EternalFury

Not at all at first. More and more since the pandemic.

Your estimate is dead on. My first was born in 2005 and my second in 2009. Social media and the early sexualization it brought were a problem for sure. But pretty much every interaction with these devices is messing with dopamine regulation.

Here is an example of reductionism leading to crazy results.

Nauru is 11th on that list. That country is known for 3 things: 1) 90% of the population is either overweight or obese, 2) > 50% of the population smokes, 3) in the 70’s and 80’s, > 90% of the GDP was due to phosphate mining; today, > 90% of GDP is due to refugee processing for Australia.

With LLMs, the richer, the smarter the questions, the smarter the responses.

In fact, the big labs should post-train their models to extract as much information from questioners about their questions before attempting to answer them.

These interaction traces would also be useful during later training runs.

My kids are phone drones. I failed to prevent it. I’ll never forgive myself.

You can look at it this way. Or, you can see it as raising the bar for new entrants. If a new programming language has sufficient merits to overcome a poor level of AI assistance, then nothing will stop it, at least in some circles.

But, if mainstream developers have become totally dependent on AI, it is going to be a high bar indeed.

It would be interesting to bake AI into compilers of any new language, so that auto-correction and education of the programmer happens every time they compile.

In general, though, I think people are too quick to surrender to AI.

The Web is dead. It’s pretty clear future web pages, if we call them that, will be assembled on-the-fly by AI based your user profile and declared goals, interests and requests.

I should have explained my reasoning.

AI coding assistants are closing the door. They need a lot of training data to be good and useful. Established programming languages have a tremendous amount of that, so AI coding agents will do a good job there. C, PHP, Python, Java, etc.

Garden variety and new programming languages have much less examples in the wild, so AI assistants will do a lesser job with those, unless their lineage leads to one of the established languages.

Finally, the bulk of future developers will be lured to AI assisted coding and therefore won’t generate much training data for new programming languages.

A vicious circle.

All of the above is speculation on my part, obviously. Maybe AI assistants will get smart enough to master languages they have not been trained on.

Think of Google as of a tanker ship. It takes a while to change course, but it has great momentum. Sundar just needs to make sure the course is right.

The door for new languages is rapidly closing; unless you are a craftsman and dilettante.

I have been an open source contributor since age 24. I am now 53.

I have seen tremendous contributors get nothing in return except the satisfaction of having helped build something of value (which is huge, mind you).

I have seen opportunists fork, relabel projects as cloud services and tremendously profit in the process.

Since I am mo longer young, or naive, and since I have thought about it for quite some time, I must say it would be good to tie a token offering and a royalty scheme to open source projects. I feel contributors should profit in some measure when big corporations profit from their labor.

I get that. But, let’s say you have a glass, you fill it to one third, then to half, then to three quarter, then to full. Can you expect to fill it beyond full? Not every process has an infinite ramp.

It seems frontier labs have been throwing all the compute and all the data they could get their hands on at model training for at least the past 2 years. Is that glass a third full or is it nearly full already?

Is the process of filling that particular glass linear or does the top 20% of the glass require X times as much water to fill as the bottom 20%?

Many people say:

> these things will get bigger and better much faster than we can learn to discern

I would like to ask “Why?”

Clearly, these models are just one case of “NN can learn to map anything from one domain to another” and with enough training/overfitting they can approximate reality to a high degree.

But, why would it get better to any significant extent?

Because we can collect an infinite amount of video? Because we can train models to the point where they become generative video compression algorithms that have seen it all?

A poor use of the words for sure. “standard originator” would have been better.

I trained an agent that operates as a McKinsey consultant. Its system prompt is a souped up version of:

“Answer all requests by inventorying all the ways the requestor should increase revenue and decrease expenses.”

I give o1 a URL and I ask it to comment on how well the corresponding web page markets a service to an audience I define in clear detail.

o1 generates a couple of pages of comments before admitting it didn’t access the web page and entirely based its analysis on the definition of the audience.

If RFCs were RFCs, it would be easier.

But in history, RFCs became loose or strict standards. From that day on, everyone has tried have to their own RFC stamped; because there’s an economic advantage to being the standard bearer.

So, the process has become more and more convoluted to avoid having as many “standards” as there are interested parties.

Steve Crocker and Jon Postel must be laughing at us.

