I think you're conflating the fact that no security measure is infallible with the idea that security measures are therefore largely pointless.
The Binance incident is a good example of why security needs to be layered. Yes, multiple signers approved a malicious transaction because they were all relying on the same compromised signing interface. That's a common-mode failure, not evidence that multi-party authorisation is inherently ineffective. Independent transaction verification and a separate trusted signing path could have made a substantial difference.
Now look at Binance's own report.
The attackers compromised an employee's development machine, leveraged an internal AI tool to gain access to other employees, obtained access to a container deployment permission token on a Wallet DevOps engineer's machine, and ultimately used stolen credentials to deploy a container that reached wallet-signing operations.
Doesn't that raise some fairly fundamental questions about privilege separation, endpoint security, credential management and production deployment controls?
Why could compromising a development environment create a route into such a sensitive production environment in the first place?
There are companies operating infrastructure at an enormous scale that take these boundaries extremely seriously. Google is a particularly good example. Its published production security architecture describes restricted production access, temporary privileges, safe proxies, multi-party authorization, and verifiable builds from peer-reviewed code.
These aren't theoretical concepts. They're established security engineering practices operating at enormous scale.
The point is that security isn't an on/off switch. It's about layers, compartmentalisation, reducing privileges, limiting blast radius and detecting suspicious behaviour.
A compromised developer workstation shouldn't automatically mean a compromised production environment. A compromised employee account shouldn't automatically mean access to deployment credentials. A successful deployment shouldn't automatically mean the ability to perform arbitrary wallet operations.
Sandboxes, isolated development environments, strict application controls, EDR, behavioural monitoring, independently verified deployment artefacts and properly configured audit logging all contribute to that separation.
None of these measures is infallible. A sufficiently sophisticated attacker might eventually circumvent some of them. But that's not an argument for ignoring them.
It's precisely why we implement multiple independent controls: so that defeating one doesn't grant an attacker the keys to the kingdom.
Saying that a nation-state could eventually find another way in is not a security strategy. It's an argument for giving up.
It pains me to say this, but it seems that Google has recently lost interest in search.
Bing literally finds far more links to new stuff than Google, which is really strange because, in the past, Google indexed anything and everything and did so much faster than e.g. Bing or anyone else did.
Google has provided AI overviews for a while now, and the search industry has moved towards not just search results, but also AI summaries, which are sufficient for the vast majority of people, as indicated by a major hit to publishers' traffic from search engines.
The fact that their devops had full access to their production infrastructure and hot wallets is fucking gross negligence.
And don't get me started on their seemingly senior devops having no qualms downloading any code from the net and running it directly on their laptops - that's quite something.
No sandboxes, no isolated environment, no VMs - it's just fucking crazy.
That's not security; it's a complete disregard for security.
Altman said that local inference, and, by extension, people having DRAM in computers at home, is "quite terrible."
He said it specifically about the local inference. Get your facts straight.
Congratulations to Francis Halzen on winning the 2026 Nobel Prize for spending $300 million to drill a cubic kilometer of holes into Antarctic ice, all to prove that high-energy cosmic neutrinos exist -- a fact literally zero physicists doubted in the first place.
What makes the decision almost comedic is that actual groundbreaking observation projects were completely passed over. Giving the prize to IceCube instead of the teams behind the James Webb Space Telescope (JWST) -- which is actively rewriting cosmic history and spotting the earliest galaxies -- or DESI (Dark Energy Spectroscopic Instrument), which is producing the most precise 3D map of the universe to probe dark energy, feels absurd. Instead, the committee played it safe with incremental particle counting.
As Sabine Hossenfelder accurately observed in her latest video, this award sadly demonstrates "just how little benefit physics now brings to humankind." We have reached the point where the Nobel committee has to hand out gold medals for incremental, big-budget data collection because the alternate candidates were theoretical models so far removed from empirical reality they border on metaphysical fan fiction (looking at you, inflation and the multiverse). But hey, at least humankind can now sleep soundly at night knowing that very energetic neutrinos are indeed passing straight through the mattress.
No, the proper headline would be:
"Google Teases Gemini 4 Argon"
It's not been released, and it's not even clear if it will be.
What will really happen is that they will keep their best models for themselves and their dear friends, including the DOD, NSA, CIA and other agencies.
No AI company would even consider slowing down their progress. This would mean losing their competitive edge and becoming irrelevant and losing billions of investors' money.
Weird how people here still think that LLMs are not "intelligent". They may not be meatbags for sure, but they are certainly capable of doing a lot of harm and good. And they can and will replace a lot of people and jobs.
False. The correct term is lose less money. Zero of these AI companies are in the black.
So, your new goalpost for AGI/ASI is being able to "make money"? Thank you for the clarification! By the way, have you ever heard the term "fundamental research"? Many if not the vast majority of things that you now take for granted have their roots in it.
Secondly, many companies that use OpenAI and Anthropic AI are profitable.
It's all quite confusing to me but whatever.
If a person dedicated a few months of research to that, I'd applaud it as an example of human ingenuity and endurance.
Yeah, right, except some of the solved problems are 70-80 years old.
But, it's nor a problem anyone needed to solve. It wasn't something keeping anyone up at night.
You're excellent at moving the goalposts.
why haven't they yet figured out optimizations that'd allow the companies making them actually make money?
The problem is you don't follow AI companies closely and both Anthropic and OpenAI have announced that their latest models have optimized their own code to run faster and more efficiently, which means that LLMs have already allowed them to make more money by reducing the amount of computing power required. And that's just one example of many. A billion-dollar company has already been created by a single person using LLMs. How's that for making money with LLMs? Look up Medvi by Matthew Gallagher.
I too want to believe I'm a special kind of meat bag. But sadly, it looks like LLMs have essentially surpassed 99% of living people on Earth.
There's nothing wrong with being a Luddite. But for some reason for some people, it takes a lot of courage to admit it.
except: why produce anything of value if AI is simply going to suck it up and reserve it to people?
If what you create/publish can be presented as a three-line AI summary, it may not be worthy content.
Recently, the search traffic to my website has decreased by roughly 10 to 20 times, and my well-researched original articles, which have not been published anywhere else, range from three to ten full pages, so Gemini cannot even summarize them properly. Some articles have been translated into multiple languages.
In the meantime Google removed me from the AdSense publisher program a couple of years ago, as they did with most Russian publishers. And they refuse to accept me back, even with a different account from a different country to which I've relocated, so I've not earned a single cent in the past two years.
So go ahead and blame me for creating something with no value.
Everyone's acting like Google stole something uniquely valuable.
The uncomfortable reality is that for most searches, people don't need your version of the answer. They need an answer. And that answer has already been written thousands of times by thousands of websites.
LLMs compress those thousands of near-identical explanations into one response. That's not a bug -- it's exactly what they're good at.
The SEO era rewarded being the 500th person to explain "What is X?" The AI era won't.
If your business depends on repackaging information that's already everywhere, your problem isn't AI. It's that your content was always a commodity.
Going forward, publishers will have to work much harder: produce something an LLM can't cheaply reconstruct -- original reporting, proprietary data, real expertise, or genuinely distinctive analysis -- and build an audience that comes back because it values you, not just the answer to a Google query.
And what if you didn't post photos of your face?
What will you verify in this case? And why? I don't understand you. Did you read the news piece?
Measure twice, cut once.
