No one has certainty here. It’s an emergent technology and no one knows for certain how far it can be pushed.

It’s reasonable that people explore contingencies where the technology does improve to a point of driving changes in the labor market.

Theranos made bold claims too according to wikipedia Theranos claimed they devised diagnostic using blood tests that required very small amounts of blood. Then their PR machine made them raise $9 billion.

Big companies indeed work hard to generate investment and interest. We can take the threat and promise of the technology seriously without taking the science fiction grade hype literally.

If you have your certainty then there are plenty of opportunities to short this space. For the rest of us who don’t have crystal balls, contingency planning (ideally through policy and not as individuals) will have to do.

I do feel with the rise of "reasoning" class of models, it's not hard to believe that code quality will improve over time.

Doesn't sound like improving dramically.

The thing is: how much

0.2x, 2x, 5x, 50x?

Who knows? It just needs to be better than the average engineer.

It needs to be better than the average engineer whose abilities are themselves augmented by AI.

The thing is that this "just" may not happen soon

It just needs to be cheaper than the average engineer, you mean.

It all comes down to a religious faith in AGI or not.

There can't be things that a human can program that AGI can not program or it is not "AGI".

While I am never a true believer in AGI, it seems to go I get a little faith when a new model comes out then I become increasingly agnostic the weeks and months after that. Repeat.

he's not, just another delusional venture capitalist that hasn't bothered to look up the counter arguments to his point of view, done by mathematicians

It’s an emergent technology and no one knows for certain how far it can be pushed, not even mathematicians.

> he's not, just another delusional venture capitalist that hasn't bothered to look up the counter arguments to his point of view, done by mathematicians

Don’t hate on it, just spin up some startup with “ai” and LLM hype. Juice that lemon.

When life gives you delusional VC's that don't understand the basics of what they are investing in, make lemonade! Very, very expensive lemonade!

Queue the Juicero reference!

Sadly my goals are more more ephemeral than opening up a lemonade stand.

Raising on hype != success, most of these lemons will rot before they can even do a small secondary for themselves. Plenty of startups died when the last hype wave ended (which was easy zirp money, before ai). I don't trust anything without a growing revenue anymore, and even then...

I'm not sure, it's an observation considering how AI improvement is related to Moore's law.

[1](https://techcrunch.com/2025/01/07/nvidia-ceo-says-his-ai-chi...)

But some say that Moore Law is dead :)

Anyway, the mumber of tiktok users coorelates with advancements in AI too!

Before tiktok the progress was slower, then when tiktok appeared it progressed as hell!

Yes I see your point, correlation is not causation. Again, this is my best guess and observation based on my personal view of the world and my understanding of data on hands at t0 (today). This doesn't spare if from being incorrect or extremely wrong, as always when dealing with predictions of a future outcome.

That's an assumption. Most/all neural network based tech faces a similar problem of exponentially diminishing returns. You get from 0 to 80 in no time. A bit of effort and you eventually ramp it up to 85, and it really seems the goal is imminent. Yet suddenly each percent, and then each fraction of a percent starts requiring exponentially more work. And then you can even get really fun things like you double your training time and suddenly the resultant software starts scoring worse on your metrics, usually due to overfitting.

And it seems, more or less, clear that the rate of change in the state of the art has already sharply decreased. So it's likely LLMs have already entered into this window.

However, an increase in computing quality doesn't necessarily mean an increase in output quality, as you need compute power + data to train these models.

Just increasing compute power will increase the performance/training speed of these models, but you also need to increase the quality of the data that you are training these models on.

Maybe... the reason why these models show a high school level of understanding is because most of the data on the internet that these models have been trained on is of high school graduate quality.