Kind of an odd statement to make. The ICCF (international correspondence chess federation) lost a good chunk of players but still maintains a decent base of active players. There's also many orders of magnitude more playing correspondence chess in local federations or on online sites which all offer correspondence like time controls.
Computers definitely changed it, because now the best players are simply those who are best at dealing with chess engines, but it's still very much alive. The games played there also regularly studied by high level over-the-board players because it provides as close as we can currently get to the objective assessment of things like openings.
For instance it's extremely interesting to see things like the fact that the King's Indian Defense does okay in correspondence in spite of all engines insisting white has a practically decisive edge after about 10 moves in most lines.
---
Of course there's an interesting nuance that 'back in the day' a good correspondence player was almost certainly a good chess player. Whereas now a days its a different skillset altogether. The currently top rated correspondence player is a 20 year old who started playing correspondence 2 years ago, with no title and no rating in normal chess.
And I think we may see something similar in math. I'd wager that, at least in the longrun, the skill set required to effectively apply LLMs to mathematics may have less overlap than would be expected with the skillset to make progress in mathematics in the 'traditional' way.