111–120 of 146 posts
Earlier quoted context omitted.
Manual vs automatic is irrelevant here. There was much less content to moderate if we talk about media. They could censor every single publication, movie or song manually. In oppressive regimes like USSR the censorship is total - the only escape is your kitchen (if you can really trust the people you talk to), in the West you are free to choose the platform and live peacefully in your bubble if you want.
[flagged]
Earlier quoted context omitted.
> no one gave a fuck about your race or age or sexual orientation or politics or anything but can this person code, do they geek out Is this really true though? There is so much evidence that shows women and other minority groups to be systematically discriminated against in tech.
I’m sure there is better data than my anecdotes. Back in the day it seemed like everyone was “different” somehow. Certainly when I ran the biggest group I ever have my boss (Ning Li, absolute pro) was a woman, and 2/4 of the managers I supported were women, and in a group of pushing 40 I was one of two who was an American white guy. I suppose it’s a judgement call if FSU folks count as “white”, but if not then “white…
Sure, I guess that's more to do with the computing industry becoming more corporate.
The point I'm trying to make is that what you refer to as "stupid political posturing" is a good thing actually. It's a sign of progress that people feel more free to speak out about issues affecting them. They can seem insignificant, but it's only by having difficult conversations and demanding better that we make the whole industry a place where all that matters is "can this person code, do they geek out".
So, in general Claude still beats newest GPT-4o model in software development category, right? [1] Although new GPT is better at small coding excercises (72.9% vs 68.4%), it falls behind in more challenging refactoring tasks (62.9% vs 72.3%). I used Claude through VPN from Europe, and from the start could see it was much better than GPT-4, in my limited very subjective use cases. [1] https://aider.chat/docs/leaderboa…
I'm genuinely curious to which extent widely published benchmarks can be used for this kind of assessment. The exercises this benchmark uses seem to be available in https://github.com/exercism/python/tree/main/exercises, and it seems that there are solutions at least at https://exercism.org/tracks/python/exercises/series/solution....
Overfitting used to be a concern when it came to comparing ML models, but it feels like it's not being talked about that much anymore, is it? What if one model has the exercism solutions in its training data, but the other model doesn't, is the comparison still fair? If it solves 70% of the exercism exercises correctly, would it perform as well on similar prompting for unseen problems (which are more likely to be what I care about)?
Earlier quoted context omitted.
Oh, it absolutely is. Censorship in USSR was manual labor. The scale at which its is operating today is incomparable, as is the amount of false positives which it captures, as well as the variety of censored topics across different institutions.
Manual vs automatic is irrelevant here. There was much less content to moderate if we talk about media. They could censor every single publication, movie or song manually. In oppressive regimes like USSR the censorship is total - the only escape is your kitchen (if you can really trust the people you talk to), in the West you are free to choose the platform and live peacefully in your bubble if you want.
They banned my account a few weeks back because I asked it to translate some Japanese text that was on a kitchen knife. Still no response from their Trust & Safety team. It's frustrating because Claude is really quite good at coding and I used it frequently.
Earlier quoted context omitted.
It's very different. A stabbing is a crime that is prosecuted, state sanctioned imprisonment is not.
the chilling effect is the same.