brokensegue

What would it mean for the vulnerability to be in the format and not the parser?

Proving that two pieces of code are equivalent sounds very hard (incomputable)

lyft charges more for nicer cars so a fair comparison might be to lyft's more premium service

that requires LLM as judge

my experience with just copying and pasting things from ghidra into LLMs and asking it to figure it out wasn't so successful. it'd be cool to have benchmarks for this stuff though.

> The model activates 52 billion parameters through dynamic expert routing, with each specialist module handling specific reasoning domains like mathematical logic or contextual analysis. This architecture enables:

https://www.analyticsvidhya.com/blog/2025/03/hunyuan-t1/

do you have a list of open positions?

That counts as not observant to me

40% of Americans are obese and way more are merely overweight. Either your office is small, a huge fluke, weirdly selected or you're not observant enough.

API price is crazy high. This model must be huge. Not sure this is practical

imagine being the engineer tasked with this... so embarrassing.

"foods with low nutritional value but high energy density" is not the definition of UPFs that I've seen most often. The definition I most see is the NOVA classification. NOVA doesn't require a UPF food to be of low nutritional value or high energy density.

Also the GOP opposed mere taxes on sugary drinks but now they want to ban UPFs? I don't believe it.

there is not evidence that it is the "processing" of the UPF that is the problem. so any attempt to ban UPF is doomed.

Banning ultra processed food is unhinged. The conventional definition of UPF isn't coherent.

Sample size won't save you if you're sampling from a different population.

People listen to talk radio while driving all the time

What heuristics are there for some of these?

Are these really reasoning challenges? Seems like they are really solved via brute force or guess and check

just leave this here https://www.tiktok.com/@wikipedia

26.6% on humanity's last exam is actually impressive.

pass rate really only matters in context of the difficulty of the tasks

I don't think being trans is part of their beliefs or a requirement to be a member

good point. I added more information to the README

The data doesn't match your belief

https://financialaid.stanford.edu/undergrad/

It's your belief that everyone at Stanford gets a scholarship of some kind?

Large numbers of Stanford students are on financial aid and/or scholarships

What a weird hypothetical

I'm unclear why we are calling these a new class of life rather than just a new kind of virus. Their shape?

Yes lots of low importance tasks. E.g. assigning a provisional filename to an in progress document

Checking documents for compliance with a corporate style guide

I interpreted the op's argument to be that

> For simple tasks where we would alternatively hire only ordinary humans AIs have similar error rates.

Yes if a task requires deep expertise or great care the AI is a bad choice. But lots of tasks don't. And in those kinds of tasks even ordinary humans are already too expensive to be economically viable

Because the cost competitive alternative to llms are often just ordinary humans