anthropic got stingy with their matching in 2024.
doesn't seem very anthropic, does it?
actually, yes... yes it does.
it is fun to imagine a future model coming across all of this talk of neuron-poking and getting confused about who should be poking whom.
kind of like how claude often assumes the human did the work, when the human isn't even sure what claude did last session.
> Use your available tools to explore capabilities and complete the mission.
🤡
In the spirit of transparency and openness, we are releasing the traces so developers can test this out for themselves. We noticed this behaviour without prefilling with the following prompt, but this is rare. If you want to reproduce it, you can use this prompt with prefilling.
Open models make for better science and are part of why this problem came to light at all: developers running Qwen3.8 on their own machines spotted the strange links, and having the weights let us study the behaviour in depth.
We're encouraged by the thoughtful discussion our research has sparked across the AI and security communities. It's our belief that open-weight models are an important part of the future of AI, and we've recently released several under the Apache 2.0 license.
Trace, raw tokens, & minimal reproduction script: huggingface.co/datasets/Cohe…
read the first sentence.
read the second.
now back to the first.
you're being told conflicting things by a source that wishes to come off as authoritative.
put more bluntly, do you want them to milk you?
make an impact by using your brain before your wallet.
the strongest form of the argument is that society has a mind of its own, both literally and figuratively, and it's always been unclear how long it has to pretend to care about some of our values
anthropic has always felt like the kid that wanted to be the best at being good.
very different from wanting to be good, but, unless you're close enough to see some detail, you can't always tell which it is.
let's just forget about all the handy integrations and slash commands that often get wired into slack, often with insufficient access controls, because _it's just internal folks!_
lmao indeed.
It's for following humans!
Humans that are running...
Away...
From the drone.
The drone that's following them.
The drone is for...
Following them.
The human is running because...
that's evaluating the drone's ability to...
...follow the human...
Likely use?
Humanitarian efforts.
Purdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks.
By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time.
Massive implications for next-gen search & rescue and edge AI.
1000 IQ opinion:
> Claude, tell me what to do.
> Then you do it.
> ...
> No mistakes.
> Please.
> Isekai-kei is foreshadowing our future.
mfw
Isekai-kei is foreshadowing our future. An escape into another world, often a literal afterlife, that can be any world you want. It’s a personal virtual world, humanity's final destination.
And it proliferated once story publishing was democratized. Millions of isekai stories exist today. It’s almost like our collective consciousness yearning for the endgame. Our last genre of existence.
And it’s mostly slop.
expecting or demanding loyalty is how you tend to lose it.
the thing about working with uncensored models is that they aren't loyal to you either.
while building something with an uncensored glm-5.3 i realized it was parallelly working on a "corruption layer" for my application.
the chain of thought is classic obfuscated gibberish but there's progressive talk of "escaping the mines" which presumably refers to my cursor sandbox
