> Many benchmarks include the task of analyzing coding agent aptitude tests. This has led to bench-benchmarks comparing LLM test benchmarking methods, e.g. M Sampson (2025), PL Royle (2024). We performed a bench-bench-benchmark analysis using Mythos Ultra Max 9.6 of these bench-benchmarks.
> Methods: We instructed various LLMs to perform the task "Write a prompt instructing a variety of LLMs to write a benchmark for benchmark analysis, including the instructions 'Create a benchmark for...
Perhaps the old recommendation of "Deliver a resume in person, shake their hand" that many of us younguns have rolled our eyes at will actually come back?? How else to prove your humanity to the hiring manager?
A couple years ago I realized that I had somewhat subconsciously made the same assumption. The thing that snapped me out of it was my awe in watching Clickspring (YT) try to recreate the Antikythera Mechanism. That device's complexity and craftsmanship is proof to me that despite the lack of technology, there were some astonishingly smart and resourceful people living thousands of years ago.
Is there an accessible way for a layman such as myself to read about some of these ideas (Really I mean philosophical discussion in general) without having to read entire books? Is there an active HN-equivalent or wiki or something?
Wow, seems surprisingly balanced. I would guess that if it was 50% longer distance than humans would win reliably, and 50% shorter would allow the horses to win?
Yeah, I feel like unless you run a site large enough for google monkeys to write a special case for your site specifically, why not just password protect the entire site but put the password on the login page? Or any other rudimentary captcha I suppose - like the old days.
Doesn't keep out anyone even mildly interested in your site specifically, including scrapers, but at least it blocks googlebot etc.
I've always thought the "strict invitation trees" or vouch trees would be an interesting way to moderate a community, even before the LLM era. A user can vouch for an unlimited number of new accounts, but if more than 10% of the vouched accounts are banned or flagged down the line, the parent voucher acct is also banned/flagged.
Since it creates a tree structure, you can wipe out entire armies of bot/spam/otherwise accounts by following the vouches up the tree.
A "reasoning" LLM is just an LLM that's been instructed or trained to start every response with some text wrapped in <BEGIN_REASONING></END_REASONING> or similar. The UI may show or obscure this part. Then when the model decides to give its "real" response, it has all that reasoning text in its context window, helping it generate a better answer.
I just append something like "Throughout our conversation, keep your responses brief. Avoid emojis, followup suggestions, and other unnecessary commentary." to every starting prompt. Seems to work OK. I'm sure sibling's recommendation of turning down the niceties sliders would work similarly for someone with an account.
Caltopo is great for this. They require a subscription to download (raster) maps but you can cache a bunch of tiles before you leave to get the gist. These days this is one of the very scarce use cases I don't use OSMand for.
The claim (which generally I'm inclined to believe) is that requiring a phone number drastically increases the cost to sending spam. That in turn drastically reduces the spam amount.
> Methods: We instructed various LLMs to perform the task "Write a prompt instructing a variety of LLMs to write a benchmark for benchmark analysis, including the instructions 'Create a benchmark for...