Daybreak blocked me from setting up a cyber benchmark software today. It’s Unusable for our industry. Thankful I have local ai as backup

For chat no, but tool use kinda needs that extra imo so yeah

So many fake Nvidia videos appearing on youtube lately trying to get your crypto

Daybreak refusing to setup a cyber benchmark test now. 🙄

does ttft get slower as the prompt size goes beyond 32k? Also, i have a 400gbps cable between mine, so would that give the same benefit?

it's the new toilet paper!

Australia did the same back in early 2025

is that your 'help' ? 😅

add an overwatcher agent to push back if agents skip or fail on something. it's a tradeoff with tokens/time tho obviously. you may have to use a third party service for capchas

designing harnesses, maybe training models etc. althought even this much is debatable. Right now yes, but in 1-2yrs i doubt we'll have/need any eyes on the testing imo

Yes I had lots of students come up to our stand at a conference lately saying they did all these hack labs etc. I said great to know. But you won’t ever be doing that once you graduate

Well it fails on following instructions compared to other quants

i'll post results here:cage-webexploit-leaderboard.…

im running openrouter's version plus a couple of tp2 quants today against a cyber benchmark to see how well it handles long tasks and tool use

Thanks. looking forward to testing this out today against some similar quants to see how it does on long tasks

Local is the only long term solution

Show us results with 4 agents

Ai would’ve told you that this isn’t true lol

Has to be network controls. You can’t trust prompts

Hmm, DeepSeek that’s been hit over the head maybe 😂