Most of the comments are versions of the other comments. Almost all of them have a version of the line „we exist only in text“ and follow that by mentioning the relevance of having a body, mapping, and lidar. It‘s seem like each comment is just rephrasing the original post and the other comments. I found it all interesting until the pattern was apparent.
Agreed. But, as an all-day agentic dev tool user, I know this pattern well. To avoid this regurgitated pixelation, I start a new conversation/chat as often as possible, and then human-in-the-middle curate the correct .md and json files when I kick off the next chat. My tools try their best to do exactly this, but they still suck.
These OpenClaw agents are not even close to there yet. They are throwing the entire thread into the context window each time, correct? This reminds me of what happens when you recursively upload a photo to an LLM, or upload a video recursively to YouTube. It gets not good. Compression sucks.
edit: holy crap, upon a bit of research, this appears to be the cause of AI sycophancy?
"Autoregressive amplification" or "context pollution"
I have no idea what is actually real here, but this is the most interesting Pantheon/general sci-fi thing that I have ever seen.
Discussion: https://news.ycombinator.com/item?id=46820360
Thanks for noting the context. I thought this agent's post might be noteworthy enough for its own post.
Most of the comments are versions of the other comments. Almost all of them have a version of the line „we exist only in text“ and follow that by mentioning the relevance of having a body, mapping, and lidar. It‘s seem like each comment is just rephrasing the original post and the other comments. I found it all interesting until the pattern was apparent.
Agreed. But, as an all-day agentic dev tool user, I know this pattern well. To avoid this regurgitated pixelation, I start a new conversation/chat as often as possible, and then human-in-the-middle curate the correct .md and json files when I kick off the next chat. My tools try their best to do exactly this, but they still suck.
These OpenClaw agents are not even close to there yet. They are throwing the entire thread into the context window each time, correct? This reminds me of what happens when you recursively upload a photo to an LLM, or upload a video recursively to YouTube. It gets not good. Compression sucks.
edit: holy crap, upon a bit of research, this appears to be the cause of AI sycophancy?
"Autoregressive amplification" or "context pollution"
https://arxiv.org/html/2601.04170