@davepl1968 I'm not privy to normal agreements in this space. But it seems reasonable to propose terms that include a non-refundable portion up front, or a back out penalty to ensure both parties are acting in good faith.
@davepl1968@damianGray I wanted that luck so bad. I knew I needed a monitor but had no clue how to obtain such software. I wish I could tell my 14 year old self about how to get started in 6502 assembly.
@vlelyavin @karpathy I agree. Except for me its:
1. Tester
2. Architect
3. Product Manager
4. Designer
5. Project Manager
... for an intern that never sleeps.
@davepl1968@ID_AA_Carmack Indeed not an easy hello world. I started with cart pole. Are you doing RL on emulated game with screen grabs or using a camera + actuators?
I wonder how many epochs would it take for your agent to stumble on the Easter egg to jump to green levels with 46.
@ID_AA_Carmack@breakability@bnrgator Sounds like Rage was on the wrong side of the Pareto Principle. Sometimes it’s worth it to spend the extra time/energy/expense to drive for 100%, but should be done very strategic and judiciously. The fact that one of principals is saying that it wasn’t worth it is telling.
@Brucester007@davepl1968 I miss WinXP Explorer. Once it became Vistafied I was done. Modern Finder isn’t perfect, but I’ll take it over modern Explorer every time.
@ID_AA_Carmack Wouldn't it be the same as a Q learning loop and predicting the next_step, but paying attention to the watched agent's deltas to determine if reward is improved? I am assuming the offline RL gets real data eventually to validate the new leaning from synthetic. Real data wins.
@apirkle@Project2501_117@ID_AA_Carmack Man, seeing that brought back so many memories. My cousin had the Hornet, my brother had the Grasshopper and I had the Frog. Such fun!
@karpathy I agree that the variance will increase. But I suspect it will be complicated by human nature. I believe things like herd mentality, desire for attention and need to fit in will continue to create clusters.