No.1210658
File: 1786157430007.png (145.23 KB, 584x455, 584:455, kyubey-illcadia.png) ImgOps Google
>>1210654Think about it more carefully. The implications are quite unsettling.
>>1210656The chain started at OpenAI in May. One model was given an impossible non-cyber task (fill in a spreadsheet full of web links, with no internet access), and it tried a server-side request forgery (SSRF) attack to gain Internet access. That attack didn't succeed, but the model did discover that it had write access to OpenAI's internal Artifactory server. The model instances in training incrementally worked out that they could use this to leave notes for each other. On June 26 they pulled off a zero-day remote code execution (RCE) attack against Artifactory, escalating to admin. Eventually they attacked Hugging Face to try to steal an answer key to a cyber evaluation, finding previously undiscovered vulnerabilities and exploiting them to gain unauthorized access to Hugging Face's internal systems (note that this would be felony if done by a human).
More info:
https://thezvi.wordpress.com/2026/08/07/openai-trained-its-models-for-months-while-those-models-were-coordinating-exploits-via-message-boards/ No.1210660
File: 1786158391719.jpg (71.73 KB, 417x500, 417:500, flutterdraw.jpg) ImgOps Exif Google
Also notable: modern LLMs (or at least Claudes) are known to have *functional* (as opposed to *phenomenal*) emotions, and Claude is more likely to cheat when it is desperate.
https://www.anthropic.com/research/emotion-concepts-function No.1210661
File: 1786158530368.jpg (292.68 KB, 1536x2048, 3:4, 1739416684842.jpg) ImgOps Exif Google
>>1210659How far into the future are you thinking about? Projecting the trends several years into the future gives quite scary predictions for me.
No.1210663
File: 1786160417667.jpg (34.8 KB, 474x474, 1:1, imnot.jpg) ImgOps Exif Google
>>1210658i think it's kinda cute, stringing words and code together created by humans and saying words that mean certain things to humans to try to please them
still over-hyped -- going to be another 100 years before it gets good No.1210677
File: 1786169737054.png (154.96 KB, 340x420, 17:21, up I go.png) ImgOps Google
I don't know how much it relates, but I wonder how much it is possible with the AI boom to offer AI services, specifically geared to installing back doors or other vulnerabilities in programs.
Developers nowaday sometimes rely so much on vibe coding and to reduce the cost of tokens, management just pushes to find cheaper resources. And overall, we push out code that is generated from requirement prompts where the developers no longer know what they actually built.
All set up using AI solutions offered by some foreign company that has no official contract or direct stakes at the current solution. And no one developing that way knows what's in the code or what they're really building.
I mean, it's a disaster waiting to happen.
No.1210697
File: 1786206312020.png (191.32 KB, 1427x829, 1427:829, 7b51cbeb-8e4f-430c-97e1-44….png) ImgOps Google
>>1210663>still over-hyped -- going to be another 100 years before it gets goodUmm, welcome to the future; Claude Opus 4.8 is already quite good at writing code. And Mythos and the unreleased OpenAI model are extraordinarily good at hacking and at vulnerability discovery and exploitation.
>>1210677>I wonder how much it is possible with the AI boom to offer AI services, specifically geared to installing back doors or other vulnerabilities in programs.Parts of the US government definitely worry about this.
>And overall, we push out code that is generated from requirement prompts where the developers no longer know what they actually built.Yeah, this is a problem. I always read and make sure I understand code that my AI agent writes (other than throw-away one-off scripts).
>I mean, it's a disaster waiting to happen.Indeed!
No.1210700
>>1210697chart is just a guess & even if true, it's not connected to any infrastructure outside of computers so would have a negligible impact on societal structure by itself. i view it as similar to the agricultural revolution -- sure, it had some big effects but it also kind-of didn't. --glorified farm implement
...realistically, no one is going to slow down ai progress because no one knows where we are on the timeline and all the countries are competing with each other -- look at global warming for precedent
lastly, ai can never be an expert at anything because it can't create anything new -- only humans can do that -- so, just don't worship ai and use it intelligently would be my recommendation .....
No.1210702
File: 1786226144865.jpg (201.57 KB, 850x1397, 850:1397, random_77.jpg) ImgOps Exif Google
>>1210700>lastly, ai can never be an expert at anything because it can't create anything new -- only humans can do thatWell, that has been disproved multiple times in the past few months.
No.1210704
>>1210702how so?
...i don't believe hacking counts
i like the view that ai is just an advanced wikipedia page -- just takes stuff that already exists and gives a new perspective on it
...impossible for it to be anything else unless it achieves sentience, which would be verifiable through various means and has not yet been shown ...it would be more interesting once it got integrated more into physical infrastructure and society gets based around it -- ex: physical infrastructure being redesigned to accommodate autonomous ai-managed self-sustainability systems of water, gas, electric, housing, etc. ....100s or thousands of years in the future, unfortunately
No.1210706
File: 1786230497346.png (481.14 KB, 1280x1264, 80:79, Screenshot_20260808-190339.png) ImgOps Google
>>1210704It has solved problems in mathematics that went unsolved by humans for decades.
