Does any engine consider this move?

[jefk]

it can also be a *book* move at move 5.
More rare than the Shirov/Zappa gambit (move 7), but nevertheless.

I will call it the Semi-slav Hilby gambit after the one who played it first at
master level. https://uschesschamps.com/player/craig-hilby/

Later in 2024 GM William played it three times and lost only once,
against Kramnik; even although Kramnik didn't play the (still new move) 5...dxc4!N

The -modern- idea of early g4 gambits (usually in d4 openings) is described by the
Russian GM Kryakvin in a book for New in Chess.

The earlier you play g4, the worse it basically is, imo.

[Werewolf]

All that's happening here is that Sadler has weakened Lc0 to play a sub-optimal move.

It's not that he's discovered a hidden gem we all had missed. If you run Lc0 BT4 in Nibbler you'll also see it finds g4 but it makes it clear it loses around 10% winning chances. The advatage of doing it this way (rather than modifying the engine) is that you still "discover" g4, and you see it's not THAT bad, but it preseves the realism of telling you you're basically picking the 10th best move.