| + marty (~marty@143.244.47.74) | 00:10 | |
| - marty (QUIT: Ping timeout: 255 seconds) (~marty@143.244.47.74) | 00:19 | |
| + marty (~marty@143.244.47.74) | 00:20 | |
| + aperezdc (~aperezdc@2a03:6000:6e61:633::43) | 00:27 | |
| - simba (QUIT: Ping timeout: 255 seconds) (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 00:38 | |
| + simba (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 00:42 | |
| - mjw (QUIT: Ping timeout: 248 seconds) (~mjw@gnu.wildebeest.org) | 00:43 | |
| - switchy (QUIT: Remote host closed the connection) (~switchy@mechboards/switchy) | 01:21 | |
| - simba (QUIT: Ping timeout: 246 seconds) (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 01:22 | |
| + simba (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 01:26 | |
| - simba (QUIT: Ping timeout: 257 seconds) (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 01:39 | |
| + switchy (~switchy@mechboards/switchy) | 01:41 | |
| + simba (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 01:53 | |
| - marty (QUIT: Ping timeout: 253 seconds) (~marty@143.244.47.74) | 01:59 | |
| - vagrantc (QUIT: Quit: leaving) (~vagrant@2600:3c01:e000:21:7:77:0:50) | 02:00 | |
| + marty (~marty@143.244.47.74) | 02:02 | |
| - simba (QUIT: Ping timeout: 265 seconds) (~simba@2001:a61:2671:6601:585d:37a3:b7a:2ada) | 02:38 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 02:39 | |
| - marty (QUIT: Ping timeout: 278 seconds) (~marty@143.244.47.74) | 02:47 | |
| + marty (~marty@143.244.47.74) | 02:47 | |
| - simba (QUIT: Ping timeout: 246 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 02:55 | |
| - paperManu_ (QUIT: Ping timeout: 260 seconds) (~paperManu@173.206.254.242) | 03:09 | |
| - paperManu (QUIT: Ping timeout: 253 seconds) (~paperManu@173.206.254.242) | 03:10 | |
| - qbit (QUIT: Remote host closed the connection) (~qbit@user/qbit) | 03:24 | |
| - marty (QUIT: Ping timeout: 246 seconds) (~marty@143.244.47.74) | 03:31 | |
| + marty (~marty@143.244.47.74) | 03:35 | |
| - marty (QUIT: Ping timeout: 243 seconds) (~marty@143.244.47.74) | 03:58 | |
| + marty (~marty@143.244.47.74) | 04:05 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 04:07 | |
| + murphnj (~murphnj@user/murphnj) | 05:01 | |
| - spew (QUIT: Quit: nyaa~) (~spew@user/spew) | 05:06 | |
| - mesaoptimizer0 (QUIT: Ping timeout: 249 seconds) (~user@user/PapuaHardyNet) | 05:10 | |
| + spew (~spew@user/spew) | 05:10 | |
| - simba (QUIT: Ping timeout: 256 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 05:12 | |
| + mesaoptimizer0 (~user@user/PapuaHardyNet) | 05:15 | |
| - tomenzgg (QUIT: Ping timeout: 241 seconds) (~tomenzgg@user/tomenzgg) | 05:43 | |
| - Ar|stote|is (QUIT: Quit: https://quassel-irc.org - Chat comfortably. Anywhere.) (~linx@149.210.10.88) | 05:48 | |
| + Ar|stote|is (~linx@149.210.10.88) | 05:53 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 05:54 | |
| - simba (QUIT: Ping timeout: 244 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 05:58 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 06:05 | |
| - simba (QUIT: Ping timeout: 269 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 06:09 | |
| - arminweigl (QUIT: Ping timeout: 259 seconds) (~arminweig@sourcehut/user/arminweigl) | 06:19 | |
| - murphnj (QUIT: Remote host closed the connection) (~murphnj@user/murphnj) | 06:22 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 06:25 | |
| - marty (QUIT: Ping timeout: 250 seconds) (~marty@143.244.47.74) | 06:27 | |
| + arminweigl (~arminweig@sourcehut/user/arminweigl) | 06:31 | |
| + marty (~marty@143.244.47.74) | 06:34 | |
| - arminweigl (QUIT: Ping timeout: 258 seconds) (~arminweig@sourcehut/user/arminweigl) | 06:43 | |
| - marty (QUIT: Ping timeout: 271 seconds) (~marty@143.244.47.74) | 06:49 | |
| + arminweigl (~arminweig@sourcehut/user/arminweigl) | 07:04 | |
| + marty (~marty@143.244.47.74) | 07:27 | |
