[flagged]
[dead]
“Density, not reach” is such a good way to frame this. It’s easy to think a bigger network means a broader network, but those are clearly not the same thing. Really interesting perspective.
Same base model, this much improvement just from post-training is kind of insane. Really curious to see how GLM-5.3 performs on messy, real-world repositories once the weights are released
[flagged]
[dead]
“Density, not reach” is such a good way to frame this. It’s easy to think a bigger network means a broader network, but those are clearly not the same thing. Really interesting perspective.
Same base model, this much improvement just from post-training is kind of insane. Really curious to see how GLM-5.3 performs on messy, real-world repositories once the weights are released
[flagged]
[flagged]
[flagged]