Mastodawn

Baldur Bjarnason Mar 9

LLMs exaggerate and exacerbate existing market and industry dysfunctions. They've hastened media's descent into fabrications and clickbait, accelerated the devaluation of writing and illustration. And in software it has fuelled an existing crisis and exposed a divide at the core of the industry

Baldur Bjarnason

Our current software crisis—we've had a few—has been ramping up IMO since the post-2007 bailouts. Instead of regulating finance, the US let the finance industry take over, which hasn't been great overall, but for software it's meant that "quality" stopped mattering

Well-funded startups capture market share with subsidised products.

Big tech is a cluster of oligopolies and monopolies.

Internal software projects are driven by their potential effects on stock prices

Baldur Bjarnason Mar 9

There is little to no downside to poor software quality and the upside of doing the job well is limited compared to tactics like lock-in, dishonest subscription models, and monopolies

Some corners of the software industry are less affected. Others, such as web dev, are more affected

For example, the stock price for Crowdstrike, even in a stock market affected by the Iran war, is up 12% today over its peak before it

Massive worldwide economic harm, no real consequences

Baldur Bjarnason Mar 9

This has led to a field whose standard practices are a cluster of bad habits and superstition. Most of the ideas of user-centred design are alien to modern devs. Misconceptions about test-driven dev abound.

When devs says that LLMs make them more productive, you need to keep in mind that THIS is what they're automating: dysfunction, tampering as a design strategy, superstition-driven coding, and software whose quality genuinely doesn't matter

Baldur Bjarnason Mar 9

And they are right. LLMs make it easier for devs to do work that doesn't matter in an industry that doesn't care, where the only thing that's measured is some bullshit measure that's disconnected from actual outcomes

Many of those most vocal about the dysfunctions of LLM-coding were ALREADY WARNING ABOUT THE DYSFUNCTIONS OF THE SOFTWARE INDUSTRY BEFORE "AI". The dysfunctions predate this particular bubble and many in software have been concerned about them for years.

Baldur Bjarnason Mar 9

Equally, most those most vocal about the benefits of LLM-coding were bullish about dev before the bubble. They didn't see the flaws of the earlier state of affairs so they don't see what's wrong with magnifying that dysfunction 10x

Hence the divide in the discourse

Both see LLMs as a mechanism for scaling up existing software practices with minimal human observation

One group thinks this'll make the world 10x richer. The other thinks it'll be a catastrophe

Baldur Bjarnason Mar 9

There is nothing either group can say to the other to shift them because the disagreement is down to a fundamental difference in world view

But if you aren't in tech and are wondering which to trust, just ask yourself: do you really think the chucklefucks of tech, the clowns who have been running the show over the past couple of decades, have got coding completely figured out?

/end

@baldur Great thread!

Baldur Bjarnason Mar 9

@1HommeAzerty Thanks!

Aedius Filmania ⚙️🎮🖊️Mar 9

/sarcasm on

Who knew people embracing LLM for code are the same that move fast and break thing.

The people that din't want to work on a page usefull once a year and just remove it along with the tests instead of fixing it.

The same people that updated some dependencies and just put them in production to see if it worked.

/sarcasm off

I have the same conclusion, it the whole industry that must change, and not just go back from using llm.

Neil Kandalgaonkar Mar 9

@baldur There is a really large group of people who think LLMs are interesting tools with new advantages and new risks. Like maybe 80% of the developers I know or work with. And they can be swayed by arguments that it is too risky or the productivity boosts are short-lived or aren’t there.

There is a solid 20% who are expecting Armageddon and picking a side (pro or con). Maybe these are the people who can say nothing to each other?

Cap Ybarra Mar 9

@baldur unfortunate realization:

team "quality matters" has all the talent, but team "race to the bottom" has all the money

Baldur Bjarnason Mar 9

@cap_ybarra Too true.

@baldur I don’t think there has been solid code quality outside of very niche industries since the 80s. And I wasn’t alive for most of that.

I don’t take joy in the plight of the industry but I also think it was enabled for far too long even at the IC level.

The small percentage here won’t shift the tide and the mass who don’t care may in fact get slight quality improvements if tooling evolves.

I’ve seen deterministically ‘safe’ code gen. But it was never productized.

Baldur Bjarnason Mar 9

@dotsie Yeah, I can believe that. I can only speak of the decline that's been visible over my career, but I know many others have pointed out that the flaws seem fundamental to the industry.

Dmitry Borodaenko Mar 9

@baldur @dotsie I'm old enough to have done ISO-9000 training. Through my career, I saw the entire trade of Quality Assurance arise, evolve into Test Automation, get diluted by Test Driven Development, supplanted by Data Scientists leveraging Internet scale experimentation made possible by Big Tech monopolies, and gradually eliminated from product selling point considerations by the time-to-market race to the bottom.

Thomas Depierre Mar 9

@baldur honestly I see it as a problem of "business management". I mean, look at what they keep investing in and the way they take decisions.

Ofc the system produced shit outcomes.

But automating the whole system with the same kind of bullshit generators that were humans before change nothing to how broken the system is.

Gaëtan Perrault Mar 10

@baldur I really like this characterization of the two groups. I have worked with both of these people at different times in my career.

And I honestly think we're spending too much time listening to both camps. Because not only are they not going to agree with each other, those extremes are not going to "accept defeat" in any way.

The productivity people are going to claim that today's problem will be figured out in 6 months. The quality people will find yet another layer to quibble about that will inevitably end with somebody, somewhere failing to write something down.

I think the reality is going to be somewhere in the middle. And it's going to be more complex than either side is currently characterizing. I think we need to spend more time talking about what success looks like. Because ultimately, that's the decision humans need to be making.

Cody Casterline 🏳️‍🌈Mar 9

@baldur re "superstition-drive coding", my favorite term for that for a long while has been:

https://en.wikipedia.org/wiki/Cargo_cult_programming

Rereading the Wikipedia definition in the new context of LLMs is enlightening:

> The term cargo cult programmer may apply when anyone inexperienced with the problem at hand copies some program code from one place to another with little understanding of how it works or whether it is required.

Cargo cult programming - Wikipedia

Baldur Bjarnason Mar 9

@NfNitLoop Yup.

Niels Abildgaard Mar 9

@baldur This is the only time I've seen someone clearly articulate the state of the industry. The way I usually frame this is in terms of wasted creative potential in developers being captured by hypercapitalized companies, making them work on uninteresting and unimportant problems, to enable the company to capture shares of some transient market... while many real, interesting problems remain unsolved.

Niels Abildgaard Mar 9

@baldur Meanwhile governments (at least the Danish one) has been pushing to educate more software engineers, at a doubling rate of around 5 years.

But there is so much untapped potential in the industry already, from lackluster training of new developers (from too few real masters of the craft and lack of interest from companies) and poorly directed effort...

Anyway, glad to see you talk about it.

@baldur I find it interesting that hardware had reached a peak around that time. We had made CPU cores as fast as possible, with nowhere to grow except by adding more cores (or taking on more responsibilities). Which for several reasons the software industry made terrible use of.

In large part because we were busy taking advantage of fast, reliable, omnipresent internet!

This was also the time GPGPUs & SSDs were entering the market.

I think that exacerbated things!

Darker Knight Mar 9

Its a run away game that leads only downwards.

The answer to this peril is to stop playing the game!

Note the difficulties with addictions and cold turkey times.