For me, the scariest part is proxy apps doing the crawling from “smart” TVs.
The inefficiency is kind of reassuring: the fact that companies which are claiming to have developed superintelligent AI tools are doing something extremely dumb confirms my (admittedly extremely cynical) worldview. It’s not that they are wasting someone else’s resources, that’s just unethical, we are long past that by now. It’s that they are probably adding two or three orders of magnitude to their own costs by parsing the HTML.
Yeah, it means that botnets are on sale and used by those companies, or middlemen.
I think it’s an open secret that there are millions, possibly billions, of devices infected as modern DDOS just keep growing in numbers when they eventually happen.
I have a deep distrust of any “smart” devices, especially innocuous-seeming things like TVs. I generally keep them in their own VLAN if possible, or cut off their internet access entirely.
But it’s not always possible. In particular, the Wiim streaming thingie only makes sense on the local WiFi…
Talking of security, love that if you follow some AI people on social media this morning, it’s a stream of comments saying that the near-Mythos Chinese open source model GLM-5.3 has been “abliterated” to remove curbs on offensive cyberhacking, closely followed by other comments that say, no, the claim that GLM-5.3 has been “abliterated” is itself a hacking scam.
I dont know why I didnt think to post this here when I saw it, but a University of Chicago Booth School of Business (because of course) professor has written more than 250 articles (25-80 pages each) in just 8 months with the help of AI:
‘In his 2006 book Radical Hope , the philosopher and psychoanalyst Jonathan Lear writes about what it is like to live through a period of profound cultural upheaval. His subject is the last chief of the Crow nation, Plenty Coups. Not long before his death Plenty Coups, who survived to see his tribe confined to a reservation by the U.S. government in 1884, told the story of his life to a white writer, Frank B. Linderman. Lear was “haunted” particularly by an author’s note at the end of Linderman’s biography, where Linderman explains that he was unable to get Plenty Coups to speak about anything that had happened after the buffalo were killed off and the Crow were moved onto the reservation. “When the buffalo went away the hearts of my people fell to the ground and they could not lift them up again,” he told Linderman. “After this nothing happened.”’
…
'Consider our concept of “merit.” Merit might be compared to counting coups for the Crow in the sense that it is fundamental to our society’s capacity to determine what is worthy of honor or shame. Of course, the question of who and what deserves public praise and reward—whether in the form of money, fame, attention or status—is hotly debated, along with the role that merit should play in society more generally. (Are we a meritocracy? Should we be?) These debates underscore the central role that the concept of merit continues to play in our idea of what constitutes a good life. But the so-far unchecked spread of LLMs and associated technologies, which frequently advertise how easily they can be used to pass off a machine’s work as one’s own, does not represent a new way of contesting the role or distribution of merit; it represents a threat to the concept’s intelligibility.
The closest we have come to grasping this problem is in complaints about how LLMs have enabled “cheating,” which is indeed a serious issue for teachers and prize-committee judges. But this is only one way everyday uses of AI will tend to corrode the distinction between work done primarily by a machine and the kind of work that those of us educated before LLMs understood to be worthy of taking pride in as our own. In a world where young people grow up being inundated with offers from machines to complete (or “collaborate on”) every conceivable task, to insist on doing many kinds of work oneself will soon be no different than choosing to use a typewriter today—a hobbyist’s predilection.’
I personally believe merit doesn’t exist has no place in modern society, so AI isn’t changing much for me. The evidence in sociological, behavioral, epigenetics, etc, keeps mounting up and it doesn’t look good for those who believe in merit or have a general liberal view of the world. It’s pretty inhumane how we still distribute wealth based on personal accomplishments and not in an equitable way, sometimes doing the bare minimum for those that drew the short straw.
It’s funny how this word was introduced mockingly in the 1950s, its authors believing a merit base society would end up badly and now is used seriously by everyone.
I think merit is just an example rather than the crux of the argument. I just picked a couple of sections to quote. I don’t know exactly how I view the article other than it made me think a bit.
I suppose a counter-argument could be that this is just a cry of pain from a literary elite over a loss of status. That LLMs are just a tool in the same way that cameras are a tool. Lots of painters were put out of work by cameras. But at the same time modernist painters actually gained status and to a large extent maintain it still. (And a little over a hundred years after it was born, photography began to get some of that artistic status thanks to MOMA, et al.)
Having said that, LLMs are producing words, just like human writers, while cameras are not doing what paint is doing, so maybe the analogy doesn’t hold.
I think speed and scale of transformations could be destabilising.
There’s this theory called the grandmother hypothesis to explain why, unlike most animals, we live beyond our fertile lives, which seems counterintuitive to evolutionary fitness as we become just another mouth to feed. The idea is that we’re worth keeping around because we pass on wisdom and cultural knowledge relevant to specific environments to our grandkids. That might be true in a fairly stable environment.
But as technological change accelerates, at least some of such wisdom becomes obsolete and so the grandkids end up teaching the older generation how to do things, (or just secretly roll their eyes at how grandad can’t even use the latest gizmo).
If it accelerates to the point that it’s making people’s knowledge and experience instantly irrelevant, or very cheap, then there might be some effects. Maybe that’s when “nothing happens” or nothing happens that makes sense in the world as you understood it.
But aren’t LLMs grandads themselves? They are trained on all the old stuff on the internet and in books, but are perpetually out of date, and favor tried and true solutions.
I wonder, for example, how a new programming language can be created in 2026. Without training data, LLMs are useless for a new language; yet without LLMs, we won’t get the training data. And how will we even judge a language’s merits without human programmers? Was Rust the last programming language ever developed? Are we now stuck with C++ and Java forever? Will Anthropic soon invent a proprietary “AI first” language and further upend our industry?
Certain models already speak a more compact and obtuse English, Claude was an example, as a way to deal more information with less tokens. Anthropic even rolled this behavior back into a more verbose but easier to understand English. Not to speak of their inner cave man reasoning
I think they could pick up programming languages instantly, just by reading the docs and keeping a compacted version in their context.
Seems like OpenAI has “stolen” a mathematicians work through his chatting with ChatGPT, used this knowledge on its frontier model training data, and then pushed through the research to solve and publish it before the original mathematician team could do it. The accusations are pretty severe, it seems like nobody else was working on this problem in this way, so it would be highly unlikely that OpenAI would target this problem in this specific matter out of nowhere, and rush to get it out.
After stealing humanity’s knowledge, what’s the problem with a few more chats?
I had a very odd interaction yesterday: my colleague is currently on vacation. And just after he left, some important thing broke on one of our machines, and we had to fix it. Before he left, he had given me access to his workstation, and told me, “if anything comes up, just talk to my Claude”.
So, I ended up talking to his Claude. It felt weirdly personal and intrusive, as there clearly was history between them. I had to explain that I was subbing in for my colleague. He had this interactive assistant setup, where the LLM was doing stuff in the background while it was also talking to me. It was doing analyses unprompted, and monitoring changes and events autonomously…
A very odd experience. I tend to run short, self-contained Claude sessions of clean question/answer loops. My colleague’s setup was more akin to a persistent coworker with their own agenda and knowledge.
When I do use LLMs nowadays, I usually do through a personal instance of Open WebUI, using openrouter as the backend (as openrouter lets you disable data collection, retention, etc).
They introduced a “Memory” feature, much like you see in commercial LLMs, where it compacts tidbits of information about you during the talks and then uses it in new conversations, it’s really creepy seeing it happen in real time, so much so that I disabled. I can’t imagine running a personalized setup like that, but I am sure he might’ve cooked something that works fine in his workflow…
