I am a bit confused by the benchmark comparison they are doing. The comparison of a domain specific "LeJEPA" on astronomy images against general models, which are not explicitly fine-tuned on astronomy images seems misleading to me.

Does anybody understand why that benchmark might still be reasonable?

The comparison is against general models which are explicitly fine-tuned. Specifically, they pre-train their models on unlabeled in-domain images and take DINO models pre-trained on internet-scale general images, then fine-tune both of them on a small number of labeled in-domain images.

The idea is to show that unsupervised pre-training on your target data, even if you don't have a lot of it, can beat transfer learning from a larger, but less focused dataset.

In my opinion needing enough slack in a system certainly rings true for many jobs that rely on high intrinsic motivation (probably also academia). Slack is also something that is essential for creative jobs, where you need the space to think and come up with new ideas.

Looking at the levels of busyness the concept of requiring slack is still compatible for me. If I am at low levels of busyness (< 4) there should be capacity for planned slots of slack, where goals can be worked on with own pace, ad-hoc meetings and guidance from peers (to fulfil the SDT motivational needs).

Even though a search based on go.mod or .sum files should definitely result in more hits, the new vulnerability management should be way more efficient, because it is already filtered for more relevant hits and should thus contain more signal vs noise as it only returns at least transitively used vulns. Am I missing something here?