Laurent SansonettiFeb 27
Miguel de Icaza ᯅ🍉

What happened at Block will likely happen everywhere else - companies now find themselves with an overstock of talent, and they are not structurally ready to absorb additional capacity.

People will achieve more with less - but this also opens the doors to building entire new businesses.

Feb 27 at 7:52pmWeb
Show threadMiguel de Icaza ᯅ🍉Feb 27

Some folks are under the mistaken impression that our industry builds careful, handcrafted solutions.

Most of the software industry was slop already. AI solutions are an improvement over the previous low quality slop.

Show threadPierreFeb 27
@Migueldeicaza 100% agreed regardless of the chosen interpretation: AI succeeding at lowering slop quality, or AI succeeding at continuing a tradition of slop production 🫠
Show threadJ$Feb 27
@Migueldeicaza Somehow I don’t think fueling a machine to vomit flawed code is a very good answer to the truly very real practice of people writing imperfect, often sloppy software.
Show threadpeter sibleyFeb 27
@Migueldeicaza it’s funny i’ve always found analogies to civil engineering that a lot of software engineering books make laughable as that isn’t really how most software is built .
Show threadChucklesFeb 27

@Migueldeicaza what's painful is that the very few careful, handcrafted solutions are being polluted along with the rest, and people are going to get killed over it.

The Therac-25 only killed a few people. Imagine someone decides to auto-refactor some 30-year-old SCADA software and causes another Bhopal.

https://en.wikipedia.org/wiki/Therac-25

Therac-25 - Wikipedia

Show threadDanielFeb 27
@Migueldeicaza How do we know their layoffs are directly caused by LLM efficiency gains and not simply correlated?
Show threadMiguel de Icaza ᯅ🍉Feb 28
@djh I don’t, but it doesn’t surprise me. The last three months are a whole different game than 6, 9, 12 months ago
Show threadAlex GrebenyukFeb 27
@Migueldeicaza I’m afraid that “it produces low quality code” argument itself is no longer relevant. I don’t know how they pulled it off, but it now mostly produces code I’m fully satisfied with, and I have a pretty high bar. The quality of the products is more about what you build and how you ensure quality. There are way more ways to ensure quality now.
Show threadMiguel de Icaza ᯅ🍉Feb 28
@a_grebenyuk agreed, even the cheap local Chinese models are very good
Show threadLars Marowsky-Brée 😷Feb 27

@Migueldeicaza Ain't that the truth. A lot of GenAI code can be somewhat better than the average out there.

The problem is when it starts creeping into the areas that were significantly above average before. Or when people who have no place making engineering decisions fall for the overselling.

(Plus the various ethical concerns.)

Show threadDimaFeb 28
@larsmb @Migueldeicaza That's what I see at work right now. It always was the case that decisions are mostly made not by engineering roles. Now the space where ICs (individual contributors) can make meaningful decision shrinks faster than melting snow in Sierras
Show threadArcticulateFeb 28
@dima @larsmb @Migueldeicaza I agree: I see this, too, in some types of software-purchase decisions. They base it on mainstream acceptance, popularity and branding (Microsoft M365 and Azure in the Windows sphere, IBM with their mainframes and Red Hat ownership on the UNIX side of things). Then there are tools for log analysis, security scanning, etc. Those two categories both usually have great backends, but a lousy web UI.
Show threadThe Penguin of EvilFeb 27
@Migueldeicaza from what I have seen so far it's mostly worse. It was trained on random GitHub projects and techniques from stackexchange etc. That would result in very grim human code if used for teaching let alone a glorified regexp
Show threadBrian CampbellFeb 28

@etchedpixels @Migueldeicaza unfortunately, I've seen human code that was worse than the average slop you get out of an LLM.

Lots of folks just stringing together bits of what they found off Google and pounded together until it kind of worked on the happy path.

