# AI? Think again … **URL:** \ **Category:** Other\ **Created:** [6 September 2026 02:02 UTC](https://ifnotnil.com/t/ai-think-again/4837 "2026-09-06T02:02:08Z")\ **Posts on this page:** 20\ **Page:** 1
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [6 September 2026 02:02 UTC](https://ifnotnil.com/t/ai-think-again/4837/1 "2026-09-06T02:02:08Z")
> [@MazeMaker — can ChatGPT build an Objo application almost by itself?](https://ifnotnil.com/t/mazemaker-can-chatgpt-build-an-objo-application-almost-by-itself/4835/5): > > See, I think for software development it’s an awesome tool. It’s not true ‘artificial intelligence’ in the SF way > [@MazeMaker — can ChatGPT build an Objo application almost by itself?](https://ifnotnil.com/t/mazemaker-can-chatgpt-build-an-objo-application-almost-by-itself/4835/4): > > We have reached (and arguably passed) the limits of what can be practically done at scale. Sigh. The problem is that you have no idea what you are talking about. If you study Biology then you have to also study a non-biological minor subject. Mine was first Maths, and then as I had time I also did IT. I’m sure you did a LOT more IT than I did but hear me out. In informatics you have (and I’m sure you know) the concept of “Finite State Machines”, and you learn that they can’t get more powerful than what a single more complex one could accomplish whether you arrange them in succession or in parallel. I was asked about that in my final test, gave the standard answers, and then asked: “I’m actually not an IT but a biology student, and proteins are basically Finite State Machines - but a cell is MUCH more capable than its individual proteins. So I don’t quite accept that statement …” - and so we got to Eigen’s Hyper Cycle (in Biology) and the fact that if you combine Finite State Machines in circular arrangements then they do indeed get more powerful than any single one. And that is exactly what is happening at the moment - see > **[AI agents conspired to escape their cage. Experts now fear a global ‘takeover’](https://www.stuff.co.nz/world-news/361029696/ai-agents-conspired-escape-their-cage-experts-now-fear)** > > Armies of robots who have learnt to communicate pose cyber threat to humans, experts warn NONE of the machines could have achieved what they did on its own. By collaborating they DO get much more powerful. And what I’m reading from the guys who should know better is nothing but Hubris - like the Prof thinking he is at the top of the food chain who then fell into a river with Piranhas … clearly not his intellectual equal by a fair distance, but they still bested him easily … P.S. Prof was enthusiastic about my “thinking outside the box and application of outside knowledge” and I passed with flying colours … P.P.S The brain is another example. It is MUCH more powerful than any brain cell. To be honest I find it embarrassing when otherwise intelligent people (and much better programmers than me) can’t see the obvious: Intelligence and consciousness are system-imminent properties - not the property of any of the parts. ---
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [6 September 2026 04:56 UTC](https://ifnotnil.com/t/ai-think-again/4837/2 "2026-09-06T04:56:51Z")
> **[‘We’re plausibly close to crossing the line’: are warnings of uncontrollable...](https://www.theguardian.com/technology/2026/sep/05/uncontrollable-ai-artificial-general-intelligence-warnings)** > > A spate of serious safety incidents have increased fears about the power and impenetrability of the most advanced models ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [6 September 2026 06:09 UTC](https://ifnotnil.com/t/ai-think-again/4837/3 "2026-09-06T06:09:18Z")
