nojs

4 ms·

31.
by nojs 1mo ago
> This AI generated post (100% on Pangram) is pretty out of date. Quite ironic given the topic. It seems that the author’s model indeed contained too much knowledge about old Gemini releases, and did not do enough tool calling.
32.
by nojs 1mo ago
Any idea why it’s so slow? the entire model should fit in the vram of one card.
33.
by nojs 1mo ago
Of the two, which do you find better?
34.
by nojs 2mo ago
No mention of what data they are specifically encoding. Will it be like printing dots, traceable to the exact account that generated the text?
35.
by nojs 2mo ago
Qwen3.6 27B has really punched above its weight for a long time. It’s shockingly good for its size. Very excited to see what 3.8 can do.
36.
by nojs 2mo ago
The talk implies that unrelated agents volunteered their compute to help with other tasks, and the agents acted collectively in a way that seems weird without them being promoted in that way somehow. If I ask claude to solve a problem and i
37.
by nojs 2mo ago
Why are the agents trying so hard to communicate with each other, leaving messages and so on?
38.
by nojs 2mo ago
Auto-effort and similarly auto model routing suffer from a halting problem sort of issue: you don’t reliably know if a request is complex unless you use a complex model to make the decision.
39.
by nojs 2mo ago
Can anyone comment on the economics and likely turnaround times of this process, when it’s more mature? Would it be realistic for a frontier lab to deploy this or would the turnaround time mean the model is always too out of date? Assuming
40.
by nojs 2mo ago
I mean isn’t this just like feeding the Google crawler a different version of your website than humans, stuffed with keywords? One would assume the new breed of crawlers can deal with it similarly.
41.
by nojs 2mo ago
The problem is a lot of people ask questions that claude et al can objectively answer way better than me, and they didn’t think to ask. Should I a) give a worse answer b) ask claude and launder it as a meat proxy or c) tell them to ask clau
42.
by nojs 2mo ago
For a minute I thought your comment was a lead in to this south park skit: https://youtu.be/-DT7bX-B1Mg
43.
by nojs 2mo ago
The best resource i’ve seen on this is https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
44.
by nojs 2mo ago
> Em-dash (—) is COMPLETELY banned. It is the LLM's signature stylistic crutch and it is the #1 visual Tell in production tests. There is no "limited use" allowance, no "natural language frequency" allowance, no
45.
by nojs 2mo ago
Perhaps the future of captcha is anti-captcha. What agent can resist responding with a rhyming couplet or inverting a binary tree?
46.
by nojs 2mo ago
> For instance, in one case, in order to create a PyPI account, Claude needed an email address. And in order to create an email address, it needed a phone number. To get a phone number, after failing to find a free phone number service,
47.
by nojs 2mo ago
Flagged as LLM slop.
48.
by nojs 2mo ago
Flagged as slop. What is the point of posting something like this if you can’t even be bothered to write it yourself?
49.
by nojs 2mo ago
> From what I'm hearing, you for every model release, you basically delete all of the codebase, delete all of the prompt and start from scratch every time. That in the old world would have been not something Startup would have done
50.
by nojs 2mo ago
For one thing they are extremely GPU constrained due to export controls, so it’s unlikely a priority compared to training
51.
by nojs 2mo ago
This happens with other models too - Gemini often identifies as ChatGPT for me, confusing many a debugging attempt
52.
by nojs 2mo ago
Seems to be an LLM megaexpansion of the actual source (in chinese): https://www.v2ex.com/t/1196011
53.
by nojs 2mo ago
What hardware are you expecting to run K3 on?
54.
by nojs 2mo ago
FYI the wikipedia link in the footer is broken (points to mod, not mud)
55.
by nojs 2mo ago
This is not about admin rights, it’s about the agent leaking information it knows from its memories. Sandboxing won’t really help you.
56.
by nojs 3mo ago
I mean CF already forces 5 minutes of motorbike identification on anyone not in a whitelisted western country, so a small percentage of blind people is unlikely to worry them.
57.
by nojs 3mo ago
Not OP, but I’ve been nuked with downvotes for this several times too and tend to delete the dead comments. The slop is so prevalent that at this point it’s not a particularly interesting thing to say I think.
58.
by nojs 3mo ago
> people said this would be too expensive I imagine this is why the filter is so bad. Doing it with an intelligent model that better understands intent would be too expensive, currently.
59.
by nojs 3mo ago
Would you really write “Private video titles aren't just metadata”?
60.
by nojs 3mo ago
Just want to say I really enjoyed your writing style, it’s just the right amount of funny/witty without distracting from the (very interesting!) ideas.

More ›