There is an H2 heading between "Lightweight JIT compiler project goals" and "The current state of the MIR project". That h2 contains the Hoover Dams image and has an id generated from empty content because the content is just image: <h2 id="-h2"></h2>. When you scroll near that image, the javascript tries to update the url but that broken ID caused it to set the empty ID ("#"), thus jumping to top.

You can verify this by open the table of contents bar and see an untitled section. Whatever CMS they are using, it didn't handle the empty H2 correctly. Few additional considerations:

1. The article author should never produce an H2 with only an image in it.

2. The image should have an `alt` label to be accessible.

3. CMS should use `alt` label or file name to generate the ID when the H2 has no other textual content.