@kimmonismus Bad retrieval is probably why half these agents think in circles. Do you think the 1B version is already enough for long-term memory, or does the 8B actually earn the extra VRAM?
@Teknium Kimi’s 1M context feels almost built for Hermes. But did it actually stay on track better during long runs or did it mostly just remember more?
A 14 year old Chinese VALORANT player made $10,000 betting against himself through his older brother.
His teammates defended him because he was only 14.
Meanwhile, his brother kept betting against the team, using Claude for quick bets in live mode.
For three matches, every missed trade, late rotation, and terrible peek looked like a rookie struggling under pressure.
By the time they realized what was happening, the matches were over
@DivyanshT91162 At 271 skills, the next skill Hermes needs is one that tells it which skills are actually worth installing.
Curated bundles for coding, research, and content would make this repo even better.
A 14 year old Chinese VALORANT player made $10,000 betting against himself through his older brother.
His teammates defended him because he was only 14.
Meanwhile, his brother kept betting against the team, using Claude for quick bets in live mode.
For three matches, every missed trade, late rotation, and terrible peek looked like a rookie struggling under pressure.
By the time they realized what was happening, the matches were over
CS teams will soon practice against AI clones of their opponents.
Feed the agent enough match footage, and it can recreate a team’s rotations, executes, and reactions.
Instead of watching how your next opponent plays, you’ll play against a simulation of them.
Again and again.
Until match day feels like a rematch
@0xVirtanen By that logic, demos are useless too. You’re not recreating the same rounds - you’re learning the habits behind them. Strats change overnight. Tendencies usually don’t.