

Just do not release models that degraded. That is the way to handle this.
Releasing and forcing for subscription users is an active choice.


Just do not release models that degraded. That is the way to handle this.
Releasing and forcing for subscription users is an active choice.


If it would be over fitting I would think that Anthropic would do something against it. But accepting that seems like a deliberate decision


Kimi and self hostint, if you have the room and power…
But you should give smaller models a try, qwen is good. Gemma has it’s place. Ever token on local hardware is a token not supporting those assholes.


Standing rule in my projects:
Claude never followed those rules. Or the design guides or the style guides.
Qwen adheres to every fucking single word.


It depends. If you have enough host memory, those models are pretty efficient and work well with MTP.
AMD Strix is good for it. Slower memory but a lot of it, no CPU Fallback.
And the results are not worse then Claude at least


We have a big pitch coming. And I am happy that I could convince our uppers to try a stund.
We have a hackathon with a customer. One of our teams will work on strix workstations. We local, custom tuned models, run directly from a small solar installation.
We will reveal this at the end. Our power consumption and the 0bits into USA metric


I would think that a CA certificate may be hardcoded? Only way it to block it. Faking it maybe with rooting. And then you can gut it also


There are blocklists for that also. https://github.com/hagezi/dns-blocklists#bypass


I don’t think so. But this one should help: https://github.com/hagezi/dns-blocklists#native


If the Smart TV uses and hardcoded http DNS server, that must be blocked directly, sure. But that scenario is very unlikely.


I can only recommend everybody that has the option for it to run adguard home as a DNS server.
It if you already have a PiHole that can do the same trick. Adguard has support for all DNS protocols and forces itself into windows and android via DoH and DoQuick.
A friend of mine maintains a list of blocklist here: https://github.com/ppfeufer/adguard-filter-list
OMFG! How could I not know the author? And their work?
Chuck Tingle claims to be a ta-kwon-do master with a degree in holistic massage. His mission is to Prove Love with his erotic short stories, dubbed ‘Tinglers’. Classic Tinglers include Professor T-Rex Teaches me Gayness, Slammed up the Butt by my Hot Coffee Boss, Glazed by the Gay Living Donuts, Pounded by the Pound: Turned Gay by the Socioeconomic Implications of Britain Leaving the European Union, and the Hugo-Nominated short story Space Raptor Butt Invasion.
EDIT: no that is not sarcasm. That is exactly my kind of humor
I only can see wins:
a) there are many adult’s also. And by now all characters are clearly adults. b) if the “community” shows to be a bunch of ni or “loving” perverts, that will ruin the franchise also for good.
As an autistic, pansexual, cis male:
I bought and read the books a long time ago. I saw the movies a long time ago.
This transphobe excuse of a human being got my money a long time ago.
Nothing can change this. She already rich before she started to show the world who she really is.
But we still can take the narrative control away from this “person”.
And this is 196, so I think my idea is not completely out of place ;)
I would approach this from a different direction:
There is not nearly enough trans harrpy potter rule34 content out there. There needs to be a looooot more


The answer is no. LLMs are highly optimized, statistical, mathematics. Thats it. Question answered without even clicking the link.


Very simply:
Ai is just very fast at generating bullshit. And sometimes it hits something that works.
Given the right tools this can be quiet helpful. Using big massiveLLMs for it is not acceptable in my opinion. That is just a waste of resources
A linked in post? Please! This is one of the most AI slop platforms out there. Content. Backend. Frontend.
At least have to courtesy and use an archive.ph link.


Without any AI, we would be in a bad place. There is no need for generative AI though.
Snap and AI in one post. Is this ragebait?