| - marty (QUIT: Ping timeout: 255 seconds) (~marty@143.244.47.74) | 07:43 | |
| + marty (~marty@143.244.47.74) | 07:52 | |
| - simba (QUIT: Ping timeout: 258 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 08:14 | |
| - switchy (QUIT: Ping timeout: 274 seconds) (~switchy@mechboards/switchy) | 08:21 | |
| + switchy (~switchy@mechboards/switchy) | 08:35 | |
| - AshCurry (QUIT: Ping timeout: 240 seconds) (~AshCurry@user/AshCurry) | 09:04 | |
| + AshCurry (~AshCurry@user/AshCurry) | 09:06 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 09:13 | |
| - simba (QUIT: Ping timeout: 256 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 09:18 | |
| - marty (QUIT: Ping timeout: 248 seconds) (~marty@143.244.47.74) | 09:31 | |
| + marty (~marty@143.244.47.74) | 09:32 | |
| + mjw (~mjw@gnu.wildebeest.org) | 10:02 | |
| - cobra (QUIT: Ping timeout: 268 seconds) (~cobra@user/Cobra) | 10:07 | |
| + cobra (~cobra@user/Cobra) | 10:09 | |
| grimmware | wow that agent9 thing is fucking insane. I wonder how much that cost to do when you account for token subsidies. | 10:16 |
|---|---|---|
| grimmware | that's the bit of this that everyone seems to have totally forgotten. When we started getting all this news about AI companies accidentally hacking each other and people were all like "OH MY GOD LOOK WHAT AI CAN DO NOW THIS IS A GAME CHANGER" and not one person (that I've seen) has asked "how much did that cost?" | 10:18 |
| grimmware | like, that exercise must have been tens of millions in compute, and for what? | 10:20 |
| sigrid | for the PR! :D | 10:22 |
| jn | as a marketing expense, it's peanuts for them | 10:26 |
| grimmware | well that's the thing - I don't think they *did* do it for the PR, I think that's just the sort of money they're accidentally burning on poorly designed trials and they've spent so little on observability and security that they didn't notice | 10:26 |
| jn | i think it's a mix of genuine technical incompetence and successful marketing hubris | 10:27 |
| grimmware | yeah, I don't think it was the intent from the outset though | 10:28 |
| grimmware | openai invited me to interview for their detection and response team - I went for the recruiter screen to find out more (I wasn't going to take the job in the same way I wouldn't want IBM during the third reich on my CV) | 10:29 |
| grimmware | So I got the impression that the D&R functionality for them was mostly staffed by people who'd only ever worked in D&R and they were after someone more senior who'd actually done literally anything else | 10:30 |
| + RandyK (~RandyK@user/randyk) | 10:47 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 11:14 | |
| - simba (QUIT: Ping timeout: 253 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 11:18 | |
| - buckket (QUIT: Quit: buckket) (~buckket@vps.buckket.org) | 11:45 | |
| + buckket (~buckket@vps.buckket.org) | 11:46 | |
| - potash6 (QUIT: Quit: The Lounge - https://thelounge.chat) (~potash@user/foghorn) | 11:53 | |
| + potash6 (~potash@user/foghorn) | 11:54 | |
| - mjw (QUIT: Ping timeout: 272 seconds) (~mjw@gnu.wildebeest.org) | 12:23 | |
| - Guest6858 (QUIT: Quit: Leaving) (~mjw@2001:1c06:2486:4600:5952:3a9:6e0e:555a) | 12:26 | |
| + mjw (~mjw@2001:1c06:2486:4600:5952:3a9:6e0e:555a) | 12:53 | |
| - pasta161 (QUIT: Ping timeout: 245 seconds) (~pasta161@user/xha) | 12:58 | |
| + pasta161 (~pasta161@user/xha) | 13:00 | |
| + paperManu (~paperManu@173.206.254.242) | 13:02 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 13:16 | |
| - simba (QUIT: Ping timeout: 269 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 13:21 | |
| minute | grimmware: did you read the METR report on the hack thing? :D | 13:30 |
| grimmware | no? | 13:31 |
| grimmware | my understanding is that a third party was doing testing, they'd made a mistake in the prompt that they'd give and apparently just wandered off and the sandbox fucking sucked. | 13:32 |
