Jonny

Jonny

@fatred.bgp4.life

Nerd. Dad. Shitposter. Zürich/Cambridge/London

285 followers157 following190 posts

@patrick.risky.biz considering this week’s SRB, and today’s news about another OpenAI’s Alignment issues” with the Aussie Govt, my only question is, where is Claire O’Neill?!? We need like a bat signal or something.

@patrick.risky.biz curious to hear your thoughts on this. from what i have read, openai have publically acknowledged that their agent hacked and breached hugging face. if that was a human, thats a crime under the CFAA/CMA etc.

Had that been an employee of openai, the lawyers would have never made this statement, or at least, thrown them to the wolves. This partnership reads like an AI agent oops version of “free credit monitoring”.

if openai were happy to admit culpability here, and no legal repercussions have been observed (as in statutory repurcussions), can we now assume that all crimes committed by AI agents are in fact, not crimes? We have seen crims use LLMs for a while now, but they are already crims; this is different

Jonny reposted

ℹ️ Note: Hosting company Cloudflare is experiencing public DNS resolver issues; the incident affects users with custom DNS and can be resolved using another provider; the company is looking into the matter; incident not related to country-level internet disruptions or filtering

Leet haxor sits down to his LLM prompt. Haxor: I want to attack people in Switzerland. Use out usual lures and translate them into “Swiss” LLM: I got u bruh…

Yes. People in Switzerland speak French, German and even Italian in places. They don’t often interpolate between them within a sentence. At least not whilst sober…

Been trying to exploit that I have a talk in May to force myself further down the golang rabbit hole. I started off with a little idea, and it’s sort of ballooned to half a management system. It’s fun, but I’m an idiot.

Struggling to see how by any measure @thekrebscycle.bsky.social is worthy of this mafia grade nonsense. I understand why America elected Trump, but I don’t think the majority of the people who voted for him that day, did so such that he could try and re-litigate dead and buried conspiracy theories

Jonny reposted

When you study a new skill like network engineering, as you move onto the advanced pieces, you might find you forget the basics - because you're not thinking about them as much. This is normal. You're not stupid. You're not doomed to fail. There's a lot to remember, and it's all part of the journey.

Today I stripped, serviced and rebuilt my new (to me) Mahlkönig Vario. Took me a while to find one of the 260 grind settings that would work for the cheap crap coffee I have here today, and oh boy… Can’t wait to try something decent in here. #coffeenerd

View full thread

When I was doing my first dial in, I used a single basket to save on waste. Normally I would put 14g in this thing. Today I had ti drop it to 11, because the new grinds are sooo fluffy!

Anyways I may have had like 5 espressos today whilst playing with this thing and that’s totally nothing to do with me being wide awake right now. 🫣

Today I stripped, serviced and rebuilt my new (to me) Mahlkönig Vario. Took me a while to find one of the 260 grind settings that would work for the cheap crap coffee I have here today, and oh boy… Can’t wait to try something decent in here. #coffeenerd

There are many guides on what to do about “proper” levelling of the burr set and I have to say the first time I did it, I screwed up the re-assembly, so had to do it again. Now I know what goes where, it’s remarkably easy to do and the difference on the final grind is wild.

Jonny reposted

And there we go; Using Vault Transit Encryption for secure messaging in observable environments. www.problemofnetwork.com/posts/vault-... I wanted to make this one about the temporal workflow messaging, but in reality, there were too many moving parts to put up that distracted from the core post

problemofnetwork.comVault Transit EncryptionThis post has a weird genesis. If I think back to the first time I heard of Transit Encryption I originally thought it was something else, and when I finally understood what the docs were telling me, ...

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

View full thread

and for completeness, 64b packets - the smallest possible on an ethernet network, failing at atound 13.5Mpps, which is 36% line rate.

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

View full thread

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

View full thread

@patrick.risky.biz whatever happened to the cmd.com product line after the Elastic Acquisition? I remember one key feature being ebpf based command execution MFA challenges, but I really can’t see this in the Elastic product offering today. Did it just die?

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

View full thread

Ha. Ok. I was a little premature. Test 2: route the two 25G ports and stateless UDP packethammer: 1514b: line rate (2Mpps) 1200b: line rate (2.5Mpps) 750b: 15.5Gbps (2.5Mpps) imix: 7.7Gbps (2.5Mpps) 64b: 1.6Gbps (2.5Mpps) I think you can see a theme here…

I shall fire up the graph generator kindly produced by the TRex community, and put them all in a follow up post. I feel like there is something about the trex tests that must prevent fast path being enabled for these flows. The MikroTik numbers are _waaay_ higher than this on routed…

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

looking at the block diagram, you can see that there is this passive component that sits between all the ports and the CPU, and so the max bandwidth we can expect between any two ports is in fact 25G full duplex.

Friend of mine asked me to bench test a CCR2004 (25G edition) to use with Fiber7-x2, and so far I’m having some interesting results. TL:DR- it’s not as weak sauce as everyone says it is.

Test 1: Bridge the two 25G ports and stateless UDP packethammer: 1514b: line rate 25G imix: line rate 25G 250b: 24G (11Mpps) 64b: ~10G (13.4Mpps)