Hacker News Reader

Comment by bjt12345 | Hacker News Reader

Serious question though - how are they going with that?

In reality, I see a lot of the current unhappiness in the Tech industry is being triggered by their pointy haired bosses wasting millions on their failed AI strategies only to embark on more failing AI strategies.

View on HN

One example: I've been working for a couple years (not full time) on a high performance FOSS address matcher: https://github.com/moj-analytical-services/uk_address_matche...

Until recently LLMs have been really bad at this task. I always knew it was coming, but with GPT 5.6 they've suddenly become good. It's pretty clear to me that it won't be long before most of my work on this is rendered pointless because the LLM can either do the classification itself (when given agentic access to the canonical list of addresses), or write a classifier itself if given enough labelled data. Of course these two are complementary

Given it is being trained on your project, the latter isn't that surprising. For the former, you could use LLMs yourself for the probabilistic matching as alternative method? Probably you don't because the trade-offs (like performance) are not worth it...

Yes - it's certainly the case at the moment that you can run a few thousand through the LLM at a reasonable price, but not, say, ten million. But the rate of progress suggests to me that this argument won't hold up forever. Eventually I think an off the shelf LLM will outperform most and probably all more traditional ML models at this task. Largely because LLMs can identify tricky ones and pick them out for more intensive effort (e.g. looking online, further searches again the canonical list of addresses)

I didn't check which model exactly you are using, but I think a complex LLM, that is capable of deciding to look online, will probably always be more expensive than most classical models. Maybe if someone invents a way that reliably strips every other ability than answering the one question one has and checking online sources, the LLM can reach an equal level.

If your tool is deterministic, I would rather rely on it, than on LLMs for the task. So I don't think your work will have been pointless, when comparing against an LLM classifying things. Also there is value in something that is battle-tested compared to something just generated on the run, and there is value in something already existing and not needing to be generated or developed anew.

We are too early to deem it failed ai strategies.

I am advising small startups. One of them went entirely to agentic developed software by a team of non technical founders.

They had their sec audit done with considerably fewer errors than had a software developer build the same (which would have taken upwards of 4 times as long).

I am sorry, the technical capabilities of the agentic systems for software development is already much further ahead than what people comprehend.

Severe disruption is imminent.

I agree. There is no point in hiring anyone into a technical-only role anymore. It's very valuable (and probably always will be) to be something-and-technical: a specialist in the target field who also understands how computers and software work. But not "just" technical, these are already completely obsolete and all of them who still have jobs, do so simply because of institutional inertia, connections, and companies being awash with money. Which is not to say that this is not a sustainable state of affairs - it may very well be so forever.

Do you realize that there is software out there that are not simple React apps? The models are clearly advanced but I think the world is a lot more complex than you seem to think.

I have basically vibe-coded a guidance system for air defence interceptors in 6 weeks. It works. Took getting my hands dirty with soldering and some mechanical work to test, but Cursor was guiding me on even that. And i started with understanding concepts like "what is a quaternion" and "why doesn't an infrared camera have a concept of exposure time even as it does have a framerate".

I had a "wow it can vibe code a React app" moment 1.5 years ago. AI evolved a lot since.

I admit i had to hire contractors to do designs for 3D printing. I don't know how to outsource it to AI yet, maybe i can't.

That is super cool. What kind of interceptor? Solid rocket motor with control surfaces?

No, a kind of electric FPV drone.

Claude Fable is very eager to design for 3D printing but I have not tested it real life yet.

Reply on news.ycombinator.com