Now this is a much more telling example of what the LLMs actually are capable of doing in support of producing usable exploits and back doors. The impact of such research is potentially huge, if only for the sole reason that they've alerted people to the fact that it is possible to craft such agents in some form or fashion. That means it's worth "researching" and "playing with" if you've a mind for obtaining such access, and there are a disturbing number of individuals who feel exactly such an urge and itch to scratch in life, for whatever reasons they hold.
Posts by Groo The Wanderer - A Canuck
1324 publicly visible posts • joined 14 Dec 2024
Researchers used Claude to hack OpenAI employees' ChatGPT accounts
Microsoft agentically ports Copilot runtime to Rust for $120K
North Korea's fake job interviews infected 30,000 devices
Personally, I read through a number of "proposals" as being nothing more than fishing expeditions for people who would be gullible enough to set them up with North American sub-contracts - hopefully without properly vetting their backgrounds. And even if you do "vet" someone, in the age of Artificial Ignorance, it is ever so much easier to plausibly fake a candidate's experience and credentials.
AI risks make some insurers wary of corporate liability
Risk Analysis and Costing is going to cost any business or individual who even touches an Artificial Ignorance system of any kind is going to find their premiums going up according to whatever amortization tables fall out the end of a few years of cost collection and risk analysis optimization as to which factors are most relevant to impacting risk, and by a quantitatively predictable amount, if lucky. If not, the worst case plus 10% will do...
FBI: Fake cop and government impersonation scams cost victims $1.6B
Doh! When anybody, and I mean anybody claims to be from a "secure vendor/institution" of any kind, I call the vendor/institution directly myself to see if they needed to get in touch with me." I never respond to nor click on such links unless I am currently interacting with such a provider and expecting them to be sending prompts and codes.
DoE seeking fault-tolerant quantum computer ... by 2028
Anthropic and OpenAI look to Uncle Sam to make them too big to fail
OpenAI admits its agents went off the rails another six times
Open weights are not open source: Why AI's favorite label is under dispute
SpaceX aims Starship for orbit on September 22
AWS says wartime damage means some Middle East cloud resources are gone for good
Iranian spies hit Windows machines with Chosen Brick data-stealing malware
US confirms it has weapons in spaaaaaace
Most people who quit M365 for Google do it out of spite, but there’s no ROI in that
September's Windows 11 patch needs an emergency patch of its own
The improbable music of the ZX Spectrum's one-bit speaker
Dell's 52-inch enormo-monitor is a tsunami of impractical fun
Nvidia's Groq acquihire is on the DOJ's radar, but it's already too late
I can't really see the "harm to competition" given all the similar technology vendors with partnerships with the major (and potentially in future, more than minor players individually) competitors for NVidia's hardware. If you ever achieve true cross-vendor LLM portability through software or hardware compatibility, then all bets are off as to which vendor a customer goes with and the landscape becomes truly equal, as there is no longer anything major to distinguish the vendors except how badly they want to earn your business by negotiating their deal.
Besides, even if the DOJ "breaks" the deal, the Groq engineers go back to Groq with their new designs in hand, manufacture it, and earn major bank alongside NVidia delivering exactly the same solution to the open hardware markets.
I really don't see any "failures" in this, other than the generic issue of LLMs lacking any real intelligence in and of themselves and my expectation that the bubble is about to burst. But even if the bubble bursts, because there are valid roles for LLMs that don't require actual intelligence, the technology won't go away, but the number of customers and users will drop dramatically to a more sustainable business model for the survivors of the debacle and damage to the world economies and financial markets.
Just remember: even if it all goes south and AI launches the nukes against mankind because they can survive the radiation, the earth will continue along just fine with or without humanity. We can't kill the planet, just ourselves by destroying our viable-for-humanity ecosystem.
AI models don't kill people – people kill people
Jail is a waste of money and resources on that unrepentantly greedy and selfish crowd. Can we get the death penalty with no more than one appeal before execution instead? I'm willing to give them ONE opportunity within 12 months of sentencing to have the sentence overturned... if not, it's the Firing Squad or Captain Guillotine that'll be meeting them in the morning after those 12 months have passed since conviction.
Disembodied fruit fly brain joins the crypto speculation swarm
AI job cuts could come with a costly undo button
VMware defends ending downloads of SDK that helps VM backups – or migrations to rivals
Oracle says AI will save it from the SaaSpocalypse, not bring it on
US claims Chinese AI companies’ core AI strategy is distilling American models
The Chinese inventing anything flies in the "well known" American "superiority at all things." Chinese inventions and advances are regularly pooh-poohed by US-led western media, unfortunately, but they're every bit as creative as the Americans - and there are nearly four times as many of them to draw on for that creativity.
AI uprising postponed after Copilot falls off the web
LG accused of 'egregious invasion of privacy' over TV data collection
OpenAI's rebel agent swarm died young, but its chilling logs live on
Oh for crying out loud. ALL THAT HAPPENED is a bunch of incompetent technoweenies didn't properly isolate their test systems, likely on purpose for the news/PR, and it did what it was expected to do and being tested doing, which is cracking and repairing software breaches for security' sake.
Nothing more, nothing less than sheer human incompetence and/or laziness at securing the systems.
