We are used to AI being a distant service. When you ask a question to a chatbot, your request travels over the internet to a massive data center, where a room-sized computer processes your prompt and sends an answer back. A startup named PrismML is trying to flip this model by shrinking artificial intelligence until it fits directly inside your pocket.
The company recently released a new version of its AI software, called Bonsai 2. It is a compressed version of an existing, publicly available model created by Alibaba. By using a clever reduction technique, PrismML shrunk this model to about one-tenth of its original size. It now occupies only 5.9 gigabytes of storage space. This is small enough to fit on many standard laptop computers and possibly high-end smartphones, meaning the AI could theoretically run without ever needing to talk to the internet.
Making big AI small
To understand how they shrink AI, think of a model as a massive, complex recipe book. The model is made up of billions of pieces of information called weights. These are the specific connections the AI learned during its initial training, and they determine how it responds to your inputs. In a typical model, each of these billions of weights is stored with a high level of precision, taking up a significant amount of digital memory.
PrismML uses a technique to simplify those weights drastically. Instead of storing every weight as a complex, high-precision number, they convert them into one of three very simple options: a positive one, a negative one, or a zero. It is essentially like converting a high-definition, uncompressed image into a simplified sketch. You lose a tiny bit of detail—the company reports about a two percent drop in performance compared to the original—but the file size becomes tiny. Because the math involved is now much simpler, a standard smartphone processor can handle the work that usually requires an entire server rack.
The shift toward running AI on your own device could change how we view privacy and cost. Today, every time you use a major chatbot, you are sending your data to a third party. If the intelligence lives on your phone, your data never leaves your device, keeping your conversations private by design. It also means you don't need a constant, fast internet connection to get smart answers. As these compression techniques improve, we might move away from the current standard where AI is a utility you rent from a giant tech company, and toward a future where it is a tool that permanently lives on the hardware you already own.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy