OpenAI Shows ChatGPT’s New AutoGPT‑Style Plugins, Browsing, and Code Interpreter

OpenAI co‑founder Greg Brockman demonstrated new ChatGPT capabilities, highlighting three new modes: Plugins, Browsing, and Code Interpreter.

Chat interface directly adds to cart, posts to Twitter

The Plugins mode lets ChatGPT connect to external apps. Using the DALL·E plugin, it generated a dinner‑scene image from a simple prompt. With the Instacart plugin, a single command created a complete shopping list and added items to the cart. By adding the Zapier plugin, ChatGPT automatically posted a tweet with the shopping‑cart link, all without the user opening Twitter.

In the TED‑style demo, a tweet was sent directly from ChatGPT, showing complete trust in the AI’s output.

The Browsing mode adds internet access for fact‑checking. Brockman asked ChatGPT to verify the publication dates of two OpenAI blog posts; the model performed step‑by‑step web searches, identified the correct dates, and corrected its earlier answer.

Brockman: It’s time to understand AGI

In a follow‑up interview, Brockman discussed why OpenAI, despite its smaller team, leads AI progress, emphasizing deep‑learning breakthroughs, collaborative research, and the belief that scaling models will eventually achieve general artificial intelligence.

"We’re standing on the shoulders of giants, seeing progress in compute, algorithms, and data across the AI industry," Brockman said.

He also addressed safety concerns, arguing that transparent, step‑by‑step development and open collaboration are essential to avoid dangerous outcomes.

Code Interpreter analyzes 30‑year AI paper dataset

Using the Code Interpreter mode, Brockman uploaded an Excel file containing metadata of 167,000 AI papers from the past 30 years. ChatGPT first clarified column meanings, then offered three analysis options: a histogram of author counts per paper, a yearly publication trend line chart, and a word‑cloud of paper titles.

Histogram of authors per paper – reveals typical team sizes.

Yearly publication trend line – shows research growth over time.

Word‑cloud of titles – highlights common research topics.

After selecting the options, ChatGPT generated the visualizations, even adjusting the 2023 data point as a forecast when the year was incomplete.

The demo illustrated how ChatGPT can now perform complex, multi‑step tasks across external services and data analysis, bringing AutoGPT‑like functionality to everyday users.