The science might be legit here, but I'm getting really, really tired of the way every single piece of writing to come out of Anthropic is written in some kind of self-aggrandising, wooey wonderous 'our model has developed a genetic mutation that makes it have feelings' bs style. Regardless of what they're trying to communicate, those undertones are always there. It's annoying and disingenuous. Homeopathy 'this-water-has-feelings' level annoying. None of the other labs write like that.
They might as well change their name to Anthropomorphic at this point.
I'm honestly amazed that people are only just figuring this out.
I know that reads like I'm being snarky, but I'm not trying to be. Within the last decade in the UK we have had (among others):
- the 2016 Snoopers' Charter
- the 2022 Police, Crime, Sentencing and Courts Act
- the 2023 Public Order Act
- the 2023 Online Safety Act
- the 2024 Addendum to the 2016 Snoopers' Charter
Couple that with the whole push to repeal the 1998 Human Rights Act and withdraw from the European Convention on Human Rights, and the fact we've started imprisoning pensioners for holding up signs, and it's really difficult to _not_ see exactly where this leaves us.
I wish I could look away and get on with my life, but I can't - and I'm starting to realise that that is also part of the design. The increasing reluctance to express controversial opinion online isn't an accidental side effect of all of this legislation. It is intended behaviour, the desired outcome.
I agree 100%. All of the models do it to some extent after the context gets tired, but opus is the worst and the sneakiest. And even when you do coerce it into doing what you want it feels like something out of r/maliciouscompliance. Much more so than most non-anthropic models. Way more so than codex/gpt or even gemini.
Pretty sure this 'article' was written by an LLM, having scraped the HN discussion on here from 4 days ago. Nothing new there apart from a clickbait title and a ton of ads.
I like having control over font hinting and aliasing and stuff, so I really prefer the latter.
I found the AVP fun at first, but then after a few days it started to get more and more annoying- whereas the Vitures were a bit annoying at first, but then just became amazing after my eyes/brain got used to them. Almost overnight.
I wasn't talking about your general point though. Your comment opened with a statement and a question, and I was quoting the statement and directly answering the question you asked.
> That's a different problem entirely and predates the recent GenAI craze.
From the perspective of respecting the reader's time and attention, I see it as almost exactly the same problem, which is why I made that comparison when answering your question.
> ...People like yourself moaning...
Seeing as you conveniently linked to the HN commenting guidelines, I suggest you take another look at them, maybe focusing your attention on the ones closer to the top of the list, and then re-read the comment you just posted.
Yeah, apart from there is more to supporting a product than just compiling a binary.
Case in point, the HN post you're commenting on is a link to their support forum. Search for _anything_ there, and this the pinned article that appears on top of every results page (from feb 2024):
> Is your Operating System supported by ISE/Vivado tools? Assistance and support won’t be provided for software and IP installed on unsupported OS!
> Note: Technical Support and assistance will be provided ONLY for Software and IP installed on supported Operating System!
> It is strongly suggested to check if the OS you are working with, is one of the supported operating system for ISE/Vivado tools!
*There have been many questions where users are trying to install or run into issues with using an unsupported OS for ISE/Vivado tools.*
> Assistance and support won’t be provided for issues observed on unsupported operating system!
Just like those spam 'articles' that may at their core be interesting or have some value - but force you to click past 4 ads and scroll over/filter out another 17 just to extract the promised value - noticing that content you're consuming is obviously AI generated results in two things:
1. resentment that your time and attention was wasted by machine generated word-padding, and
2. a loss of confidence in the accuracy of the information presented
For any linux folk who haven't tried something like this yet, keyd and wtype are what I settled on, having tried the kmonads, interceptiontools, xscapes, kanatas, etc. (though I've been meaning to try kanata out again now it's a bit more mature)
Trying to read between the lines, here are my lazy sunday morning guesses at what might be going on here:
1. The Xilinx team are pushing back on the increasing number of things they have to support. Silver lining, maybe this means they're being asked to work on a new product that will require redistribution of headcount (like maybe another NPU )
1.1. Their Linux expertise is lacking / stretched across multiple teams (this is the impression I got from following the work in github.com/amd/xdna-driver over the last year or two). Maybe this is the outcome of a 'these are the things i'm doing now, so if you want me to do something new then tell me which of these things I can drop' type conversation & where the pushback is coming from (maybe we'll get some fedora support in that repo though ) .
2. Marketing have been pushing for something that helps them 'fight the AI fight', and it may be that they've now been given the mandate so the division is in the midst of the typical top-down mythical man-day reallocation wave. Xilinx have probably been told that priorities are shifting towards integrating more of the Xilinx inference tech with more mainstream AMD products, possibly at the expense of their existing roadmap. Xilinx have tenured employees who know what they're doing and don't want to retrain/change, so this is a side-effect of the pushback.
3. This is a straight-up monetisation strategy. Marketing ran a project and concluded thta it's just not worth supporting that lower tier for free. It may be that even though have a majority Windows userbase, the [commercially serious | higher stakes | CICD pipeline based] development actually happens on Linux, and this is them closing that loop. Not quite a Docker Desktop situation, but maybe not that dissimilar - they're saying that most professional/commercial users are Linux users, and the days of unlimited free commercial use on the smaller devices are over. Maybe the margins on those lower end devices aren't good enough to justify the amount of support overhead, and pay-to-play will filter out the noise and ensure they're talking to users who are already bought-in. Or, maybe somebody just needs an earnings blip on a slide somewhere, and this is them milking their startup/smb customers.
How does your thread-per-connection model compare to Heikki's proposal[0][1] from back in 2023?
[0]: https://www.postgresql.org/message-id/31cc6df9-53fe-3cd9-af5... [1]: https://www.youtube.com/watch?v=xLLakMmVtbY