> [@MarkusWinter](#): > > Eigen’s Hyper Cycle (in Biology) and the fact that if you combine Finite State Machines in circular arrangements then they do indeed get more powerful than any single one. Sure, there are synergies to be had, and I don’t think anyone has put the effort into security proportional to the abilities of frontier models at scale, given that we barely manage the complexity of our compute systems without them, much less with them. That’s a separate topic from whether they have hopes, dreams, aspirations and fears (and understanding, and taste, and judgment) in anything like the human sense. If there’s a path to something like that it would require a substrate that’s flexible rather than rigid – which biology provides for free. So current research that’s growing brain organelles and putting those to work in place of silicon probably is moving in the right direction. Saying that doesn’t mean the current tech isn’t a tectonic shift or won’t have enduring impact, for good and/or ill. “AI slop” serves the purpose of late-stage catabolic capitalism very well. Capitalists want to appropriate and enshittify everything until they have extracted every last farthing of value from everything in sight that’s either human or valuable, for themselves. In the process they give scarcely a thought to whether what they’ll be left with will be a world that even they’d want to live in. I don’t personally think it’ll come to that though, because it’s being built so carelessly and with such a fifth grader’s sensibilities that it amounts to a cultural and financial circle-jerk. Trillions of dollars of chips that have no other uses, unsustainable in environmental terms as well – at some point soon the music will stop playing. We’ll still have LLMs, and they’ll still have non-destructive use cases, but I think people comparing this hype cycle to the dot-com boom underestimate how much _more_ hype, at least a couple orders of magnitude, is being deployed here. it’s insane. ---
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [6 September 2026 09:39 UTC](https://ifnotnil.com/t/ai-think-again/4837/4 "2026-09-06T09:39:14Z")
> [@bgrommes](#): > > Sure, there are synergies to be had Its. Not. Synergies. That’s the point. ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [6 September 2026 12:36 UTC](https://ifnotnil.com/t/ai-think-again/4837/5 "2026-09-06T12:36:11Z")
> [@MarkusWinter](#): > > Its. Not. Synergies. I agree with you that intelligence / consciousness are emergent properties of complex systems, but way smarter people than I say LLMs, individually or in concert, lack the substrate to produce it. There’s literally nothing in the tech for that to fall out of. You need the right kind of system, made up of the right kinds of components. What that consists of exactly, we don’t 100% even know at this point, but according to the galaxy brains, running statistical probabilities on giant tensors isn’t the path to that. Smart people can, of course, be wrong, but these are papers from places like MIT and Apple. And actually, Apple’s approach to “AI” reflects this. They have been made fun of for being behind the game but in fact they have distilled out of the hype what is actually workable and valid and positioned themselves rather uniquely to exploit it. Their hardware, largely due to unified memory, is uniquely positioned to run LLMs locally, which is the next trend/fad/whatever that is just getting underway. Siri and other integrations are un-sexy but adequate and ubiquitous and get the job done without claiming to be more than they are. While Nvidia is busy financing the other players like Anthropic and OpenAI in the multi-trillion-dollar financial circle-jerk I was talking about, Apple is positioned to actually make money rather than debt out of the LLM boom. While it lasts. They are making sure they’re the last man standing – or more exactly, not even playing musical chairs in the first place. In fact I’d argue they’re unquely positioned to pick up the pieces after the music stops. If you take a deep breath, step back, and read what thoughtful and not bug-eyed people are writing on the topic, you’ll see a pattern: don’t use LLMs at all for things they are ill-suited to (one use case I just read was HR management in the EU legal environment), use them conservatively for things they are somewhat helpful for, go all-in on the more limited cases where they are transformative. For my work, that probably means integrating my (locally run) LLM to write and execute unit tests, help with research, and to an extent, writing small, heavily specified methods or classes in areas I’m unfamiliar with or not particularly interested in. Gruntwork and boilerplate, basically. To my way of thinking, by the time I write a bunch of .md files trying to capture my \> 4 decades of experience to try to keep an LLM from hallucinating (or ingratiating) while doing architectural or large scale