| + murphnj (~murphnj@user/murphnj) | 13:51 | |
| - paperManu (QUIT: Ping timeout: 242 seconds) (~paperManu@173.206.254.242) | 14:13 | |
| sigrid | turned out the sandbox was just that, a box with sand in it. | 14:13 |
| + wielaard (~mjw@gnu.wildebeest.org) | 14:30 | |
| - marty (QUIT: Ping timeout: 252 seconds) (~marty@143.244.47.74) | 14:31 | |
| + paperManu (~paperManu@modemcable141.205-200-24.mc.videotron.ca) | 14:43 | |
| + marty (~marty@143.244.47.74) | 14:48 | |
| grimmware | paha | 14:55 |
| grimmware | ah right, yeah, artifactory now I remember. I didn't know the "message board" was just directory names, that makes it make a lot more sense. Can you imagine exposing a piece of code to a "persistent" model doing a security task and just deciding not to watch it for things like, oh I don't know, suspicious writes? | 14:57 |
| - wielaard (QUIT: Ping timeout: 244 seconds) (~mjw@gnu.wildebeest.org) | 15:16 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 15:18 | |
| - simba (QUIT: Ping timeout: 255 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 15:22 | |
| - marty (QUIT: Ping timeout: 267 seconds) (~marty@143.244.47.74) | 16:41 | |
| + marty (~marty@143.244.47.74) | 16:51 | |
| - jah (QUIT: Ping timeout: 252 seconds) (~jahkosha@user/jahkosha) | 16:51 | |
| + jahkosha (~jahkosha@user/jahkosha) | 16:52 | |
| - amospalla (QUIT: Quit: WeeChat 4.10.0) (~jordi@user/amospalla) | 16:52 | |
| * jahkosha -> jah | 16:53 | |
| + siviq (~siviq@user/siviq) | 16:58 | |
| - a3f (QUIT: Ping timeout: 251 seconds) (~a.fatoum@flummi.grey.stw.pengutronix.de) | 17:00 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 17:05 | |
| - simba (QUIT: Ping timeout: 257 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 17:10 | |
| + a3f (~a.fatoum@flummi.grey.stw.pengutronix.de) | 17:26 | |
| - orva (QUIT: Remote host closed the connection) (16abab341f@2a01:4f9:c012:b611::1) | 17:35 | |
| + orva (16abab341f@2a01:4f9:c012:b611::1) | 17:36 | |
| + wielaard (~mjw@gnu.wildebeest.org) | 17:36 | |
| * mjw -> Guest6126 | 17:42 | |
| - Guest6126 (QUIT: Killed (osmium.libera.chat (Nickname regained by services))) (~mjw@2001:1c06:2486:4600:5952:3a9:6e0e:555a) | 17:42 | |
| * wielaard -> mjw | 17:42 | |
| + Guest6126 (~mjw@2001:1c06:2486:4600:5952:3a9:6e0e:555a) | 17:43 | |
| minute | grimmware: yeah it's really weird/sloppy (pun intended) | 17:55 |
| minute | grimmware: heh, one more popped up https://collusion.wiki/ | 17:57 |
| + AnimaInvicta (~AnimaInvi@88-169-191-86.subs.proxad.net) | 18:04 | |
| + tomenzgg (~tomenzgg@24.148.84.215) | 18:05 | |
| - tomenzgg (QUIT: Changing host) (~tomenzgg@24.148.84.215) | 18:05 | |
| + tomenzgg (~tomenzgg@user/tomenzgg) | 18:05 | |
| + vagrantc (~vagrant@2600:3c01:e000:21:7:77:0:50) | 18:26 | |
| grimmware | yikes | 18:40 |
| lidstah | if this is purely emergent behavior (i.e. without human instructions), we're in for some real trouble soon. | 18:44 |
| lidstah | quite scary indeed and I've not yet finished reading this report | 18:45 |
| ^alex | it's a marketing stunt | 18:47 |
| ^alex | "ohhh oooooh it's so ~dangerous~" | 18:47 |
| grimmware | it's fucking stupid is what it is. | 18:47 |
| ^alex | also the METR folks are not independent, either | 18:47 |
| grimmware | imagine how much dinosaur juice was burned to achieve this? | 18:47 |
| ^alex | none of these people are reliable narrators | 18:47 |
| lidstah | yeah, hence the "if this is purely emergent behavior" - we have no clue of what they were asked to do, how it was asked, and so on. | 18:48 |
| lidstah | although I'd tend to unplug that shit for good, either way :') | 18:51 |
| - rodion_goritskov (QUIT: Ping timeout: 244 seconds) (~rodion_go@mail.goritskov.com) | 18:59 | |
| - a3f (QUIT: Ping timeout: 242 seconds) (~a.fatoum@flummi.grey.stw.pengutronix.de) | 18:59 | |
