LeCun said Facebook AI had created and open sourced SEER, a billion-parameter self-supervised computer vision model that he described as proven to work efficiently with complex, high-dimensional image data.

Judge as of 2021-03-04. Tested parts: Facebook AI created SEER; it is a self-supervised computer-vision model; it has about one billion parameters; it was open sourced by the date of the statement; it handles complex, high-dimensional image data. True: Facebook AI's SEER announcement or paper (around March 2021) shows a self-supervised CV model of about 1 billion parameters (about 1.3B counts) and a public code or model release (e.g. the VISSL library or released weights) by 2021-03-04. Mostly true: everything holds but the open-source release was of the training code or library rather than the full model, or had a small timing gap. Mixed: the model existed as described but nothing related had been open sourced by that date. Mostly false: SEER existed but differs in a key qualifier (not self-supervised, far from a billion parameters). False: no such Facebook model existed. The "works efficiently" part counts as met if the published results show strong performance on large image datasets such as ImageNet or Instagram images.

Set before the research began.

1. E1 Facebook's new computer vision model achieves state-of-the-art performance by learning from random images VentureBeat · 4 March 2021

   “Facebook claims to have made a step toward this with a computer vision model called SEER, which stands for SElf-supERvised. SEER contains a billion parameters and can learn from any random group of images on the internet without the need for curation or annotation.”

2. E2 Facebook's new computer vision model achieves state-of-the-art performance by learning from random images VentureBeat · 4 March 2021

   “VISSL, which Facebook is open-sourcing today, allows for self-supervised training with a variety of modern machine learning methods.”

3. E3 Facebook's new computer vision model achieves state-of-the-art performance by learning from random images VentureBeat · 4 March 2021

   “That said, Facebook doesn't plan to share the images or the SEER model itself, in part because the model might contain unintended biases.”

4. E4 Facebook's new computer vision model achieves state-of-the-art performance by learning from random images VentureBeat · 4 March 2021

   “When trained with just 10% of the examples in the popular ImageNet dataset, SEER still managed to hit 77.9% accuracy.”

An independent report from the announcement date says Facebook built SEER, a self-supervised computer vision model with a billion parameters (E1). It also reports strong ImageNet results, 77.9% accuracy when trained on only 10% of the labelled examples (E4). That meets the "works efficiently" part. However, what was open-sourced that day was the VISSL training library, and Facebook did not plan to release the SEER model itself (E2, E3). The criteria call this case mostly true.