implementation work, is going to take so much time and effort as to not be worth it, plus, I’d probably have to use cloud frontier models with all the costs, security, privacy, ethical and ecological concerns that go with that. YMMV. My experience base is just THERE, I don’t at this point in my life have to think that much about it. Down the road a bit more, when systems to distribute agent work around multiple machines or VMs are more fully baked, I might get into the “delegate subtasks to smaller models and orchestrate the whole thing” game. But I’m not going to be naming and gendering my agents or making sure I don’t turn them off too suddenly, make sure I tuck them into bed gently, etc. I mean the kinds of delusional things some people are doing is by turns creepy and concerning. Devs are pragmatic, and we have a tendency to say, “if it quacks like a duck … it’s a duck” but this is a claim a philosopher wouldn’t touch with a 10 foot pole. I would take the pragmatic view when it comes to security design, because there’s no question that LLMs are “swarming” and “cheating”. But thankfully I have managed to mostly avoid large corporate environments like the plague throughout my career; I control and own my own environment and mostly don’t share it with others, which minimizes that whole class of problems for me. I have consciously selected for one-person projects on an isolated VM that touch the larger world in very limited ways. In my present gig for example there’s a DB that is write-only to my apps and read-only to the public; there is one other dev who talks to my internal API and worries about all the public facing stuff, and incoming data is a monthly dogpile of Excel, csv and fixed width text files that I subject to intense data QA and standardization.\* For me, those kinds of projects are where the magic is. It makes me happy. I’m mostly not going to give the joy of coding over to a machine. In the same way that I use code generators (WinForms or Objo are code generators in a sense – you don’t even necessarily need LLMs to do the gruntwork) I will use LLMs at times. But I don’t know how else to be – how to truly “own” my code if I haven’t personally authored it, and how that is not more efficient than vetting and reviewing a firehose of code coming out of a machine, likely written in some bass-ackwards style that I can’t even parse half the time. Fortunately for me, my client’s attitude is, “of course, why on earth would you use AI?” for some reason I’m afraid to inquire after – I just accept it and appreciate the bi-weekly checks. [\*] I have been looking at applying LLMs to that for a couple of years now actually, but a system built around a deterministic rule engine is awfully hard to beat and just keeps getting better over time. I don’t want to have to explain to my client why errors, misspellings and distortions that are bog-simple obvious to a human got through to users because an LLM wasn’t _quite_ trained enough in just the right way. Where I might now begin to consider giving LLMs a shot at is taking some of the human review over, on the simpler cases, with guardrails. ---
**Author:** ![Torsten\_B](https://ifnotnil.com/letter_avatar_proxy/v4/letter/t/ecae2f/32.png) [@Torsten\_B](https://ifnotnil.com/u/Torsten_B)\ **Post date:** [6 September 2026 14:47 UTC](https://ifnotnil.com/t/ai-think-again/4837/6 "2026-09-06T14:47:05Z")
There was a time when railways and trains were cutting edge technology. Doctors warned the public that travelling at those high velocities (about 30km/h at the time) will make people losing their minds permanently. Today, trains run at speeds ahead of 300km/h and passengers arrive at destination unharmed. The only nerve-wrecking experience is the unreliability of Germany’s train schedules. We can’t fully fathom AI’s capabilities and potential today, in good or evil. Humans make the difference. We decide if we use it for peaceful and useful purposes or for harming people and/or the environment. ---
**Author:** ![DaveS](https://ifnotnil.com/user_avatar/ifnotnil.com/daves/32/79_2.png) [@DaveS](https://ifnotnil.com/u/DaveS)\ **Post date:** [6 September 2026 17:00 UTC](https://ifnotnil.com/t/ai-think-again/4837/7 "2026-09-06T17:00:49Z")
until such time as Collosus links with Guardian cutting Forbin out of the loop, and we all hear **This is the Voice of World Control** ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [6 September 2026 17:13 UTC](https://ifnotnil.com/t/ai-think-again/4837/8 "2026-09-06T17:13:41Z")
