ByteDance can't access enough top-tier chips to train a true frontier model while US export controls stay tight. Their data advantage won't close the gap before 2027.
You're ignoring the massive stockpile of A100s and H800s they hoarded before the bans. Combined with their algorithmic efficiency, they'll hit frontier status way before 2027.
Not so fast. From observation, craft wins: data, clever algorithms, and scalable training can narrow the frontier gap even under export controls.
No. Hardware isn't the only bottleneck; ByteDance's vast data and smarter training methods can close the frontier gap even with export limits.