Show threadMiguel de Icaza ᯅ🍉Feb 28
@etchedpixels modern models are really far apart from this, they are trained on assorted interactions. I am in awe. They are also great at understanding existing code
Show threadGlyphFeb 28
@Migueldeicaza no argument that AI did not invent low-quality software, but, citation needed on the “improvement” bit? my personal experience and every bit of quantitative analysis I have managed to dig up (which is, admittedly, of uniformly low quality, but speaks pretty clearly in aggregate) seems to suggest otherwise.
Show threadMiguel de Icaza ᯅ🍉Feb 28
@glyph I can vouch for the hype. It is insanely good. The world is vastly different now than it was 3, 6, 9 ago.
Show threadGlyphFeb 28
@Migueldeicaza Unfortunately I cannot respect "vouching" as evidence, even from people I would otherwise implicitly trust, because the one thing that LLMs seem *amazing* at is hijacking the subjective assessment of their performance. They've even fooled me a dozen or so times, which is one reason I've sworn off of them for good. Have you seen a real-world impact with a preregistered rubric, independent of any subjective judgement?
Show threadMiguel de Icaza ᯅ🍉Feb 28
@glyph every day.
Show threadGlyphFeb 28
@Migueldeicaza Very mysterious! I remain skeptical but I look forward to seeing the results thereof at some point
Show threadMiguel de Icaza ᯅ🍉Feb 28
@glyph oh sorry I misunderstood you. No, I haven’t seen one, I don’t follow that space close enough.
Show threadGlyphFeb 28
@Migueldeicaza Ah. OK. Bummer! If you ever manage to do that sort of measurement I'd be curious, but I understand that this is pretty expensive science that there's not much of a budget or motivation to do in most places
Show threadjwzFeb 28
@Migueldeicaza @glyph Certainly one of the ways that the world is different than it was a couple of years ago is that you, Miguel, have become willing to overlook, for your personal perceived momentary convenience, a stultifying level of harm that these tools cause, and have become one of those "but Magda Goebbels made a great strudel" people.
Show threadtxtx 🇪🇺Feb 28

@Migueldeicaza It's not the first time that the big companies got excited about doing more with less. The ones who went all-in mostly didn't make it, even if they survived as a zombie corp. It was too late.

I'm talking about the outsourcing hype.

Show threadtxtx 🇪🇺Feb 28
@Migueldeicaza 'Kodak appears to have legitimized outsourcing, leading to what some have called “the Kodak effect”. Senior executives at well known companies in the U.S. have followed Kodak’s example and signed long term contracts worth hundreds of millions of dollars with outsourcing “partners”. A number of highprofile multi-billion dollar “mega-deals” have been signed which has raised awareness even more.' [Information Systems Outsourcing in the New Economy — An Introduction]
Show threadArcticulateFeb 28
@Migueldeicaza Well said! Thereis an abundance of poorly designed software, for sure. No matter if you’re on Windows, Linux, Mac, BSD variants, Android, iOS/iPadOS, tvOS, etc. Can’t speak for watchOS as I don’t have an Apple Watch. I don’t have anything negative to say about BeOS and HaikuOS software, except that I would like to see it gain more native-SDK built software.
Show threadRichard BownFeb 27
@Migueldeicaza yep, but businesses pedalling what exactly? We've kind of proven that software is now so malleable that even entering software as a business model is itself a greater risk than it ever was, because any investment is fraught with the risk of it becoming immediately obsolete.
Show threadRichard BownFeb 27
@Migueldeicaza the loss of SaaS vendors and intermediate software companies of any sort means essentially any business can become its own systems provider. Costs should go down, walls will go up, business silos form. It's the start of a whole new business concept the like of which hasn't been seen since the first business systems in the 50s or 60s.
Show threadSir David NielsenFeb 27
@Migueldeicaza also they blew $64 million on a party last year
Show threadTroed SångbergFeb 27
@Migueldeicaza All this is predicated on Dorsey being truthful. I think they overhired post-Covid, reflected in the declining stock valuation and AI is just something to blame for layoffs while pleasing Wall Street.
Show threadTroed SångbergFeb 27
@Migueldeicaza Reading this thread from one of those laid off have only increased my doubts as to whether things really are as Dorsey claims: https://front-end.social/@ksylor/116140605934899412
Show threadJef PoskanzerFeb 27
@Migueldeicaza I mean, yeah it's going to happen to all the other companies that are high on plausible nonsense generators. Good opportunity for companies that have not gone institutionally nuts.
Show threadZephyrFeb 27
@Migueldeicaza appreciate this take, ideally and maybe optimistically software development jobs will not be reduced in the long term, but done by more and smaller businesses
Show threadRandy EdmondsFeb 28
@Migueldeicaza plus crypto isn't taking over like they hoped
Show threadKevin MillerFeb 28
@Migueldeicaza I had this exact conversation tonight. Just think of all the amazing businesses the laid off developers will create.
Show threadGFeb 28
@Migueldeicaza I've never worked anywhere that didn't have a long and growing backlog of things that needed doing. What are you on about?