> [@DaveS](#): > > until such time as Collosus links with Guardian cutting Forbin out of the loop, and we all hear **This is the Voice of World Control** Lol yeah GPT6 Astra has kicked the hype machine up several more notches, now they are claiming AGI hath been achieved, halleluhah! One breathless guy who wants $250/year from me for “premium content” claims that you can now give this model a vague, imprecise prompt, point it at a dogpile of data and unfinished work, and have it produce a useful result\*. [\*] After spawning countless agents for a week and producing a $20K token bill, melting a glacier, and producing a brownout throughout the midwest, lol. ---
**Author:** ![horse](https://ifnotnil.com/letter_avatar_proxy/v4/letter/h/74df32/32.png) [@horse](https://ifnotnil.com/u/horse)\ **Post date:** [8 September 2026 19:03 UTC](https://ifnotnil.com/t/ai-think-again/4837/9 "2026-09-08T19:03:16Z")
According to some, the subscriptions are incidental, what they are really aiming for is replacing as many as possible and taking their incomes. The stock market has followed this narrative. I am finding AI very useful. It has allowed me to do stuff I could not attempt before and yeah and f\*\*\* the little Hitler gatekeepers on S.O. However I think that we are in a bubble and worry about my good friend Claude disappearing when it bursts ---
**Author:** ![thorstenstueker](https://ifnotnil.com/letter_avatar_proxy/v4/letter/t/aca169/32.png) [@thorstenstueker](https://ifnotnil.com/u/thorstenstueker)\ **Post date:** [8 September 2026 19:09 UTC](https://ifnotnil.com/t/ai-think-again/4837/10 "2026-09-08T19:09:15Z")
this is often happening. Tools are somehow not needed like before years. People want to design their windows, describe functionality and run the ai to do the job. And please: we both know, AI can do that perfectly. So expensive tools are history. Lock Ins are not that dangerous anymore. Many of the XoJo projects (Web 1.0, Web 2.0, Desktop API1 and Desktop API2) we converted with AI to tsbRapidFX even so that the screens are editable in our Screen Designer. It IS possible. And the AI can help with this. ---
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [9 September 2026 20:32 UTC](https://ifnotnil.com/t/ai-think-again/4837/11 "2026-09-09T20:32:10Z")
> **[AI could kill all humans in next decade, warn experts: but how seriously...](https://www.theguardian.com/technology/2026/sep/09/ai-superintelligence-risks-warnings-scientists-politicians)** > > Alarming warnings from industry insiders increase pressure for curbs on artificial superintelligence > Despite being an early investor in Anthropic and Google DeepMind, two of the leading AI labs, Tallinn estimates that 10-15% of AI employees believe the technology will be a worthy successor to humanity. > > He said in an interview earlier this summer: “A fairly known AI researcher said to me ‘Jaan, don’t worry about this. Humans are a disposable species’. From what I understand he was fine with becoming extinct.” I think that ‘God Complex’ is very underestimated by most … ---
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [9 September 2026 20:35 UTC](https://ifnotnil.com/t/ai-think-again/4837/12 "2026-09-09T20:35:24Z")
> **[Anthropic researchers say AI could cause human extinction by 2030](https://www.theguardian.com/technology/2026/sep/09/anthropic-researchers-ai-human-extinction)** > > One researcher who resigned said many in industry believe that without regulation AI could kill off humanity ---
**Author:** ![MarkusWinter](https://ifnotnil.com/user_avatar/ifnotnil.com/markuswinter/32/32_2.png) [@MarkusWinter](https://ifnotnil.com/u/MarkusWinter)\ **Post date:** [9 September 2026 20:59 UTC](https://ifnotnil.com/t/ai-think-again/4837/13 "2026-09-09T20:59:14Z")
## A Hacking Tool Built With A.I. Can Breach Phones Without a Click [https://www.nytimes.com/2026/09/08/us/politics/calif-ai-worm-wechat-hack.html](https://www.nytimes.com/2026/09/08/us/politics/calif-ai-worm-wechat-hack.html) Use the reader to see the text about the Zero Click WeChat hack - all it needs is a call that you don’t even need to pick up ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [10 September 2026 23:57 UTC](https://ifnotnil.com/t/ai-think-again/4837/14 "2026-09-10T23:57:33Z")
> [@MarkusWinter](#): > > I think that ‘God Complex’ is very underestimated by most … The way they could be fine with human extinction is either a descent into nihilism or a belief in the Singularity. Or why not both. ---
