@cabinsmountain Dude, your picture contains an answer why it wouldnt work. Living in a village you will regularly get a family with more than two kids. At that point the amount of houses you have to build becomes... difficult. You really should read some lit you know try "the Land" Kobylianska
@johnjhorton@sinanaral Imo the issue would come rather from post-training, because there are RL-loops explaining to the model how to be "a useful assistant". And indeed being a "useful assistant" stereotypical matches with suggesting to take 45, instead of 40.
@johnjhorton@sinanaral I like your finding, and agree with the practical conclusion, but im not sure it is a good explanation. You may underestimate the diversity of sources that enter models (granted, they still be written sources, which is a "bias") + the capability of LLM, to escavate obscure stuff
@HarshithLucky3 Dude, google *is* out of frontier race for a long while at this point. At best it would be the same as 5.5.
But probably it will be yet another outstanding model from google google google google google google google google google google google google google google (google)
@haider1 I'm sure it will match some some some some some some some some some some some some some some some some some some some some some (some)
You got it. For the last year at least gemini seems like a bad joke, not a model that can match anything sota-ish
@jmkeegin And having fun can consist in discussing a question that came to your head during the reading can be externalised, when you feel a pull of it, and not day(s)\weeks later when the other person in question will have enough time to put their thought to it.
@goodworse > you easily generate THOUSANDS of lines of high-quality code
You have a mistake here, it should be:
> you easily generate THOUSANDS of lines of the same token (cus nobody cares to set you up properly)