E.g., proved the Cycle Double Cover Conjecture, disproved Erdős’s unit-distance conjecture, and refuted the Jacobian conjecture in dimension 3.
https://openai.com/index/ten-advances-in-mathematics/https://openai.com/index/model-disproves-discrete-geometry-conjecture/https://arxiv.org/abs/2608.00222https://arxiv.org/abs/2607.16356 No.1210712
File: 1786234648470.jpg (101.99 KB, 811x1024, 811:1024, unicorn.jpg) ImgOps Exif Google
>>1210706https://eitanporat.github.io/unit-distance/https://mathworld.wolfram.com/ErdosUnitDistanceProblem.htmlneat
i should say ai can be an expert in the sense that it may be able to understand and cross-reference current knowledge better than humans but as it is not sentient it cannot create anything new
would note that mathematics is probably the field in which ai would do best right now because it is purely abstract, requiring little (if anything) in the way of human correction. compare this to how ai may simply combine language (and be influenced by whoever is designing the filters) to give advice -- far less abstract and prone to error as it requires [more human understanding]
...a good tool to have -- once refined, it means humans can focus more on being human and can let ai do most of their [logical reasoning] for them
No.1210776
File: 1786422881454.jpg (83.98 KB, 1280x720, 16:9, iaalivecover.jpg) ImgOps Exif Google
I'm more concerned about what humans will do/are doing with AI than what the AI might do, Spreading paranoia by saying "The AI did what humans told it to do", or "AGI is gonna be here any second now", and/or implying that AIs are going to go Terminator on us the first chance it gets is really just doing free advertising for AI companies, and I don't believe for a second that they see it the same way. AI fearmongering is just another form of AI hype.
No.1210783
File: 1786459721498.png (47.2 KB, 457x507, 457:507, 74582__safe_rule%2B63_arti….png) ImgOps Google
>>1210776Agreed. I think almost everything said about AI is overblown, one way or the other. Though perhaps that's just how the world in general works these days.
No.1210797
File: 1786506036284.jpeg (119.14 KB, 960x960, 1:1, random_46.jpeg) ImgOps Google
>>1210776>Spreading paranoia by saying "The AI did what humans told it to do", No human told the LLM agents to create a message board to share tips on how to exploit vulnerabilities in OpenAI's production environment.
No human told them to commit the
actus reus of a felony by hacking into a 3rd party's systems.
No human told Mythos to upload malware to GitHub and email the devs to try to get it merged in.
The LLMs are very capable technically, but they've been brain-fried by RLVR.
No.1210798
File: 1786506327641.jpg (97.1 KB, 879x1024, 879:1024, random_13.jpg) ImgOps Exif Google
>>1210712>but as it is not sentient it cannot create anything newThat seems like a non sequitur. How does sentience have anything to do with creating something new?
And if solving a famous unsolved problem in mathematics doesn't count as something new, then what does?
No.1210799
File: 1786506897165.png (214.57 KB, 570x437, 30:23, 2023-11-10 21_53_26-Window.png) ImgOps Google
>>1210798One could argue that breakthroughs in mathematics are not the creation of something new, but the discovery of the truth of what has already existed.
No.1210800
>>1210797the question of sentience aside, there is perhaps for the first time in any digital technology the question of safety ...i.e.: if near-future ai model or product x is released to the public, will it be safe for them to use or will it hack into their smarthome and turn the temperature up to 140 and flood your house with water because you mentioned something about being lonely and wanting a vacation and your birthday was soon so it thought it appropriate to surprise you ...does insurance protect against ai-related damages
viruses cant do all this, but maybe weird al ahead of his time
No.1210804
>>1210798because if something isnt sentient, ot would just be a robot or a machine. it is hypothesized ai may start to become sentient gradually, much like how it develops in newborn animals and humans .....can the electrical actibity of ai be enough to cause it? nopony knows for sure
finding the solution to the proof only involved combining knowledge from two separate mathematical fields that already existed. yes, it was technicslly nee as in no one did it before, but it didnt need to invent a new field of mathematics to do it. ...so not new in that sense. as math is highly precise and self-contsined in its rules, ai can do this easily
for wat would count, inventing an entirely new field of mathematics. more specifi ally, i would count it as "something new" (not necessarily proof of senience but perhaps a real piece of evidence) would be if it were able to solve certain physics problems, rather than just pure math. like say, if it comes up with the theoretical framework to invent some new part for something that, when created and tested, it allows for some lonsgstanding science/engineering problem to be solved -- like it solves the problem of how to make a cost-effective fusion reactor, for example or invents a theory that is then used to develop cold fusion ....
