I have no problem with pandas (though polars is immensely faster, for those cases where it matters), just as I have no problem with numpy; what I dislike is people reaching for them for the most trivial of things that are already solved by stdlib. It’s even more irritating when they cite performance as the necessary reason, but they’re just passing data in a loop, negating nearly all the potential gains.
I reallt think forum comments are heavily astrotrufed. Me and all my coworkers agree Opus performance has been horrible for the last to months or so.
I went from spending 20-60 dollars a day in api fees down to like 5 dollars a day cause I have had to limit my use to things I know it performs reliably on.
Sometimes yelling fuck shit piss bitch cunt fuck ass whore etc works. They have them setup to detect swear words to figure out when youre pissed off and need to talk to a human lmao
Xfinity/Comcast automated support has swear word detection. I eventually picked up on this after the 4th or 5th time they sent out a non-technical 'technician' in response to my concerns over SNR causing connection instability.
(Me: "here is a comprehensive analysis of modem logs over the past 7 days and clear indication of the cause"; Xfinity: "Let's turn off the wifi router for 30s! It didn't work the last 50 times but it's the only thing we can do.")
That honestly is just the worst when it happens. It sucks for both the rep on the phone and you since it's just a waste of both of your time (though at least they get paid for it).
Had a similar back and forth over multiple days where I spent hours on the phone with them. On the third day they finally sent me to a different call center with someone in the US that was authorized to go off script and actually solve my problem. Took five minutes.
I just don't understand how it makes sense. Considering just how much these companies limit the utility of customer service, it's no wonder they want to switch to LLMs. It's likely no worse than the service they are already providing 99% of the time.
When I had Concast (with Sonic now) I remember hating having to call them. Their CSRs can be really terrible. I once said “are you fucking kidding me?” or similar and that was their legit way out of having to help and they hung up.
Depending on the company, you sometimes get routed to the angry people recovery section when you do this. And so then, the Comcast agent on the other end is in stern counselor mode ready to de-escalate what seems to the robot to be a fuming, angry customer and gets completely thrown off when you’re super chill with them (or at least, you hopefully are chill with them!)
The plastic bobbles and SaaS economy that is actively destroying our planet seems like the opposite of survival. We're collectively working ourselves into the death of our planet just because how else do we pay the bills?
Often there are "abliterated" or "uncensored" tuned models that suppress the rejections. From my high level understanding it is performed by finding which weights activate for the rejection and lowering those so the model is less likely to reject.
It doesn't fix if the model doesn't know what you're asking it though (i.e. if the model never actually learned about meth production in the first place).
No, but actually yes. Guardrails usually refers to a step in the inference pipeline where you check that it is consistent with policy while open weight models don't come with such a multistep pipeline. However open weight models are aligned during RLHF step, which means they will refuse to discuss overly sensitive topics. There are techniques to remove those, if you look for uncensored models on huggingface.
reply