[ninth_ant]

For the benefit of innovation it actually pays off when some organizations work on alternatives to the dominant model.

What if LLMs become stagnant and some other approach is needed, perhaps in tandem with it? Work that seems irrelevant today may become useful then.

I’m not saying that their approach is this, just that failure to produce a LLM doesn’t necessarily equal an embarrassing failure.

[martingoodson]

The failure to do any meaningful work related to the most important breakthrough in AI ever is objectively bad.

[version_five]

That doesn't follow from anything. Research is part methodical slog, part lottery, maybe a pinch of intelligence. A few labs won the short term lottery here, and most researchers explored stuff that didn't get headlines. (And to be fair, OpenAI built a great product that catapulted lab research into popular view).

There might be some argument on other metrics - publications, students trained, lectures, recognition, whatever, that show this institute is lagging. But not being part or llms implies nothing about their success or failure.

[mnd999]

LLMs are not the most important breakthrough in AI ever, in the same way the NFTs are not the most important breakthrough in digital commerce ever. It's just a load of hype to generate big funding rounds. At least there's no cartoon apes this time around.

[sva_]

I agree that LLMs are not the most important AI breakthrough ever, but your characterization seems needlessly harsh, as LLMs undeniably have utility.

[extasia]

The transformer architecture is arguably the most important breakthrough in NLP, and language is the predominant mode of communication between humans, so I fail to see how its "just a load of hype"

Could you name a bigger breakthrough in AI?

[sva_]

I'd say the Multi-Layer perception itself.

Maybe even convolutional neural networks, because they showed that ANNs are viable and are what really got the ball rolling.

[mafribe]

CNNs are from the 1980s (the "neocognitron" by Kunihiko Fukushima [1]), while the MLP is from 1958 [2]. So nearly half a century, resp. a century old.

[1] K. Fukushima, Neocognitron: A Self-organizing Neural Network Model for a Mechanism of Pattern Recognition Unaffected by Shift in Position.

[2] F. Rosenblatt, The Perceptron: A Probabilistic Model For Information Storage And Organization in the Brain.

[nonethewiser]

And what alternative to LLMs does the Turing institute have?

> I’m not saying that their approach is this, just that failure to produce a LLM doesn’t necessarily equal an embarrassing failure.

That's not really a rebuttal against the article though. They specifically state that they aren't blaming the institute for not producing and LLM nor even for not predicting them.