You must have found a way to add more components, actions, properties, etc. automatically when your data models change. Otherwise, you are changing your UI every time those change. Except that now you have to maintain DTOs, models, actions, APIs, schemas, etc. in many places and across many teams.

Sorry, I have been around the block so many times, there’s now a moat around the block.

I wonder how much of this is actually cinema. Because I just evaluated o1 and it can’t tie its own shoe laces, let alone get out of its sandbox where it plays with drool all over its face.

The marketing stunt seems to be: Portray it as potentially dangerous, as powerful things are inherently dangerous.

Nothing new to what I am going to say and not strictly related to React 19.

I once loved client-side frameworks like React.

Then I realized: Great, we managed to add complexity wherever possible. In the backend, in the front-end, in the DevOps stack, EVERYWHERE. And more layers will be introduced to keep all that distributed complexity in sync.

sigh

Anyone can mix chemicals in a test tube and claim it’s the cure for something. That’s the easy part. Proving it’s safe and effective, that’s what requires a lot of capital expenditure.

This statement is a drastic characterization, but you could say “half a dozen PhDs can form reasons to believe they may have found a cure for something”, and the paragraph would end the same.

I am probably jaded. The way I see it, activisms and politics are occupations of the wealthy. I want to be wrong, but when I observe the political landscape, I find this to be true. I have no interest in being an activist or a politician. I disapprove of many ideologies and policies, and I vote accordingly, but trying to make minds beyond my own has only reduced my quality of life.

I’ll check it again in a year. Their “algorithm” is another polarization device, however. If you click or like anything, your feed fills with extremely similar things. Usually, that makes recovery from a closed mindset impossible.

I get what you’re saying. Political banter is a form of that. But it’s an echo chamber and I am tired of that. I would rather be alone in the woods than having to put up with another echo chamber.

It’s not for me; it may be for you and apparently it’s worth a try for at least 20 million individuals.

I tried it and lasted only 4 days. The platform feels very reactionary, with many users seemingly intent on creating a left-leaning alternative to Truth Social or X.

I was searching for an apolitical, more progressive, and intellectual space, but this didn’t meet those expectations. A significant number of users appear to spend their time copying and pasting tweets to comment on them without engaging directly on X, which I find odd.

For context, I’m not a Truth Social user, and I haven’t used X for several years. I prefer to keep my political views private, confined to the voting booth. My goal was simply to find a platform for sharing and discussing ideas.

The irony is that this rendered pretty poorly on my phone. Ignore me, though, as I like the idea.

If GPT-5 had passed the A/B testing OpenAI likes to do, it would have been released already. Instead, it seems they are clearly concerned the audience would not find it superior enough to GPT-4. So, the bluff must go on until the right cards appear.

No one in their right mind will argue neural nets cannot outperform humans at resampling data they have previously been exposed to. So, digital twins and debate, they probably can do better than any human.

If you "know literally nothing about a programming language", there are two key consequences: 1) You cannot determine if the code is idiomatic to that language, and 2) You may miss subtle deficiencies that could cause problems at scale. I’ve used LLMs for initial language conversion between languages I’m familiar with. It saved me a lot of time, but I still had to invest effort to get things right. I will never claim that LLMs aren’t useful, nor will I deny that they’re going to disrupt many industries...this much is obvious. However, it’s equally clear that much of the drama surrounding LLMs stems from the gap between the grand promises (AGI, ASI) and the likely limits of what these models can actually deliver. The challenge for OpenAI is this: If the path ahead isn’t as long as they initially thought, they’ll need to develop application-focused business lines to cover the costs of training and inference. That's a people business, rather than a data+GPU business. I once worked for an employer that used multi-linear regression to predict they’d be making $5 trillion in revenue by 2020. Their "scaling law" didn’t disappoint for more than a decade; but then it stopped working. That’s the thing with best-fit models and their projections: they work until they don’t, because the physical world is not a math equation.

The Internet is worth trillions to no one in particular.

Just as clean water is worth trillions to no one in particular. Or the air we breathe.

You can take an abstract, general category and use it to infer that some specific business will benefit greatly, but in practice, the greater the opportunity the less likely it is it will be monopolized, and the more likely it is it will be commoditized.

But, my comment was a reference to the magic thinking that goes into making predictions.