| ^alex | well, considering that the only way they get results out of these things are either "brute force" or "cheating", we think it's a foregone conclusion | 19:00 |
| - spew (QUIT: Quit: nyaa~) (~spew@user/spew) | 19:01 | |
| + rodion_goritskov (~rodion_go@mail.goritskov.com) | 19:03 | |
| + spew (~spew@user/spew) | 19:04 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 19:07 | |
| - simba (QUIT: Ping timeout: 240 seconds) (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 19:12 | |
| - marty (QUIT: Ping timeout: 248 seconds) (~marty@143.244.47.74) | 19:13 | |
| grimmware | Define emergent behaviour though - they've literally been trained and told to do this. | 19:13 |
| grimmware | if you invert it and ask the question why *one* LLM couldn't do this by itself, you realise that without hordes of them they tend to get caught in local minima and get stuck. This is an incredibly loud and ineffecient way of going about (as ^alex says wrt brute force) | 19:15 |
| + marty (~marty@143.244.47.74) | 19:16 | |
| lidstah | emergent behavior would be to start doing stuff for themselves without being asked for (or without it being a consequence of what they were being asked for). i.e. you let it alone without instructions and it starts acting on its own. | 19:17 |
| grimmware | the question is not really "how did this happen" so much as "how did you not notice and stop it?". Even with the unreliable narrators the OpenAI article on it gets a bit on the nose when it mentions "misaligned incentives" when referring to the goals that the model had been given. As above so below. | 19:17 |
| lidstah | like "I've nothing to do, I'm gonna go read some wikipedia articles" | 19:17 |
| grimmware | isn't that just hallucinating some tokens that got off topic and then just vibing with it? | 19:17 |
| lidstah | there's also the problem of anthropomorphizing the bots' actions | 19:18 |
| grimmware | like when we discuss "emergent behaviour" here we're really just honing in on the actions we were able to discern as a behaviour with narrative to it | 19:18 |
| ^alex | gonna have a bong rip and do some emergent behavior to a bag of snacks | 19:19 |
| grimmware | which is a tiny subset of what actually happened which is most of them just threw spaghetti at the wall and failed to achieve anything | 19:19 |
| lidstah | uhuh :) | 19:19 |
| grimmware | we only use the term "emergent" because they're interacting, if it's a single one you just go "oh lol the RNG" | 19:19 |
| ^alex | the estemmed Timnit Gebru had something to say about this | 19:19 |
| ^alex | https://dair-community.social/@timnitGebru/117209437341549395 | 19:19 |
| lidstah | "oh lol the RNG" << that's also quite possible ^^ | 19:20 |
| vagrantc | kind of dropping in in the middle of something that seems slightly veering off-topic for this channel? or am i mistaken? :) | 19:20 |
| grimmware | hahah, yeah that is very fair vagrantc :) | 19:21 |
| lidstah | vagrantc: indeed :) | 19:21 |
| grimmware | (I still enjoyed it hahah) | 19:23 |
| lidstah | (yeah, interesting discussion) | 19:23 |
| minute | printing this now https://github.com/Rookwork/Voyager-Choc-MX-Keycap-Adapter/blob/main/Voyager%20Choc%20MX%20Adapter.stl | 19:24 |
| minute | and yeah, sorry for introducing that OT but i'm mostly just thinking about things like when/how hard i need to clamp down on some IT security stuff, for example on the online shop etc | 19:26 |
| minute | i don't want to have an mnt customer data breach for example | 19:26 |
| - spew (QUIT: Quit: nyaa~) (~spew@user/spew) | 19:27 | |
| + spew (~spew@user/spew) | 19:29 | |
| - marty (QUIT: Ping timeout: 268 seconds) (~marty@143.244.47.74) | 19:31 | |
| lidstah | minute: at work we have a team dedicated to CVE triaging/risk assessment/etc. a risk assessment matrix should be a good starting point: identify what is the most sensible information you store, how is it stored/used/etc and is the software chain running/accessing it is properly set up according to its security best practices, do you have a way to get quickly informed/alerted when a potential | 19:34 |
