neuronic

My hunch is that AI tools are probabilistically good enough to deceive most casual users in the beginning, so beliefs are built on these foundations. LLMs handle simple tasks reasonably well so their perceived 'intelligence' is easily convincing for many users (often managers and executives).

However, this illusion immediately falls apart during actual stress-testing like writing code for a complex, battle-tested production environments. Non-technical folks can effortlessly spin up simple web apps or tweak a frontend so now they conclude that software engineers are becoming redundant.

And even if there are obvious examples of all of this backfiring no one believes it...

My brain was already perfectly conditioned not to pay attention to screens 30 years ago.

Now my JOB is to look at screens, I am confronted with screens in public and I am expected to be responsive on my mobile screen device both in social life and at work.

Sure, I can employ all sorts of techniques to curb screen addiction but it is the same issue as all these weight loss programs. The Yo-Yo-Effect with screens is unavoidable no matter what state of mind you reach. I know people who take week long vacations in silent monasteries with immense benefit to their mental health. That benefit is curb stomped by modern society within days of their return to "normal" life!

This is absolutely not why there are no leading AI, other important silicon tech, or relevant space companies in Europe. To some degree they exist but are all B-Tier in comparison to US/China. You'd be surprised just how lose money can sit in Europe, I guess. Just not the way it needs to be for this.

The financial structure of the EU is nowhere close to enabling these capital devouring endeavors based on lofty future bets. Operating at a loss for years and years is simply unacceptable in European markets and the EU is not authoritarian enough to randomly divert capital based on political orders like China because the EU doesn't try to be a superpower controlling a hemisphere.

Xenophobic from whose perspective...? For Europeans, every EU alternative should be strongly considered to help reduce dependency risks wherever possible.

I work with LLMs extensively and daily and they are very useful. BUT dear god, absolutely nothing about them is intelligent.

If you work at the edge of context you know what I mean. Even within context, if the system was truly intelligent, the way that Euclid was intelligent, why do I need /superpowers and 50 cycles to get a certain implementation right?

Why is the AI not one-shotting obscure but simple business logic cases with optimal code? Whoops pattern never seen before! There is no thought to it, zero. The LLM is just shotgunning token prediction and context management until something sticks. The amount of complexity you get out of language is certainly fascinating and surprising at times but it's not intelligence - maybe part of it?

Sell it as skills or whatever, but all you do every day is fancy ways of context management to guardrail the token predictor algorithm into predicting the tokens that you want.

And why would it materialize? Anyone who has used even modern models like Opus 4.6 in very long and extensive chats about concrete topics KNOWS that this LLM form of Artificial Intelligence is anything but intelligent.

You can see the cracks happening quite fast actually and you can almost feel how trained patterns are regurgitated with some variance - without actually contextualizing and connecting things. More guardrailing like web sources or attachments just narrow down possible patterns but you never get the feeling that the bot understands. Your own prompting can also significantly affect opinions and outcomes no matter the factual reality.

This is a problem with nearly all predictions about the future. Everything is just a linear extrapolation of the status quo. How could a system have predicted the invention of the transformer model in 2010? At best some wild guess about deep learning possibilities.

Or the impact of smartphones in 2003? Sure smart phones were considered but not the entire app ecosystem and planetary behavioral adaptation.

Just goes to show that biology is WAY more complicated than "if you want to prevent X then do Y" - especially at microbilogical scale. Genes influence each other for example, so by up- or downregulating stuff you are interfering in a highly complex, non-linear system with complex consequences.

Just look at this example of a gene regulation network: https://www.researchgate.net/figure/Different-visualizations...

You are responding to someone with a sex-based racist slur as their username.

This is all a great way of describing why art works as a money laundering or tax evasion scheme.

Yes, absolutely but I found it funny anyway!

Congrats, first time I noticeably hear my MBP M4 Pro fans and can feel the temperature through the keyboard.

Amazing work, as always. I love neal.fun

Edit: also good to know that paper airplanes have officially beat the SR-71, F-104 or X-43B with altitude record.

Uh yea, thats because we tax everyone to hell EXCEPT the rich. Wealth inequality is a serious problem and we are moving to catastrophe sooner rather than later on the current path.

It's obvious why the ultra-rich are building bunkers and hide-outs. Those are of course scams by the building companies, as they give a false sense of security, but the idea of what is REALLY going on is obviously out there.

I even curated a list of 6-8 sources in NotebookLM recently, asked a very straight-forward question (which credential formats does OID4VP allow). The sources were IETF and OpenID specs + some additional articles on it.

I wanted to use NotebookLM as a tool to ask back and forth when I was trying to understand stuff. It got the answer 90% right but also added a random format, sounding highly confident as if I asked the spec authors themselves.

It was easy to check the specs when I became suspicious and now my trust, even in "grounded" LLMs, is completely eroded when it comes to knowledge and facts.

That's only really value if the chips are useful and if there are people buying the chips for something they want to do with them.

It's entirely based on the perception that LLM training & inference is here to stay at ever growing scales when the shortcomings of Artificial Dreaming are increasingly scrutinized. Not all businesses want to end up paying refunds to their clients like Deloitte [1] because the LLM hallucinated crap into their reports (and they failed to correct it).

[1] https://www.theguardian.com/australia-news/2025/oct/06/deloi...

