kennywinker

I don't disagree with that at all, but I do think that line of thinking about ones own issues can lead to a sort of self-blame - thinking things like: if I were stronger I wouldn't have been hurt by that neglect or mistreatment. No question, who you are has a huge impact on how something effects you, but nobody reacts well to e.g. neglect or mistreatment. I might end up with depression, while my sibling ends up with anxiety. That's our predisposition interacting with the objective reality, but neither outcome shifts the blame for the neglect or mistreatment to us. I know that's probably not what you were trying to say, but I do think that idea is one possible place downstream of what you are saying.

For me, making a distinction between the big Trauma with a capital T and little trauma with a lowercase t has been helpful. Any experience that has a lasting negative impact on you could be considered a trauma, but only very extreme stuff - violence, sexual abuse - would count as Trauma.

A car accident is a car accident. Sometimes they are fatal, sometimes the paint gets scratched. People sometimes get long term health effects from a fender-bender.

Interesting. I don’t have a counter to the coining a new term evidence in that article. Potentially that’s solid evidence of the first aspect.

But I do have a counter for the second point:

Context is not a part of the model. It’s an ephemeral blend of input and output fed back in as input.

Online learning is potentially a way for incorporating and evolving coined terms to happen. But since that’s not what any of the models are doing, something would have to change in how we do things before it could happen.

How does that give the llm a way to speak in meaning that humans can’t understand?

You input text into the system, that moves through layers as chains of weights. Sure it doesn’t ever have to be output as text, but for the text-in-text-out machine to have any impact on the world around it, text must be generated.

I’m no LLM lover, i believe they are much more limited in their usefulness than the hype says.

But your example feels like a bad illustration of your point. LLMs operate on words, and can’t read your mind - if you had given me (a human) this task with these prompts i too would have failed because i can’t divine your intent even slightly.

How? Either you prompt it with “make me a ____” or you prompt it with “Generate a non-random juxtaposition”

In the first case you are directing the creativity, in the second you can ask for non-random all you want but the output will be definitionally random based on the training weights and the temperature setting of the prompt.

I’ve never heard of that paper - I went to that example because it’s an example of creativity that most people agree required both a deep understanding of a complex and abstract concept, and unusual and creative thought.

Perhaps I don’t understand relativity and its precursors well enough to see how it was incremental, but either way i’ve yet to see anything remotely like relativity, or even anything like more traditional examples of creativity like picasso or van gogh, come out of an llm

So you’ve claimed that they have half of my precondition for them to speak in ways that we don’t understand.

Coining a term on the fly: yes

Remembering that term and incorporating it into its mind: no.

But, beyond saying they aren’t stochastic parrots you haven’t really given anything to substantiate the idea that they aren’t (very complex) stochastic parrots.

Is there a way for an llm to coin a word, and absorb it into its model? During training maybe… but not after - not the way they’re designed now, anyway.

For it to have new vocabulary we dont understand, it needs to have novel ideas that need words coined for them, and a way to persist those ideas and words into the future. I don’t think that exists.

To me this hypothetical make it clear this won’t happen, not unless there are fundamental changes to what llms are. It doesn’t suggest it will happen. To me, anyway.

Special relativity?

Color me skeptical. LLMs seem to make writing code faster, so of course that means that people can iterate on ideas faster, but I have yet to see actual creative output from an LLM that wasn't coached into it or random juxtaposition.

That law is literally privatized profits, public losses.

Companies get a tax break on risks. If they pan out, they keep the profits and the tax break.

Neoliberalism is a scourge.

Skeptical 4B will ever rival GPT5.5, with its estimated 1.5-10T parameters.

You just can’t compress 1.5T into 4B without losing useful stuff.

But that doesn’t matter that much. If 4B can compose tools, retrieve info, and not get stuck in loops that is going to handle a lot of use cases

There are a few edge models, Spark-X2.5-4B, LFM2.5- in 8b-a1b and 2.6b variants, that are useful for on-device agent-y stuff, but I am doubtful that’s a valuable category. How many iPhone users have never opened Shortcuts in their life? Automating on-phone stuff seems niche.

I don’t expect the economics of local vs cloud ai to change until either the bubble pops or new ai chips land that can run big models fast with low power demand.

The only thing preventing this switch from starting in earnest is the data center buildout monopolizing all current and future GPUs

This is a prediction about the future. It’s not a true fact about the world. For example, software like Jev is betting there is big money in not-very-intelligent intelligence.

Even very llm-pilled coders i know sometimes back away from the “smartest” models, since they aren’t always better at the job at hand, and definitely not when you account for cost.

