[HN Gopher] NLP Course - For You
___________________________________________________________________
NLP Course - For You
Author : mjakl
Score : 32 points
Date : 2023-12-23 19:15 UTC (3 hours ago)
(HTM) web link (lena-voita.github.io)
(TXT) w3m dump (lena-voita.github.io)
| victorbjorklund wrote:
| Definatly not responsive design but syllabus looks promising
| hackernewds wrote:
| Definatly not the best critique
| light_hue_1 wrote:
| As an ML researcher I'm sad to say that this painfully out of
| date. It's definitely a course from 5 years ago. Totally
| irrelevant as an intro to NLP today.
| linooma_ wrote:
| Has much changed as far as parts of speech tagging in the last
| 5-10 years?
| screye wrote:
| The sad truth is - all of classical nlp is Dead, with a
| capital D.
|
| The bottleneck for accuracy was always data quality and human
| effort, not model architecture.
|
| Llms make the data and human problems so much easier, that
| the benefits of supporting different architectures just
| doesn't make sense. With quantization, I'm not even sure
| classical models win out on cost anymore, and they had
| already lost on (real world) accuracy.
|
| LLMs are the O365 subscription that you just can't fight
| against with bespoke mini solution. An all in one solution is
| simply too appealing.
|
| Also, if you have to learn pre-2020 NLP I would just learn to
| use spacy. It pretty much covers all of pre-2020 NLP out of
| the box in a well documented package with strong GPU and CPU
| support.
| behnamoh wrote:
| So if anyone wants to learn language modeling stuff, do you
| recommend starting with transformer and just learn how to
| deploy and finetune LLMs (given that ordinary people can't
| train these LLMs)?
| nvtop wrote:
| A lot. POS taggers used to be linear classifiers + features.
| In 2018 they switched to BERT and similar encoder-only
| models. In 2023, POS tagging is largely irrelevant, because
| it was used as a part of a larger pipeline, but now you can
| have everything end-to-end with better accuracy by fine-
| tuning a sufficienly large pretrained model (LLM or encoder-
| decoder like T5)
| fantispug wrote:
| It covers a lot of the fundamentals in some detail (attention
| and transformers, decoding, transfer learning) that are
| underneath current cutting edge NLP; this is still a very good
| foundation likely to be good for several more years.
|
| What might be missing is in-context learning, prompt
| engineering, novel forms of attention, RLHF, and LoRA (though
| it covers adaptors), but this is still changing rapidly and the
| details may be irrelevant in another year. If you have a look
| at a recent course like Stanford CS224N 2023 there's a lot of
| overlap.
___________________________________________________________________
(page generated 2023-12-23 23:02 UTC)