LLMs, and other transformer models before them like BERT for example, have the benefit of "transfer learning."

If you want to make a ticket classification model, you don't need to train across many customers doing ticket classification, who might not want you to train on their data and whose data is impossible to obtain unless you already are a large company. Instead, you can train on a completely different task which forces the model to learn the syntactic and semantic structure of the English language — say predicting missing words in Wikipedia articles, which you can't do successfully without a good understanding of English — and then fine-tune that model on your specific task with much less data.

Because the initial task is so general, you can have one company do the expensive pretraining, share the base model, and then have everybody else just do the finetuning on their (small) proprietary dataset.

LLMs and models like Jev go one step further, letting you use them as classifiers with no training at all, with nothing but rules in the prompt. And with a few tricks, you can indeed use them to get pretty high-quality labels, which you can then distill into a fine-tuning dataset for a much smaller model.

Agree to all of this but I would make a minor correction: the interpretability is true under certain conditions, e.g., you use something like n-grams and train an SVM with a linear kernel. You could change the representation back then too, e.g., LDA (topic models), Latent Semantic Indexing, and of course, you could change the SVMs kernel as well, e.g., use a sequence kernel [1]. Some of these were much less interpretable. But the common pipeline back then was indeed n-grams + Linear SVM. Today it is easy to modernize the pipeline by just changing the representation step to use an embedder like ModernBERT. I have a "one page" classifier if you want to try it out - [2].

[1] I went down the rabbit hole of using sequence kernels quite a bit back then, see https://arxiv.org/abs/1910.13800, but the timing was terrible. RNNs and variants like LSTMs were beginning to become popular, and everyone (including me) switched.

[2] https://pocket-text-lab.abhishek-ghose-82.workers.dev/

And you can now get that world knowledge out of llms and generate synthetic data to train your svm and fine tune on your own dataset!

I have seen Phd in stats validating models with data the models have been trained on. And I don’t live in a country with dubious universities.

So between llm and pseudo experts, at this stage, I prefer to go with a LLM.

> Building an SVM isn't a tricky job, but you probably need a data scientist involved, and to think carefully about a labeled dataset.

Well not anymore... You can just get AI to build it.

non-LLM models just aren't visible to consumers.

Saying that the field got reduced to only LLMs is like saying that the field of combustion engines got reduced to cars. The trucks, combine harvesters and peaker plants are still here, they're just in the background, invisible to people who are not part of the industry.

Banks still use random forests and SVMs, medicine and image processing still use CNNs (and people who take photos with their phone don't even realize how much image processing is being done), TTS, ASR and predictive typing models are in wide use, and there's probably still plenty of BERT in spam classification and such.

I was thinking about this recently as well. Smaller, specialized models that can run on device just make so much more sense to me than these huge generalized models that need huge infrastructure run.

I am not sure a best practice in one area translates nicely to another. LLM's power comes with it being a monolith.

> single model that can do everything

"AGI"

You can can't build an SVM-classifier for a ticketing system without domain data. You can ask an LLM to do it, and it will do it very well.

With the LLM you don't even need a data scientist to train the model, you just need someone who can connect to APIs. And the LLM is fine with adding and removing categories. No retraining.

It makes the initial faces incredibly fast. Then it is slow and expensive to run, so once the system settles, it becomes the target for model optimization, e.g, a dedicated, or semi-dedicated model.

I can spin up a classification solution with an LLM in ca. 5 minutes that is likely to do an okay job. To build a custom classifier, I need lots of training data first.

That is the reason.

Once you then have a system that is used enough, it starts making sense to look to replace that with a leaner classifier, but LLMs have a place in vast number of places where we'd previously never use a classifier because it was too much effort.

There are absolutely good reasons to use smarter things for classification. Svm with tdf idf for example is super fast but will make very bad classifications at times in ways entirely unfixable with that system (thinking back to bats being an acronym for a theology degree classification or similar)

This is a bit too radical and misses some nuance. In order to start building classifiers you at least need some ML domain knowledge, like what to google at least. Meanwhile, point anyone at the jev/open-jev-clones API and have them start classifying in <10 minutes. There's some value there. (same used to be valid for LLMs before jev became a thing. You can use llms as zero-shot classifiers with the right API wrapper, w/ no domain data).

1000x more expensive run can be overall cheaper. You avoid upfront training/tuning cost. It all depends on your scale.