Unit tests actually need to be correct, down to individual characters. Same goes with API calls. The API needs to actually exist.

Contrast that with "high level design, rough outlines". Those can be quite vague and hand-wavy. That's where these fuzzy LLMs shine.

That said, these LLM-based systems are great at writing "change detection" unit tests that offer ~zero value (or negative).

That’s not at all true in my experience. With minimal guidance they put out pretty sensible tests.

Yes and no. They get out all the initial annoying boilerplate of writing tests out of the way, and the tests end up being mostly decent on the surface, but I have to manually tweak the behavior and write most of the important parts myself, especially for non-trivial tricky scenarios.

However, I am not saying this as a point against LLMs. The fact that they are able to get a good chunk of the boring boilerplate parts of writing unit tests out of the way and let me focus on the actual logic of individual tests has been noticeably helpful to me, personally.

I only use LLMs for the very first initial phase of writing unit tests, with most of the work still being done by me. But that initial phase is the most annoying and boring part of the process for me. So even if I still spend 90% of the time writing code manually, I still am very glad for being able to get that initial boring part out of the way quickly, without wasting my mental effort cycles on it.

These tests are also very time consuming to write, with lots of boilerplate that AI is very good at writing.

But code should change. What shouldn't change, if business rules don't change, is APIs and contracts. And for that we have integration tests and end to end tests.

"speaks volumes" lol

Hard to describe, easy to spot.

Some people write tests that are tightly coupled to their particular implementation.

They might have tons of setup code in each test. So refactoring means each test needs extensive rewrites.

Or there will be loads of asserts that have little to do with the actual thing being tested.

These tests usually have negative value as your only real option as another developer is to simply delete them all and start again.

That's what I would interpret the GP as meaning when they use the phrase "change detection" tests.

That is not due to people choice but due to what actual code being tested does.

I think integration tests and end to end tests are much better.

I will start to care when integration tests are failing, because that is an actual bug. Then I will fix the bug and move over.

Try disabling it for this website. It's incredible. The content is difficult to see between all the various ad surfaces. My browser came to a screeching halt.

It's completely at-odds with the strengths of LLMs (fuzzy associations, rough summaries, naive co-thinking).

They will inevitably hallucinate interactions and observations and therefore decrease reliability. Worse, they will inject a pervasive sense of doubt into the reliability of any tests they interact with.

Yes, you are correct that it entirely lays in the reputation of the AI.

This discussion leads to interesting question, which is "what is quality?"

Quality is determined by perception. If we can agree that an AI is acting like a user and it can use your website, we can assume that a user can use your website and therefor it is "quality".

For more, read "Zen and the Art of Motorcycle Maintenance"

I do not want additional uncertainty deep in the development cycle.

I can tolerate the uncertainty while I'm writing. That's where there is a good fit for these fuzzy LLMs. Anything past the cutting room floor and you are injecting uncertainty where it isn't tolerable.

I definitely do not want additional uncertainty in production. That's where the "large action model" and "computer use" and "autonomous agent" cases totally fall apart.

It's a mindless extension something like: "this product good for writing... let's let it write to prod!"

The word "reasoning" is being used heavily in this announcement, but with an intentional corruption of the normal meaning.

The models are amazing but they are fundamentally not "reasoning" in a way we'd expect a normal human to.

This is not a "distinction without a difference". You still CANNOT rely on the outputs of these models in the same way you can rely on the outputs of simple reasoning.

Nothing like that is happening. This false equivalence originates from several types of people:

1. Journalists that want/need to foment the largest possible catastrophe.

2. Political pundits which want/need to discredit some field.

3. Social scientists playing defense.

It was called natural philosophy for Christ's sake.

Modern medicine's forefather was heroic medicine, based on modulating the 4 humors (blood, phlegm, yellow bile, black bile).

We didn't zap into existence with the hard sciences being hard. We made them that way through centuries of intellectual effort, almost all of which turned out to be wrong!

What do you mean by this? the field of psychology is perfectly capable of policing itself, and it's rejected much of the conclusions of its historical predecessors.

> In order for the same to be true for the hard sciences we would need to be failing to replicate experiments which hinge on germ theory, atomic theory, the standard model, etc.

"hard sciences" also fail to produce results relevant to most people. Sure, they can maybe make better batteries, but how can they explain how dysfunctional society is?

EDIT: We can also directly blame the poor communicational skills of "hard sciences" for diet culture. "hard sciences" have failed in their efforts to produce a population that can reason about nutrition in an evidence-backed manner, and this can be traced directly to how scientists choose to present their data.

This is false, an obvious counter-example is an explanation that happens to be coincidentally correct (without being derived from any science)

-

This seems like an attack on the notion of objective truth which while interesting in the abstract, is not particularly interesting to me personally.

How can you establish "correctness" without something like the scientific method? How do you even bind loosely-defined english to verifiable claims to real-life referents without agreeing with others on terms? No, you're chosen simple, comforting delusion over anything resembling objective truth.

> This seems like an attack on the notion of objective truth

Yes, truth is an apriori concept; objective truth is a silly delusion. Coherence is generally a much stronger concept anyway.

I don't see how you can say this, would you prefer to live 200 years ago before hard sciences had started changing peoples lives? Almost every convenience you see around you exists thanks to hard sciences.

My point was not to reject hard sciences so much as to emphasize you can't easily extract the consequences of them in isolation. It's nearly futile to even try. My apologies for poorly articulating this.

There are tons of non-replicable findings way, way further down the stack than psychology, and those tend to have a lot more relying on them than psychology/sociology studies. If you're upset about scientific validity, consider directing your ire to where problems are more likely to actually hurt people -- the "hard sciences."

Nice ad hominem but I'm none of those things. I work in clinical trials, one of the few areas where we actually do have to know things, and a very good empirical demonstration of exactly how incredibly difficult that is.

I am a hard science maths / data science guy, but unlike a lot of my peers I have a great interest in softer reasoning (philosophy, ethics, political science etc). But I am constantly disappointed by how tainted by ideology psychology and psychiatry feel (and economics, but this is a different discussion).

Do you think that psychology and psychiatry are held to the same rigour as harder sciences and should be considered as valid?

If we take those two observations: a) science is done by humans and b) humans have motivations, obviously the way to arrive at truth is to allow for most things to be wrong most of the time. This is the process by which we've learned every single thing about the universe.

I don't know what you mean by "held to the same rigor." I don't think any psychologist on the planet would tell you we understand psychology as well as we understand basic chemical reactions.

I wonder how much of the AI spend is driven by the above situation.

I bet you could use those puzzles as benchmarks as well.