This brilliant browser tool purposely makes AI chatbots worse

A new browser extension just debuted that’s designed to be used in tandem with an AI chatbot. Its goal is to make the experience worse.

“Are you concerned that you or your loved ones might be experiencing a LLM-induced psychosis? Or participating in a massive de-skilling event? Or outsourcing cognitive and emotional functions to auto-complete?” designer Sam Lavigne asks in a YouTube video introducing his new product. “Then you should install ‘Slow LLM’ on your computer.”

Lavigne is an assistant professor of synthetic media and algorithmic justice at Parsons School of Design, as well as an artist and web designer. Slow LLM is his latest creation, and its entire purpose is to make using AI chatbots as excruciating as possible. Whereas ChatGPT might answer a simple prompt such as “What day is it?” in a matter of milliseconds, Slow LLM intervenes to stretch that out to the nails-on-a-chalkboard pace of 30 seconds or more. 

Slow LLM is a small yet eye-catching act of digital protest against the proliferation of AI tools, and it joins a growing catalog of similar efforts like “Your AI slop bores me,” a human-powered chatbot, and “Slop Evader,” a Chrome extension built to turn back the clock to the days before AI-generated search summaries.

Those behind these projects, which Lavigne describes as “tiny tools for digital sabotage,” know that they’re not exactly combating the wave of artificial intelligence and large language models. Instead, they’re asking users to reconsider how they’re using AI tools—and what they might be missing out on in the process.

Designing a tiny tool of digital rebellion

Lavigne is no stranger to creating tools of digital sabotage. His past projects have included “Slow Hot Computer,” a website that does exactly what the name suggests; “Zoom Escaper,” a tool to help you escape Zoom meetings by self-sabotaging your audio stream; and “The Good Life,” an email service that sends you 225,000 emails from the Enron email archive in chronological order. For Lavigne, these sites are all about turning the mission of most web developers on its head by creating more friction rather than removing it. 

In the case of Slow LLM, Lavigne explains: “Like many other people, I’ve been thinking a lot lately about the political, financial, ecological, and emotional effects of LLMs. I became interested in exploring ways to disrupt their usage, either as something that people might wish to do to themselves, or as a kindness to others.”

He decided that, to achieve this goal, it would be best to take a softer approach—making LLMs worse, rather than blocking them altogether. In a matter of days, he pulled together Slow LLM. 

The browser extension, available for Chrome and Firefox, currently works to stunt both OpenAI’s ChatGPT and Anthropic’s Claude. It functions like a football player intercepting a ball mid-pass. Typically, when a user chats with an LLM in a web browser, it sends that prompt to a server using a built-in JavaScript function named “fetch.” Slow LLM replaces fetch with Lavigne’s own version that buffers all network requests, making them appear to run at a trickle.

AI adoption feels effortless when it’s fast, easy, and convenient. Slow LLM invites reluctant users to reconsider their choices while watching ChatGPT painstakingly thread together a response that they could have just Googled.