
fate amenable to change @hospitable // @OmniscopeAI // @BananaDev_
Kardashev Scale
Joined September 2018
- Tweets1.3K
- Following124
- Followers632
- Likes1.2K
Opus 5.5 > GPT 6.1 Sol... BUT if you have something like a JOURNAL.md for the agent to frequently write notes/progress, Opus 5.5 can give you a great start and then GPT can perform well after reading the notes. Not really coordinator/orchestrator. More like shoulders of giants.
1
47
i thought rlhf was sophisticated... sure, the algorithm and math are cool, but the high-level conceptual implementation is so simple it's almost disappointing
Q-learning and PPO explanation by gpt 6.1 sol in codex from the actual learning runs in progress by opus 5.5 in claude code
zooming out a bit, this still feels like mind blowing magic sometimes, compared to where we were less than 5 years ago
1
116
learning reinforcement learning fundamentals by watching + guiding claude code while it trains a bot to do musty flicks in rocket league
hardstuck plat for too long, so I'm vibe coding a Rocket League ghost coach
opus 5.5 doing well so far, got a ghost octane with boost rendering in game + proof of concept step-by-step mechanic mode
1
67
Hey, that鈥檚 us! @hospitable
Look ma, I鈥檓 on TV!
1
124












