SLOW LLM: Chrome Extension Delays ChatGPT to Fight AI Dependency

New York-based artist Sam Lavigne has released SLOW LLM, a Google Chrome browser extension that deliberately stretches AI chatbot response times to frustrating lengths. The project is framed as a tool against growing over-dependence on generative models.

The SLOW LLM project interface. Source: slowllm.lav.io

What Is SLOW LLM and Why Was It Built

The idea came to Lavigne after watching his students and acquaintances rely on generative tools for even the most basic tasks. The open-source application currently works with ChatGPT and Claude.

An "enterprise" version is also available via DNS servers, supporting a broader range of models. This deployment option can throttle chatbot speeds across an entire local network — suitable for educational institutions and corporate environments.

How It Works Under the Hood

SLOW LLM does not interact with neural networks directly. Instead, it exploits a JavaScript mechanism by overwriting the Fetch function responsible for delivering data to the browser. When a user visits a chatbot domain and submits a prompt, the modified function stretches the response delivery over an extremely long period.

Why It Matters

The project taps into a growing debate about AI's impact on human capabilities. In an interview with 404 Media, Lavigne said that many people are outsourcing their cognitive and emotional functions to AI tools, forgetting fundamental skills they once mastered. He believes that as LLM reliance grows, so does the trend of skill degradation.

Lavigne argues that chatbots have taken the concept of eliminating "friction" to an extreme. They offer to outsource every difficulty to "Silicon Valley's thinking machines" — even when overcoming those obstacles is what makes human creativity meaningful and valuable. Anything that removes friction from solving hard problems prevents learning and erodes existing knowledge, according to the artist.

So far, no one has reported successfully deploying SLOW LLM in workplace or educational networks, Lavigne noted. He joked about testing the extension on unsuspecting users, though he conceded they might "quit their master's program in rage."

Lavigne's Track Record of Digital Sabotage

The New York artist has a history of provocative digital performances. During the COVID-19 pandemic in 2021, he released Zoom Escaper — an application designed to help users "escape" video calls by flooding their audio stream with irritating sounds.

In 2018, Lavigne scraped publicly available LinkedIn profiles to build a database of ICE agents. The incident sparked controversy, and the published information about government employees was subsequently removed from GitHub and Medium.

A frequent collaborator of Lavigne's, working under the pseudonym Tega Brain, has released several browser-based utilities including Slop Evader — a filter that removes "junk" generative content from search results by excluding anything published after November 2022, when ChatGPT publicly launched.

Lavigne describes his work as "small experiments in digital sabotage," aimed at creating tools that marginally disrupt computational systems. He does not consider himself an anti-technology maximalist, however — he admitted to using the Claude chatbot while writing the code for SLOW LLM itself.

Earlier in March, analysts at ActivTrak published research concluding that rather than lightening workloads, artificial intelligence has so far only accelerated and complicated workflows, increasing the overall volume of employee tasks.