wren6991

7 ms·

91.
by wren6991 3mo ago
Replying inline because this hit max depth: > You are confusing CPU-level security with filesystem-level security. > I'm curious - how do you expect an LLM harness to build and test executables without being able to build and exe
92.
by wren6991 3mo ago
Here: https://en.wikipedia.org/wiki/W%5EX This doesn't require cooperation from the shell. If writable paths are non-executable, and executable paths are non-writable, then executing `cp $(which git) my-new-favour
93.
by wren6991 3mo ago
I mean, ideally enforce W ^ X, and enforce stricter permissions on executables where the agent has flipped the W to X. My broader point is we should be using operating system primitives instead of regex and wishful thinking.
94.
by wren6991 3mo ago
Sure, the two main approaches are revoking execute permission from the executable itself, and write-protecting the things that it would modify (like $PWD/.git). Both of these are achievable with sandbox-exec or landlock_restrict_self(2
95.
by wren6991 3mo ago
Yeah, this is one of the weaker parts of the post because I didn't really make my position clear: that sandboxing should be a first-class integral part of every harness, not an afterthought that risks putting something you care about o
96.
by wren6991 3mo ago
Prefix caching is inherent in how global softmax attention works. Modifying the prefix invalidates the suffix because the suffix's cached KV projections are a function of the hidden state after the model has evaluated the prefix. There
97.
by wren6991 3mo ago
Yeah, I don't think that line quite landed. My point was that LLMs are inherently adversarial (read, "relentlessly proactive") and sandboxing should be a first-class integral feature of your harness, not an afterthought. The
98.
by wren6991 3mo ago
yes, and they could all be closed by the stale bot
99.
by wren6991 3mo ago
If you don't know what OpenCode is, imagine a dark and stormy night.
100.
by wren6991 3mo ago
Currently this is still just going to Moonshot, since they didn't release the weights yet (scheduled for 27th).
101.
by wren6991 3mo ago
> Is there any good reason to believe there is a lot of headroom or there is not? It's hard to answer quantitatively, but for example Qwen3.5 -> 3.6 was a significant step in capability, arising from continued post-training of th
102.
by wren6991 3mo ago
Cryptography *2*? We're still over here trying to implement Cryptography 1 without side channels, and they went and invented a new one?
103.
by wren6991 3mo ago
LLMs are hopelessly confused about which model they are. Ask DeepSeek V4 Flash which model it is, and it's 50/50 between "I am DeepSeek (深度求索)" and "I am part of the GPT-4 series developed by OpenAI." Ask Claud
104.
by wren6991 3mo ago
$230 for a macropad with an exposed PCB and no washers under the Allen screws. I'm not sure what the joystick is for, and neither are they apparently: the only example they give is something that could just be a keybind.
105.
by wren6991 3mo ago
The fact that RLVR rewards successful trajectories. Notice I didn't use the word motivation , which you decided to use. The reason LLMs use a technique is simply because it works, and that comes straight from training (specifically RL
106.
by wren6991 3mo ago
Crashing instead of raising an exception or bubbling a return code sounds like a legitimate bug ticket IMO. One reason I'm in favour of handling capabilities in the OS is so we can stop having trivial symlink traversal and /.. tra
107.
by wren6991 3mo ago
You're right to push back on this. The honest take is it's not a smoking gun -- that's a sharp critique.
108.
by wren6991 3mo ago
It tends not to improve things. Besides the generally bland and muddied style, and the low-fidelity reinterpretation of your points, they also have a habit of randomly deleting sentences that didn't spark joy for them but were actually
109.
by wren6991 3mo ago
Right, reasoning models are almost more notable for what they don't do, which is: try to generate the first token of the response from a single forward pass through the weights. Non-reasoning models were still surprisingly good at ge
110.
by wren6991 3mo ago
...by inferring both the imports and the script body from the same context? I think you're suggesting there's some kind of information flow from the anticipated body of the script back up to the imports, but I don't see why t
111.
by wren6991 3mo ago
I think you hit the nail on the head here. Having subagent dispatch in the loop for RLVR is something we've already seen in open models, like Kimi K2.5 and later, so it's no great stretch to assume OpenAI are doing it too. If you
112.
by wren6991 3mo ago
> What effects provide beyond DI is entirely in their ability to abort (resume zero times) or (in the case of multi-shot effects) resume multiple times Thanks, this answers my question and does sound handy. Also I do see the value in hav
113.
by wren6991 3mo ago
Generally excessively verbose and failing to account for LLM failure modes. Example: when dispatching a subagent with an instruction prompt, I see the agent refer to "phase 5 of the plan" without actually pointing to the plan file
114.
by wren6991 3mo ago
Given how poorly LLMs do with writing prompts for LLMs, I'm not sure I'd trust their judgement in designing a language for LLMs. > and the runtime requires explicit permission to touch the filesystem, network, etc This feels li
115.
by wren6991 3mo ago
It's still on the backburner. I'm gonna finish it... one day...
116.
by wren6991 3mo ago
> And, on the other end, C's warts, footguns and ancient quirks also matter less if you have an LLM plow through it. Warts and ancient quirks, I can kind of see, but in what way have LLMs solved the footguns?
117.
by wren6991 3mo ago
Why do you want an engine? Just write games
118.
by wren6991 3mo ago
Thanks, I wasn't aware of that paper! I think the popular use of distill (one-hot rather than logits) is not due to Anthropic but due to the DeepSeek-R1 tech report: https://arxiv.org/abs/2501.12948
119.
by wren6991 3mo ago
> At some point, you need to watch out for cynical/nasty series. In fact, all of the books we purchase are ranked against peers they have read in the past, or those we know to actively avoid buying due to cynicism, sarcasm, or open
120.
by wren6991 4mo ago
From the obviously slop-generated website, to the wild technical inaccuracies like "Git makes you clone the entire history before you can open a single path", to the weird focus on performance when runtime is usually dominated by

More ›