Never, or 10 Minutes From Now, Whichever Comes Sooner: What Computers "Can't" Do, Part 3 Million and One
Part 1 of an excellent interview with Baadur Jobava is available on the ChessBase website, and includes a number of Jobava's best and most interesting games. The games are terrific, and I would also endorse his thoughts about self-training, though I don't think self-training makes having a coach pointless. (Just ask players like Karpov, Kasparov, Kramnik, Carlsen and many others about this.) That said, one should not use a coach for what one can do for oneself.
There is no question but that Jobava is a very strong, and very creative player, and the move 15.d5 which he played against Evgeny Bareev (see the link above) was a great and subtle novelty. Jobava is justly proud of it, but when he says of this that "the silicon monster can never find such an idea" he will have to add his name to the long and prestigious list of people who have grossly underestimated what computers in general and chess engines in particular can do. I have no doubt that evaluating 15.d5 as a strong move was way beyond the capability of chess engines back in 2004, when the Bareev game was played, but nowadays Stockfish puts it as co-number one within ten minutes, while Komodo takes a bit longer before making it the top choice. It does take a while and one will miss it if they have an itchy trigger finger over the space bar, but finding it in ten minutes, compared to the years it took just one solitary grandmaster to find it, is a mere blink of an eye.