academic malpractice complaint viable?

I’m curious about this one, legal repercussions against the perpetrator seem pretty clearly viable (it would seem to me that Science has a viable fraud case, and Williams for probably a few too). why wouldn’t anyone do that? is that not the done thing in academia?

the network effects of reputational damage are pretty oof though :|

this shit should be regulated into the dustbin of history. relatedly, the recent release of unredacted docs from the NYT vs OpenAI case showing the clear knowledge of theft and substitution… it’s all willingly done

I don’t know enough about Chinese state security apparatus to comment about that aspect but I’ve seen some mentions of how the state there is also heavily pushing prompts everywhere, and it’s apparently just as shit a failure as it is in the western sphere…so, could be fucked either way

network access control list, but I am now definitely going to start making references to salt circles when I handle the promptfans

but, who the hell gave them $100k for this?

the third name is a vc-type, and I bet it won’t take much scratching to find that linked to our good friends and the ongoing public image adjustment campaign that’s seeing funding from a few corners

I remember this showing up here a while back

think of the monetisation opportunities!

but yeah it’s pretty weird, people are out of ideas for what to do I guess. endless growth mindset can’t have a thing just be done 🤷‍♂️

that dyson toothbrush, huh

Speaking of wrong predictions… the entire US military budget would have to be spent on OpenAI/Anthropic to keep the bubble from bursting.

well that really is rather a tell, isn’t it

I don’t get the second one

“muskin’ the truth” ≈ “kirkin’ the kobayashi maru”

a great sneer found in the wild:

the one titled “interesting” (archive) also smells

idly, I do wonder how well an ocr-only clodjeepeetea would work. every time I look in on all these “agent session” tools they appear to be a very mech turk thing of a phenomenal bit of hidden state, and a tiny little bit of visible distraction. feels like ocr-only might break a bunch of the usual tricks that are used to build on for the shiny demo

the networking part is fine (I do a looooot of networking), but the software quality and choices (esp with containers) is just extremely meh

re the gapping, eh, there are ways. just off the top of my head: rig up the serial console, screen+camera and usb keyboard signalling (which you can data diode on transmission later via usbnet). and that’s before you go for esoteric side channels and utter silliness

but “I turned off spectre and friends mitigation” is just some extremely funny shit

I’d seen that link and didn’t even go looking, but wow this is way worse that I would’ve expected

it’s been a while since I last looked (I usually just deploy whatever network I need) but iirc even the typical desktop config distributions of qemu-kvm are a macvtap or link+nat setup

It then detected I run my host kernel with mitigations=off

I wonder if this person even thought to run a nessus/openvas/whatever scan against the same machine

also, quite seriously, what fucking clowning is this to have this be some “I slapped together a test on my desktop” rather than having a lab bench? is it too hard and costly for trail of bits employees to afford a 1u they can pxeboot for experiments? wild

The standard approach for QEMU is to either wire up a TUN/TAP interface or to use passt.

(e: adding)

The standard approach for QEMU is to either wire up a TUN/TAP interface or to use passt.

other typical modes are bridged (e.g. having various vlans available as bridges and then managing vm exposure to vlan by interface attachment), direct interface assignment (hw offload to the guest)

passt is also commonly used for podman, and it’s moooostly userspace[0] IP and no real good l2 options

1. I had an absolute rager[2] of a time figuring out how to make n-many podman containers[3] spin up into an assortment of vlans for realtime multi-vlan testing
2. of the graaaargh kind
3. don’t do containers, kids. they’re bad for your health

(I’m only half joking. it’s stunning how badly container-management software and tooling all suck)

just wanna re-up one from last week’s stub

@o7___o7 said:

We need a word for “my condolences for that cool person you liked becoming a milkshake duck”

my proposal: “autorowling”

I propose: “autorowling”

Anthropic also said that in testing, auto mode proved safer than manual review — in a study with 1,053 paid testers, auto mode caught 89% of harmful actions, while human review only caught 13.6%. (Perhaps that’s because “manual review can become habitual: users approve 97% of permission prompts in Claude Code.”)

having just yesterday had the displeasure of reading a PR (just shy of 2000 lines) made by Fable 5 (which the promptfondler felt it was most necessary to be specific about), a definite real other problem is simply the literal fucking overwhelm resulting from the slop tsunami

also “harmful” actions sure is doing a lot of legwork there

it started out container-flavoured easier APIs and then grew a bit from there, still relatively liked for its ease of dynamic shittery. seconding @dgerard’s position on apache httpd as ssl terminator. it’s also got some other cases where it’s just a worse performer and/or a nuisance to deal with (in comparison)