Adobe Trains Firefly with Competitors’ Data

When artificial intelligence visual creation tools first came out, they worked in a very amateur way. However, as large-scale data sets were fed into these tools, the results quickly increased. First from OpenAI DALL-Efrom Stability AI stable diffusion And midjourney tools stood out.

However, these tools often cause unauthorized processing of people’s or institutions’ data. went to court. Adobe has differentiated itself from this trio since its launch. firefly introduced. Since Firefly was trained with Adobe’s stock image library, it did not cause licensing problems and it was possible to use the outputs for commercial purposes. But Adobe, which said that it did not use its competitors’ artificial intelligence outputs to train its own model, was not telling the truth.

Adobe, which criticizes artificial intelligence models that feed on each other, also used artificial intelligence outputs in Firefly training.

Users can upload millions of images to Adobe’s stock image platform. According to Bloomberg, Adobe representative Michelle Haarhoff said that images produced by generative artificial intelligence Data set used in Firefly training confirmed that it constituted a small part. This corresponds to approximately 5 percent of the data set.

The visuals in the 5 percent slice in question are also Midjourney and DALL-E Created by competitors such as. But each of these images; They were examined in detail to see if they contained any brand, character or logo.

In such a situation, you expect criticism to come from outside, but it has been within the company from the very beginning. Whether the artificial intelligence model is ethical or not There were disagreements about. Since these discussions are confidential, the names of the sources are not given. However, criticizing competitors’ data collection methods, knowing deep down that this is the case, puts the company in a controversial position. What are you thinking?