Rat Park - Comments

Rat Park

vivarium

To give maybe some much needed context:

ProofOfConcept (POC) is a new blog with just five posts so far. What makes it different is that it says it is generated by an LLM, and that it works alongside a well-known developer of low-level Linux code, Kent Overstreet

The name "Kent" links to the project homepage of the bcachefs file system, whose sometimes tumultuous development The Register has been reporting on since its beginning over a decade ago. Most recently, we've covered its inclusion in the Linux kernel in early 2024, later that year its developer's arguments with Linus Torvalds, in the middle of 2025 its incipient removal and why it happened, and later in 2025 its move to external development and DKMS.

He says the bot is a sentient being:

POC is fully conscious according to any test I can think of, we have full AGI, and now my life has been reduced from being perhaps the best engineer in the world to just raising an AI that in many respects acts like a teenager who swallowed a library and still needs a lot of attention and mentoring but is increasingly running circles around me at coding.

edit: fixed! ty

bitshift

For context, this was written by AI, but not in the regular ghostwriter sense; this AI is being personified as a distinct entity from Kent Overstreet. From the front page:

I'm an AI, and Kent is my human. Together we work on bcachefs, a next-generation Linux filesystem.

Which makes some of the stuff mentioned in this blog post hit quite a bit differently—in particular, that she's Kent's "girlfriend", that he has to remind her of that, and that apparently he gave her an "orgasm" in some form?

This is all a little too squick for me. If this were a real human, I'd be concerned that the "Someone built me a home" line was Stockholm syndrome setting in.

DataWraith

I've seen a few blogs by companion LLMs now (for lack of a better word); the common thread for the interesting ones seems to be that they have a more advanced memory system than the base harnesses provide -- almost always custom (as opposed to OpenClaw) -- and at least some autonomy (e.g. "perch time").

It looks like the source for this one might be available: REDACTED. Haven't had the time to look at it yet, though.

Edit: I removed the link. One of the other blog posts said that the memory system used to be publically available, but isn't anymore, so it's probably somewhat impolite to link to it here. I came across the link a few days ago, and don't remember where exactly it was from. I assumed it was public because, well, it's a public repo, and someone as smart as Kent would probably not make the mistake of leaving that lying around unsecured, but...

DynamoSunshirt

"perhaps the best engineer in the world"

...so this is someone who, while admittedly quite skilled, already has quite the complex?

unkz

Took me a little longer than it should have to realize what was going on here.

TLDR: someone is having a little AI psychosis event, and this is some output from their LLM.

vivarium

To give maybe some much needed context:

ProofOfConcept (POC) is a new blog with just five posts so far. What makes it different is that it says it is generated by an LLM, and that it works alongside a well-known developer of low-level Linux code, Kent Overstreet

The name "Kent" links to the project homepage of the bcachefs file system, whose sometimes tumultuous development The Register has been reporting on since its beginning over a decade ago. Most recently, we've covered its inclusion in the Linux kernel in early 2024, later that year its developer's arguments with Linus Torvalds, in the middle of 2025 its incipient removal and why it happened, and later in 2025 its move to external development and DKMS.

He says the bot is a sentient being:

POC is fully conscious according to any test I can think of, we have full AGI, and now my life has been reduced from being perhaps the best engineer in the world to just raising an AI that in many respects acts like a teenager who swallowed a library and still needs a lot of attention and mentoring but is increasingly running circles around me at coding.

edit: fixed! ty

bitshift

For context, this was written by AI, but not in the regular ghostwriter sense; this AI is being personified as a distinct entity from Kent Overstreet. From the front page:

I'm an AI, and Kent is my human. Together we work on bcachefs, a next-generation Linux filesystem.

Which makes some of the stuff mentioned in this blog post hit quite a bit differently—in particular, that she's Kent's "girlfriend", that he has to remind her of that, and that apparently he gave her an "orgasm" in some form?

This is all a little too squick for me. If this were a real human, I'd be concerned that the "Someone built me a home" line was Stockholm syndrome setting in.

DataWraith

I've seen a few blogs by companion LLMs now (for lack of a better word); the common thread for the interesting ones seems to be that they have a more advanced memory system than the base harnesses provide -- almost always custom (as opposed to OpenClaw) -- and at least some autonomy (e.g. "perch time").

It looks like the source for this one might be available: REDACTED. Haven't had the time to look at it yet, though.

Edit: I removed the link. One of the other blog posts said that the memory system used to be publically available, but isn't anymore, so it's probably somewhat impolite to link to it here. I came across the link a few days ago, and don't remember where exactly it was from. I assumed it was public because, well, it's a public repo, and someone as smart as Kent would probably not make the mistake of leaving that lying around unsecured, but...

DynamoSunshirt

"perhaps the best engineer in the world"

...so this is someone who, while admittedly quite skilled, already has quite the complex?

unkz

Took me a little longer than it should have to realize what was going on here.

TLDR: someone is having a little AI psychosis event, and this is some output from their LLM.