If this is not slop, it's certainly written in that style. It's also not true. For example:
Separately, Nicholas Carlini published a 15-round reverse shell exploit here, crediting “Claude” without specifying the model version. I read every file in that repository — the 601-line exploit.py, the write-up.md, and the claude-prompts.txt containing 44 human prompts across roughly 8 hours of work. What I found in there complicates every narrative about this exploit.
As a point of comparison, recently [link]an independent vulnerability research company[/link] showed that Opus 4.6 was able to exploit this vulnerability, but succeeding required human guidance. Mythos Preview did not.
So. That repository was not published by Carlini, the model version was specified, and it was specifically called out by Anthropic in their post as a notable example of a place where the advancement of Mythos over Opus was that Mythos was able to do the same thing autonomously.
I'm going to stop reading at this point. Why are people upvoting this? Is that just that much demand for anti-LLM stuff that people aren't even reading it?
Unfortunately this article is LLM-written-or-edited, so I don't trust the contents either. In my experience, an LLM can write an argument for or against anything that's convincing at a glance, but at the end of the day I'll always remember arguing with Claude over some Python minutiae that it (anthropomorphising language warning) refused to accept was true until I showed it the code. I will never trust LLM output to be true unless it compiles on my machine.
bakkot | 5 hours ago
If this is not slop, it's certainly written in that style. It's also not true. For example:
That repository is linked from Anthropic's announcement post, with the text being
So. That repository was not published by Carlini, the model version was specified, and it was specifically called out by Anthropic in their post as a notable example of a place where the advancement of Mythos over Opus was that Mythos was able to do the same thing autonomously.
I'm going to stop reading at this point. Why are people upvoting this? Is that just that much demand for anti-LLM stuff that people aren't even reading it?
WilhelmVonWeiner | 5 hours ago
Unfortunately this article is LLM-written-or-edited, so I don't trust the contents either. In my experience, an LLM can write an argument for or against anything that's convincing at a glance, but at the end of the day I'll always remember arguing with Claude over some Python minutiae that it (anthropomorphising language warning) refused to accept was true until I showed it the code. I will never trust LLM output to be true unless it compiles on my machine.
doctor_eval | 2 hours ago
Even if it compiles it is, in my experience, equally likely to be correct, fragile, insecure or just plain wrong.
addison | 2 hours ago
I think we have to mark such generated articles as spam, they are simply attention farming at this point.
Yogurt | an hour ago