Based on my experience with running models locally, there is a threshold of intelligence required to be useful. But it’s possible there is also a ceiling where smarter isn’t necessarily better. If you ask a 4B parameter model to fix a bug, it might e.g. fix the bug but fail to fix a compilation error created by the fix. If you ask a frontier model, it might fix the bug, re-write your unit tests, and update the readme. Maybe you wanted those things but maybe you didn’t. “Smarter” is often shorthand for more proactive, and guessing more about your intent. Which is great when it gets it right, and annoying when it gets it wrong.

I suspect smaller models, tuned to a specific task, will do a VAST majority of the llm jobs. High capability huge models will be what humans want to interact with, the bare minimum that gets the job done will be everything else.

At the moment, their main "customer" is investors, not users. But even users want to use the most dangerous model, because dangerous is just another word for powerful.

Sports cars are advertised with how fast they can accelerate from time to time, or tesla's "ludicrous mode" - that is them advertising how dangerous it is.

Notoriously, Ayn Rand only believed deregulation. Her books are just the phrase "deregulation is good" over and over and over.

I think the Ayn Rand thing is more about "the I'm getting mine, f everyone else". However the next 5-10 years shake out, if you're ultra rich today you'll almost definitely be ultra rich in 10 years. Probably ultra-richer. Ayn Rand wasn't just about free markets, she was about individualism over collectivism. Don't worry about how your actions will effect society, worry about what's best for yourself and society will work itself out.

> the motivation to finish a product in India is mostly about tariff bypassing

I don’t think that’s the whole story. Apple started making phones in india in 2017, back when the tariff situation was merely a twinkle in epstein’s best friend’s eye. I think back then it was about general supply chain diversity, more than tariffs specifically. Useful in cases of natural disasters, or trade embargo, and leverage in business negotiations.

By 2024 they were making 36 million iPhones in India, you don’t get to that capacity if you start in november, after the election. Way I see it india was the backup system, and tariffs are the reason they are using the backup system - not the reason they built the backup system in the first place.

> A good metric is the difference in value between the precursor and the final product;

I agree, that’s a useful metric, especially in this case. But it’s also not perfect. E.g. it’s easy to have a >50% “made in USA” product if all the parts are made in china but the last assembly step takes time. Labour in the US is more expensive, so 5-10min of clicking parts together can easily cost more than the parts themselves. You could increase the capacity of that type of manufacturing indefinitely without ever gaining any real independence.

“Made in” has no clean and clear meaning, because our world is so interconnected that everything of sufficient complexity is made from parts that are made from parts that are made in other places.

The iPhones that are “made in india” contain sub-assemblies that are made in china, that contain chips that are made in taiwan, made of metals that were mined in africa, wrapped in plastics derived from middle east oil.

America hasn’t had real antitrust law for >40 years. It should be impossible for a review site to be purchased by a company that makes things reviewed on that site. Reagan era neoliberalism defanged antitrust, and brought on the corpo-hellscape we live in today.

The crux of the issue isn’t hacking - all the big US labs have hacked someone now (https://www.felonybench.com/)and none have faced charges. It also isn’t copyright violations, they all seem equally guilty and equally uncharged.

The issue is openai and anthropic both make similar products in similar ways and yet only one has been designated a “supply chain risk” for entirely opaque reasons that based on timing are probably just “they didn’t kowtow correctly”.

The gov putting their finger on the scales for the company that showed the most deference to them is not a good sign for democracy.

I guess I don't know what "national champion" means. Do they receive different legal treatment from other similar companies? If so, that would be a worrying signal. A common feature of fascism being present doesn't mean you've definitely got fascism, but it does raise a pretty big red flag.

Is this a joke? Volkswagen was literally created by the nazi party.

This kind of partisanship is corruption. The legal system is being corrupted to serve political ideologies.

> Cancelling people for merely trying to open conversation

But you must believe there is a line, no? Sure, open conversation is good, but there are things you can say in open conversation that will make people not want to associate with you. Is that unfair? I don’t think so. Cancellation isn’t bad, imo, unless it’s unfair. Words taken out of context, distorted by mob mentality, that kind of thing. Otherwise, to me “cancellation” is just the consequences of your words and actions.

Sure it’ll be overturned. Tell that to these people:

https://www.theguardian.com/us-news/ng-interactive/2026/jun/...

And while you’re at it? Tell it to Renee Good and Alex Pretti

I know for a fact that it has gotten simpler. Which means that the number of people who can do a version of the thing i’ve been doing most of my life got way way way bigger in the last couple years.

Sure, not everybody can do it - i take your point. But the dynamic between can and can’t has changed. Projects I wouldn’t take on because they were too much work for not enough reward have become something I can pop off in a couple hours. People who know nothing about coding but a lot about their specific need can build complex apps in days.

Sure, but changing the oil means buying a bunch of stuff and getting dirty. Replacing your vibe coded garbage app with my vibe coded garbage app is much much easier than changing my oil.

That said, I think this only applies to apps that can be replaced by vibe coded garbage. Anything with a network effect, or anything that requires deep domain knowledge is much safer