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

▲

luke5441 5 hours ago | parent [-]

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...

	▲	RobinL 5 hours ago \| parent [-]
		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)