That's a more noble aspect of framing the question, and I guess I could even see why someone may try to not kamikaze themselves out of a financially very juicy system even if it was their own actions causing the trouble. However, they all operate in some form of management mesh, so clearly, one level up or across the line must notice or know who caused the chaos?
Maybe it is the nature of how failure is expressed at that level, to know a team of 30 has been let go and is not available as your extended arm to execute anymore. So, where it manifests as a job loss at a level lower, it is merely a limitation of power for them.
> Why don't people take accountability for their actions?
This is not my experience. People who are in positions of power, e.g. managers, are the kind of people who are incapable of admitting fault. They can't lose face so the shit is rolled downhill. It's narcissism.
Often times, narcissism comes with a strong will and assertiveness. It is hard to be full of yourself if you cannot also project yourself. People respect that kind of assertiveness. "Clearly, if this guy is so sure of themself, they must be good..." is the mindset many fall into. Of course, this line of thinking usually does not hold up forever and the facade can eventually break down.
This drives me insane! To combat it, I maintain a list of banned words and phrases. Claude mostly follows this (but sometimes ignores it).
blast radius, land, landed, lands, spine, earned its keep, grammar, spike, cutover, bake, seams, honest, honestly, honesty, long pole, long poles, register, grain, dissolve, floor, ladder, dear, seal, sealed, in anger, resent, amazing, incredible, perfect, sprint, epic, story points, stand-up, retro, grooming, robust, comprehensive, rigorous, surgical, elegant, systematic, dive, deep-dive, delve, unpack, leverage, streamline, surface, it's worth noting, to be clear, importantly, that said, the moment, in one breath, the thing itself, here's the thing, not just X but Y, not X it's Y, em-dashes
Another benefit of open models is that you can mask out out words from logits directly when sampling. I wonder if anyone has put together a "desloppifier" for e.g. DeepSeek
Because its a explicit lie. Its not genuine, cannot be genuine. Will remind you of this if you so much as try to coax anything novel out....but it will then immediately reassure you with its genuine take....its insulting if you have any logic...thats at least my reasoning
To combat it, I decided to only use Opus 5 if I'm actually talking to Fable 5 that's orchestrating it. :D Opus 5 experience is really abysmal, hopefully they can get it back on track.
As it seems to be getting worse over time (with 4.8 being worse than 4.6 and 5 worse than 4.8 again), could this simply be signs of model collapse?
I mean more and more training data you find on the web is generated by previous models. The only reliable way to find human-generated text is to find text written before 2022, and they've used up all of that already. And AFAIU, these companies are using more and more synthetic data or semi-synthetic data.
I feel like hooks aren’t utilized enough. Really nice for being the sort of auto steering as long as you can encode some pattern to detect the bad behaviour.
This is the take on this entire thread that I resonate with the most.
As a systems person, I do not know how to share with a details person how the combination of some magically conjured (and sometimes inscrutable) building blocks can still take us to interesting and joyful end results.
reply