5 ms·
I suspect that Grok has been ironically lobotomized by pressures to correct its political views. Similarly, I could imagine the Gemini folks working in a signi
by arw0n 4mo ago
I suspect that Grok has been ironically lobotomized by pressures to correct its political views.
Similarly, I could imagine the Gemini folks working in a significantly more complex corporate climate, with different parts of Google pushing for different capability focuses. They are only lagging behind less than a year, so it isn't too large of a gap yet.
That said, the fact that Anthropic is currently the top dog suggests that talent and execution is incredibly important. A year ago none of my normie friends new them, and when i suggested using Claude looked at me like when I recommend Linux.
- janalsncm 4mo agoThat shouldn’t affect Grok’ coding ability. How often are people discussing politics with Claude code? Writing decent code is just hard and it’s not just Grok.
- bwhiting2356 4mo agoIt affects their ability to hire and retain talent.
- janalsncm 4mo agoIf training a good model requires talent then that’s the answer to the question this thread is trying to answer: is training a good model actually that hard?
- buthowjejddjeu 4mo agoTalent to do.. what? This could mean a lot of things. Navigating astronomically huge fundamentally not so hard but still really tangly and hairy projects requiring both excellent short- and long-term vision in an overheated domain with angry people and lots of money is a skill all of its own.
- janalsncm 3mo agoTalent to train high quality LLMs, especially coding LLMs.
- black_knight 4mo agoWhy would these be independent?
- janalsncm 4mo agoMore specifically, political lobotomy shouldn’t affect coding ability.
- Hamuko 4mo agoIt's all a bunch of weights isn't it? Why wouldn't fiddling with some parts of the weights have cascading effects?
- Discordian93 4mo agoYet empirically it does
- girvo 4mo agoYou’d be quite surprised, I think. Fine tuning a model on one axis can have drastic impacts on another that as a human we would expect to be completely unrelated.
- janalsncm 3mo agoI have never seen anyone argue that this cannot be overcome with more high quality RLVR data. The practical reality is that the Chinese and American models might have very different politics. But the most relevant factor in model performance is the quality and volume of training data, not ideology of the base model. Unless you are suggesting something very particular about the way Grok was neutered.
- thot_experiment 4mo agoNot true, aggressive post training makes models notably dumber.
- KaiserPro 4mo ago> That shouldn’t affect Grok’ coding ability. If you are spending all your time having to re-train because the boss doesn;t like the output, it will hamper coding
- Oreb 4mo ago> A year ago none of my normie friends new them, and when i suggested using Claude looked at me like when I recommend Linux. Isn’t that still the case? Normies haven’t even heard about Claude, in my experience.
- buthowjejddjeu 4mo agoIn my experience it has improved a bit, but 90% of my normies still have no idea. (It was 100% before)