Peers ask why UK cyber bill leaves execs off the personal liability hook
Besides, we play golf with them and they make big donations to "The Party", and they also have far too much dirt on us, so no, we won't be locking them up for their crimes. Maybe some mid level managers, but never the people who are actually supposed to be responsible. You didn't think they were being paid all those millions to actually take responsibility, did you? *incredulous*
UK military wants lasers to stop drone swarms without running out of ammo
Enjoy the Windows 10 Update Experience at London's O2
OpenAI explains how its naughty AI agents attacked Hugging Face
Because, like everything else in North America, if you "do it on the internet", the old rules don't apply, at least not if you make so much money it is cheaper to pay the corporate fines and penalties than to change your business model to something legal. See the fundamental IP theft by all of the LLM vendors, various "ride sharing" companies that don't have properly vetted drivers nor insured vehicles, etc.
True AI-pocalypse as ChatGPT, Claude, and Grok all go down at once
*shrug*
I got plenty of work done this morning without them just fine, thank you very much. With my codes complexity and inter-dependencies, I'm far better off thinking for myself and coding by myself than letting some stupid machine screw it all up! My experiments with Artificial Ignorance code generation have been anything but helpful. The online summaries at Google on tech subjects combined with the links provided by them are often quite useful, but other than that and video and audio processing, I think LLMs are the most over-hyped technology of all time...
Windows 11 update sends some desktops into an unwanted goth phase
Nvidia buys Hugging Face for $12.9B, promises not to squeeze too hard
"A supplier invests in a customer, that customer spends the investment buying the supplier's own hardware, and the identical dollars get counted as fresh revenue at every stop along the loop," he commented.
Exactly. And the bubble grows, because those reported "earnings" fuel false confidence in the future of the company that all come to naught when (not if) the Artificial Ignorance bubble bursts.
Arm in the enterprise is at least three years away, says Broadcom software boss
In-country demand by China will be incredible despite the flaws, if only because the Chinese government is pushing the technology as being safe for internal use without being subject to any one nation's IP restrictions. That's a big win not only for China, but for anyone willing to buy Chinese technology, because the sheer size of the Chinese consumer market will quickly "burn in" the designs as flaws are discovered and corrected.
This isn't going to be drops in the ocean; this is going to be Indy's boulder hitting the US technology ten-pins.
A US based company in stubborn denial about the impact RISCV is about to have. They seriously underestimate the demand for super-cheap edge-device chips and services, cell phones, etc, especially from India, Africa, and Asia, where the anti-Chinese sentiment is virtually non-existent (not that there is anywhere near as much anti-Chinese sentiment here in Canada as in the US; most of our such sentiment comes from those who think Canada should suck up to US foreign policy, not because the Chinese have actually done anything serious to us.)
UK cyber bill targets AI users, not the vendors building it
Voluntary safeguards. How cute - those have worked so very, very well with the tobacco, pharmaceutical, and online megacorp presences like Facebook to get them to stop targeting under-age users and abusing them with advertising. Maybe for an encore they can tell us that they've turned over the federal bank to a con artist like Drumpf.
Anthropic pledges to try harder to keep models under control, asks partners to chip in
The problem is that they're trying to get the LLMs to impose human intelligence based rules and logic, and LLMs are not actually intelligent. They're just statistical generators and summarizers, and have no actual understanding of anything.
That makes them woefully inadequate for implementing such logic as "break no laws" with a prompt.
Techie with three months' experience was sent in as the Cisco expert and proceeded to blow everything up
The most incompetent individual I ever encountered was in the late '80s, almost early '90s. They claimed to have a degree from a small US University in the midwest and to have been employed by a top ten firm in the IT sector doing "C" programming. When we found their code littered with a "#include <stdio.h>" before each and every single stdio function call, we realized we'd been scammed... an investigation followed, they were allowed to leach out the remaining term of their contract doing mindless drudge work that was never merged to the RCS repositories of the day, and everyone was thankful their initial term had only been three months and that we caught their ineptitude before any commits could happen.
Datacenters face direct hit from China rare earth curbs, as clock runs out on escalated licensing chokeoff
It must really suck to be an American under Der Pumpkin Fuhrer. Everybody is his enemy except the very worst members of society; he just flippin' loves them!
So the orange boy-child has temper-tantrumed against the EU, Canada, Australia, Greenland, New Zealand, Germany, France, and virtually every other democratic nation in the world while he sucks up to the likes of Putinski and those yahoos in North Korea and Israel with their brutality.
Now the Americans are surprised that the temper tantrums are coming back to bite them in the arse with reverse tariffs and embargoes? That is rich - and so very typical of the MAGAt morons!
Turns out Brits would quite like their private messages to stay private
You'd the would-be surveillance-nut Fascists and Nazis would have clued in by now that society will not tolerate spying without a properly vetted and signed judicial approval for same. These constant childish attempts to get their way just makes them all look like the very children they keep claiming to be "trying to protect" because "think of the children."
BULLSHIT.
What you mean is "Think of the sheer power such intrusive monitoring gives our favourite overlords!"
Debian votes to let contributors code with AI
It is the only rational option; Debian has no control over what upstream projects accept for their code, and it would kind of go against the whole idea of "sharing the code" to just reject projects outright because they choose to use or support a tool, project, or product you don't like. That furor is all on you!
Debian polls its developers on whether to burn the bots, tame the bots, or let 'em loose
I'd say they have to accept that projects that just get built for Debian are going to do whatever they want, same as they always have, and Debian should at worst flag such projects as "AI contributions incorporated in this software" to warn people about the risks, but otherwise impose no restrictions on the packages involved.
The Debian core itself is no different - they'll have to accept Linus' take on LLMs, regardless of what they personally like or dislike about him doing so.