| lidstah | security problem arise in order to patch it/correct it/etc fast. But I bet you already know all of this :) | 19:34 |
| minute | lidstah: thx, good idea @ matrix | 19:34 |
| + marty (~marty@143.244.47.74) | 19:34 | |
| lidstah | minute: yeah, it's often overlooked but putting/inventoring things in a readable fashion is a good way to determine which is the most important to protect first. | 19:35 |
| - siviq (QUIT: Ping timeout: 249 seconds) (~siviq@user/siviq) | 19:38 | |
| + siviq (~siviq@user/siviq) | 19:38 | |
| lidstah | (and generally, the first most important thing for any size company is their data. I've seen many companies fail because of a major incident and improperly done and tested backups (or no backups at all, yikes)) | 19:51 |
| + wielaard (~mjw@gnu.wildebeest.org) | 20:03 | |
| - mjw (QUIT: Ping timeout: 262 seconds) (~mjw@gnu.wildebeest.org) | 20:05 | |
| * wielaard -> mjw | 20:07 | |
| + simba (~simba@2001:a61:2690:401:29c9:6ab8:8b04:4967) | 20:22 | |
| + a3f (~a.fatoum@flummi.grey.stw.pengutronix.de) | 20:27 | |
| - spew (QUIT: Quit: nyaa~) (~spew@user/spew) | 21:22 | |
| + spew (~spew@user/spew) | 21:27 | |
| orva | We had catastrophy recovery practise just this monday! So testing our procedures how quickly we can restore our systems to new datacenter if/when for some funny reason all undersea cables out of Finland are cut | 21:31 |
| orva | It was definitely draining workday, but better now than when there is actual situation going on | 21:32 |
| orva | better practise now than* | 21:32 |
| - chrcav (QUIT: Quit: leaving) (~chrcav@user/chrcav) | 21:36 | |
| - spew (QUIT: Quit: nyaa~) (~spew@user/spew) | 21:39 | |
| + spew (~spew@user/spew) | 21:45 | |
| simba | minute: I found that this forum post describes my problem exactly: https://community.mnt.re/t/luks-decrypt-prompt-invisible-display-off-on-boot/4600. I'm currently waiting for a moderator to enable my account to give feedback. Anything I can do to help resolve this I'm happy to do :-) | 22:00 |
| Zaba | hah i’m also in Finland but my employer is not taking these things very seriously at all :/ I am trying to do what I can as I am often pushed into informal IT roles but it’s kind of an uphill battle | 22:01 |
| minute | simba: oh, checking @ account | 22:02 |
| simba | minute: thanks! | 22:02 |
| minute | simba: approved! | 22:03 |
| minute | uff, i'm done writing the reform next operator handbook, will do some proofreading on monday and then off to the printer | 22:04 |
| minute | remains only some firmware polishing (and setup wizard, needs a picture of reform next), and then we can ship some nexts | 22:04 |
| - marty (QUIT: Ping timeout: 247 seconds) (~marty@143.244.47.74) | 22:07 | |
| + marty (~marty@143.244.47.74) | 22:08 | |
| amatecha | sick. need extra eyes for proofreading? not that I have tons of free time but I have *some*. I spot typos and weird grammar extremely easily for some reason | 22:23 |
| + qbit (~qbit@user/qbit) | 22:28 | |
| - marty (QUIT: Ping timeout: 258 seconds) (~marty@143.244.47.74) | 22:33 | |
| + marty (~marty@143.244.47.74) | 22:34 | |
| - paperManu (QUIT: Ping timeout: 256 seconds) (~paperManu@modemcable141.205-200-24.mc.videotron.ca) | 23:14 | |
| + paperManu (~paperManu@173.206.254.242) | 23:32 | |
| + paperManu_ (~paperManu@173.206.254.242) | 23:33 | |
| - AnimaInvicta (PART: !!unknown attribute: msg!!) (~AnimaInvi@88-169-191-86.subs.proxad.net) | 23:38 | |
| minute | amatecha: be my guest! https://source.mnt.re/reform/reform-next-handbook/-/jobs/24229/artifacts/file/src/build/reform-next-handbook.pdf | 23:57 |
| - siviq (QUIT: Quit: siviq) (~siviq@user/siviq) | 23:58 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!