pixelmelt

I'm inclined to agree given how unstable Gemma 4 is when not using the "official" sampler settings

Aren't consumer class dumb TVs gone?

Can confirm it is also useless for building tools defending against reverse engineering work (unless asked to do code review for some reason?)

Ironically making a stink about it online is likely to have a larger impact then using their dedicated feedback or support channels (which go to claude, not a person)

It has about the same problems with finding content that the www has, its just in the earlier stages

Reminds me of this story https://marshallbrain.com/manna1

Try blackbox, its a puzzle game that will have you messing with device settings to beat levels

Its already happening, some of my friends have (seemingly) inherited LLM verbal tics, specifically overuse of the word "highlight."

Im in the community reverse engineering web CAPTCHAs, it's because they are too easy to reverse engineer with Claude now.

I've seen multiple people break botguard (the obfuscation used by recapcha) within the last year when before it was considered a huge technical envour.

Devices like phones don't have this issue since Google owns the client attestation end to end and can fingerprint you without the risk of receiving spoofed values.

I've been working on my own misaligned model and grok is definitely different enough with a syspronpt compared to all the other frontier models that I've considered using it to generate synthetic training data, however it leans really heavy into LLMisms which makes it not really worth it. Tangentially I also really like the idea of llms as librarians they are trying out with grokapedia.

There was an old game called cursors.io which was the same concept but collaboratively traversing a maze where you would sometimes have to leave other players behind to reach the next level

I opened this thread expecting a bunch of "kids these days..." posts, kind of surprised not to see any. People have been raising themselves up by putting down other generations since the very first I assume, the temptation towards the fallacy of composition is too irresistible.

I liked this, reminds me of some other discussion on recycling/global warming etc being pushed as the comsumers fault

I've been getting into making and breaking these antibots recently and it's funny to me how the person who wrote this post gave so much attention to what LinkedIn was doing and left the other antibots on the page as a footnote. They grab way more, they just don't let you see it. I haven't reversed PX or Recap yet but the antibot on twitch and Nike similarly checks if you have any of these 53 apps installed (when loaded on a WebKit browser) https://pastebin.com/raw/KACvjpTK

And so did everyone else, making it... Dead? If you've got some good alternative communities then do tell!

I've already been doing this haha, got Claude to install VR mods into a game that most certainly should not have been working with VR mods on linux

It sounds weird because Claude wrote it

Some of the more recent versions of it had memory leaks so you couldn't just leave it on in the background

As one of said generation, I would chalk it up to instant communication creating innumerable shallow remote relationships that significantly replace time spent with others in person.

maybe not overrun by spam, but the amount of bots I see on popular subs is definitely not 0

I found a new one in claude recently with "Fair enough, ..."

Wow this is big, what's the best device/remote stack to use it with?

Hoarding/archiving/collecting is quite fun. I don't think it's a stretch to say he likely read more then the average person, and from his own collection too!

They have a support article on how to fix this!

Relevant username

Had a friend who worked on designing systems to pick these signals up around airports

Its effectiveness is even more apparent with older smaller LLMs, people who interact with LLMs now never tried to wrangle llama2-13b into pretending to be a dungeon master...

Check out Unsloths REAP models, you can outright delete a few of the lesser used experts without the model going braindead since they all can handle each token but some are better posed to do so.

You should add an RSS feed so I can follow it!

I've been enjoying smashing through those panes of glass with LLMs. Just spin up a browser, go and export all the requests to undocumented user endpoints to curl commands, specify what you want out of it and after a few minutes you have an api endpoint up and running!