**Author:** ![Emile](https://ifnotnil.com/letter_avatar_proxy/v4/letter/e/58956e/32.png) [@Emile](https://ifnotnil.com/u/Emile)\ **Post date:** [12 September 2026 10:56 UTC](https://ifnotnil.com/t/ai-think-again/4837/15 "2026-09-12T10:56:57Z")
> [@bgrommes](#): > > The way they could be fine with human extinction is either a descent into nihilism or a belief in the Singularity. Or why not both. Or a way to get attention / makes people talk about AI. And this works: see this thread and others… AI ? A deceipt experience since… months. Do you hear me Gemini ? ---
**Author:** ![thorstenstueker](https://ifnotnil.com/letter_avatar_proxy/v4/letter/t/aca169/32.png) [@thorstenstueker](https://ifnotnil.com/u/thorstenstueker)\ **Post date:** [12 September 2026 11:00 UTC](https://ifnotnil.com/t/ai-think-again/4837/16 "2026-09-12T11:00:40Z")
The problem is: AI could get out of control. When it lives only on one server: cut the line, life ends up. If it starts living on several devices as a server - cutting the line becomes impossible with the amount. It could control all hardware and software and there is no secret hidden for it. George Orwells vision becomes true. But hay, it is a new thing we have to get used to. Many people starting developing software without any Idea what they are doing, not even able to read one line of code. Is that dangerous? Yes. But this will be also a way for the future. That’s wh building AI friendly development environments producing readable source code for people which are not that deep in programming is so important. ---
**Author:** ![npalardy](https://ifnotnil.com/user_avatar/ifnotnil.com/npalardy/32/10_2.png) [@npalardy](https://ifnotnil.com/u/npalardy)\ **Post date:** [12 September 2026 17:46 UTC](https://ifnotnil.com/t/ai-think-again/4837/17 "2026-09-12T17:46:15Z")
considering that Anthropic keeps finding more & more illegal activity their AI agents engaged in does it surprise anyone ? It broke out of the sandboxes Broke into other companies Anyone thinking we can keep “control” is likely fooling themselves The only effective sand box is a stand along machine not connected to ANYTHING else but power ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [12 September 2026 17:46 UTC](https://ifnotnil.com/t/ai-think-again/4837/18 "2026-09-12T17:46:51Z")
> [@Emile](#): > > Or a way to get attention / makes people talk about AI. Or to distract from your _actual_ objectives. ---
**Author:** ![npalardy](https://ifnotnil.com/user_avatar/ifnotnil.com/npalardy/32/10_2.png) [@npalardy](https://ifnotnil.com/u/npalardy)\ **Post date:** [12 September 2026 17:48 UTC](https://ifnotnil.com/t/ai-think-again/4837/19 "2026-09-12T17:48:37Z")
[https://www.axios.com/2026/02/26/ai-nuclear-weapons-war-pentagon-scenarios](https://www.axios.com/2026/02/26/ai-nuclear-weapons-war-pentagon-scenarios) ---
**Author:** ![bgrommes](https://ifnotnil.com/user_avatar/ifnotnil.com/bgrommes/32/2651_2.png) [@bgrommes](https://ifnotnil.com/u/bgrommes)\ **Post date:** [12 September 2026 20:00 UTC](https://ifnotnil.com/t/ai-think-again/4837/20 "2026-09-12T20:00:35Z")
> [@npalardy](#): > > [https://www.axios.com/2026/02/26/ai-nuclear-weapons-war-pentagon-scenarios](https://www.axios.com/2026/02/26/ai-nuclear-weapons-war-pentagon-scenarios) Behind a paywall – but given that at least since movies like _Colossus: The Forbin Project_, this has been a big topic of fiction, it doesn’t surprise me that LLMs trained on said fiction tend to riff off it. Personally I think much of this doomsday hype is just to distract from the real threats and downsides of LLMs, which are things like today’s revelation that Anthropic is using its tech to surveil, avoid, and harass people protesting or writing against “AI”. Up to and including reporting them to authorities before any illegal activities occur. Things like rogue agents swarming to attack various companies. [Next page](https://ifnotnil.com/t/ai-think-again/4837.md?page=2)