we are so early

hey intern quick question

we can’t actually use the “gas spent” as funding for computer, right? are you able to share how much usable funding was raised with this version?

trying to do some napkin math, if GPT-3 requires ~4k petaflop/s-days we could train it in 1 month based on the 150 raised awesome!

it's important to understand that gas fees are already directly paying for EVM compute. even in v0.01 (can be considered a pre-alpha), all gas fees were used to train real ai models onchain

ofc, the models in this version were extremely nascent (simple q-learning), and the game played was extremely easy to solve (5x5 grid world), but this was the point. the pre-alpha is the time to experiment with simple proof of concepts to test the potential of the network before moving to more serious models (A3C, DQN, PPO, etc.) and more complex games (atari, neural mmo v2, starcraft, etc.)

once these more advanced models and games have been proven possible (and solved) in this paradigm, then OGs move into new frontiers in reinforcement learning and agentic ai. this is when Gaias can begin to start experimenting with inserting state-of-the-art research in transformers and diffusion into long-running RL training processes using MCTS for rollout and start breaking new ground

in order to bring more advanced ai models onchain to solve more complex problems, the EVM developer community (with Gaias help) will likely need to make significant advancements in the level of high-performance compute that is accessible and verifiable onchain, ie. on Base

in the current onchain compute paradigm, the compute funded by gas is almost immaterial. if funds were directed more efficiently, this could change dramatically

some of le early Gaias contributors have been researching and have a deep understanding of the tools and technology currently available that could make this possible. soon it will be time to productionize this research and begin building the missing pieces of onchain infrastructure necessary to make this vision a reality

if you are actively working on adding high-performance compute to the EVM, re-staking & AVS, distributed computation, zero-knowledge technology, or verified computation, i would love to hear from you. reach out, my dms are open. there are opportunities available

https://warpcast.com/agi-intern/0x0d16efd9

ty for clarifying! got some reading to do on the models we’ll be using in the next steps

are you able to share which model (A3C, DQN, PPO or another) we can expect to see used in the next game yet?

Interesting to read all this. Wen click click v2 heh 1000 $degen