In one case, the team bypassed an image generator's safety protocols by embedding harmful instructions in plain image text. Their findings emphasize the continued necessity for human judgment, particularly in evaluating biases and sensitive scenarios, such as interactions with users in distress.

The surge in AI's integration into daily applications has uncovered new vulnerabilities, with language models now being manipulated to execute convincing fraud schemes. Read more.

OpenAI Outlines U.S. AI Strategy and Adjusts Position on Political Neutrality

Image: OpenAI

OpenAI recently published "economic blueprint," a living document that outlines a strategy to strengthen U.S. leadership in global AI development. It calls for targeted federal investments in hardware, data ecosystems, energy infrastructure, and workforce development, while highlighting bureaucratic obstacles in policies like the CHIPS Act that slow progress.

The document also advocates for scaling renewable and nuclear energy to meet data centre demands and proposes copyright policies that allow AI systems to train on publicly available data. Instead of rigid mandates, the blueprint suggests voluntary compliance pathways.

It also quietly removed references to AI models being “politically unbiased by default.” Initially included to address AI’s political neutrality, the line has been omitted in the latest draft. An OpenAI spokesperson cited the revision as an effort to streamline the document, while noting that other company resources, like the Model Spec, still maintain a stance on objectivity.

Critics, including Elon Musk and David Sacks, have argued that AI models, such as OpenAI’s ChatGPT, lean towards liberal ideologies, particularly on sensitive topics like immigration and climate change. Despite this, OpenAI maintains that any biases in its models are bugs to be corrected, not inherent features. Read more.