What does Ellisons personal wealth have to do with this? The concern is that the circular pattern of shifting money between these companies is artificially inflating the stock market to heights that will crash very very badly when this bubble finally pops.

This is why I am really looking forward to PIDs in the European Digital Identity ecosystem (EUDI) [1]. This works with the OpenID Verifiable Credentials spec built on top of Oauth2. There are open source solutions in the competition for building the EUDI Wallet and the architecture and reference framework is openly accessible [2]. All credentials are kept with the holder (you) at all times. Basically implementation of the EU eIDAS 2.0 regulation, obviously subject to GDPR.

Mandated to be accessible to EU citizens by 2027 when all Member States have developed a Wallet solution.

Not associated but learned through it at work recently, just awesome project and thought I'd share in this context.

[1] https://commission.europa.eu/strategy-and-policy/priorities-...

[2] https://eu-digital-identity-wallet.github.io/eudi-doc-archit...

[3] https://github.com/openwallet-foundation/credo-ts

Probably a "the dose makes the poison" kind of thing? Constant inflammation and exposure to inflammatory agents could eventually raise the likelihood of cell damage in affected tissues, no?

The immune system is highly highly complicated and directed by huge networks of genes and molecules all up- and downregulating each other depending on internal and external factors. If things go "off balance" in this system the consequences could be dire.

You dont want firefighters hosing down your house from the inside when there is no fire anymore either.

In the scale of the universe this is bound to happen, likely infinite times anyway and this is what feels rather weird to me. Not just the perceived "special circumstances" but that independent of the rarity it will still happen many many times and then any conscious lifeform developing technology to realize this be subject to the definition of survivorship bias.

This lie is still being spread in economics classes?

How are you getting these results? Even with grounding in sources, careful context engineering and whatever technique comes to your mind we are just getting sloppy junk out of all models we have tried.

The sketchy part is that LLMs are super good at faking confidence and expertise all while randomly injected subtle but critical hallucinations. This ruins basically all significant output. Double-checking and babysitting the results is a huge time and energy sink. Human post-processing negates nearly all benefits.

Its not like there is zero benefit to it, but I am genuinely curious how you get consistently correct output for a "complicated subject matter like insurance".

LLMs are just a stock price preserving excuse to do layoffs from previous overhiring.

Is there a reason why you feel compelled to compose a comment before reading or even just skimming the source? Why ask questions that are specifically answered by the provided link?

The source specifically tells you the answer.

The way the 2020s are going I would not be surprised if it happens by next Tuesday.

For anyone wondering like me:

What is the 4-2-6 breathing technique?

The 4-2-6 breathing technique is a calming exercise. First, inhale slowly for four seconds. Then, hold your breath for two seconds. Finally, exhale slowly and steadily for six seconds. This technique helps by making your exhale longer than your inhale, which is a signal to your body to relax. It's particularly useful when you need to settle your mind before sleep or if you're feeling anxious and need to steady your nerves.

Source: https://www.calm.com/blog/breathing-exercises-for-anxiety

I tend to agree, especially because the most harming political influencers are NOT (pseudo-)anonymous. But... it's not just a "bad thing" that is happening. It is the foundational destruction of free societies as we know them. Debates and democratic discourse are replaced with hate, oppression and violence.

I believe that social dynamics like shame and consequences are disabled by pseudo-anonymzation. Pretty much the same effect as people becoming more aggressive and vocal in the confines of their cars. You'd never flip off random people in a supermarket as some would do for getting cut off in traffic.

This substack posts a few interesting theories and ideas how that comes to be. However, the most concerning to me is the asymmetric impact of emotional manipulation due to social media enabled network dynamics.

In particular:

Online discussions are dominated by a surprisingly small, extremely vocal, and non-representative minority. Research on social media has found that, while only 3% of active accounts are toxic, they produce 33% of all content. Furthermore, 74% of all online conflicts are started in just 1% of communities, and 0.1% of users shared 80% of fake news. Not only does this extreme minority stir discontent, spread misinformation, and spark outrage online, they also bias the meta-perceptions of most users who passively “lurk” online.

The brain responds to alarmist, negative and distressing information with much higher priority. At the same time, very few radical and extreme influencers can utilize this mechanism, amplified by social media trying to boost ad revenue. Counterfactual information which directly appeals to the biases and psychology of users is posted and wrapped into click-baity designs to maximize attention and revenue. Tribes are forming and very few elite users can steer the information consumption of users - not just what but also how.

This is highly damaging to society and there is no more institutional trust anywhere to retrieve reliable information on which discussions can be based. Everyone selectively chooses their "reliable sources". This is the absolute opposite of how PKI works, it's like everyone just picks the Root Certs they like (for us techies).

This is of course ironic because all studies and knowledge humanity has to offer are a single search prompt away. But it simply doesn't matter if institutional trust is gone and studies are dismissed because they are coming from "woke" or "radical right wing" sources - completely obliterating what we are trying to achieve with peer review and so on.

Nice, while Shinjuku is much larger and more complex than others it has been really intuitive using it in the real world.

Finally going to get a mental model of Jungfernstieg, Hamburg after a decade of living here. Wow.

Yes, now the cheap cooking stoves have touch interfaces which is an OBVIOUSLY bad idea, much worse even than touch buttons in cars. The expensive professional stoves however...