Alphabet’s Google Combines AI Research Units Brain and DeepMind | Hacker News Reader

Alphabet’s Google Combines AI Research Units Brain and DeepMind

wsj.com

70 pointsmfiguiere21 comments

It looks more like a re-org to show to investors Google is doing something to fix this. "let's combine departments and move people around". Google is a technical company and they have not had much use of LLM as a technical feature in any of their products. Honestly, what is the use of LLM other than write your essays and summarize web info? For technical nerds, LLM is boring.

Google is approaching this very consumer-oriented (Apple-esque) LLM model with a very technical genius mindset. And this is incredibly miscalculated. The kinds of advances from DeepMind research and Brain research is a puzzle piece of advancement that needs a good product development to seam together. I doubt DeepMind founder has this talent.

It needs to first accept LLM is NLP on steroids. And it is way behind on the magical consumer-oriented technology. Then develop further research for a reasonable product in a year. In the meantime, start giving more technical information without the nice language skills as search results (similar to what it is doing today, but on a broader scale). Let users enjoy this feature, while they use ChatGPT LLM for their office documents.

Both Brain and DeepMind had more researchers than OpenAI and more money too, yet Google couldn't use all that to produce anything. I don't think the merger will be very useful here, since I doubt google has any kind of resource constraint in place. Deepmind was not even making any products, or even any plans. They did lots of flashy stuff like microing blink stalkers in Starcraft but it was always hard to see where would all that lead to.

OpenAI’s flagship products are literally named after work done by Google researchers.

Yes, that even adds more weight to GP's argument which seems to be "they had everything and someone else ate their lunch".

Reminds me of what happened with Xerox PARC

Or Kodak

> someone else ate their lunch

How? Is OpenAI expected to be more profitable in the search landscape in the next year?

That’s even more embarrassing for Google. They couldn’t make a product out of their initial innovations.

I think Search was such a strong engineering feat that sold itself as a product, that they never had to develop the product chops that other tech companies had to.

I'm also thinking of Waymo, and the disciplined way they went about developing their technology, working with municipalities, first responders, etc. But, there is already a clear-cut product goal (drive better and safer than a human driver), existing legal and regulatory framework, and very visible and high stakes for errors (humans getting injured or killed). It's the kind of conditions that engineering excellence and mediocre product design does very well.

Uhm, Alphafold?

DM and "Open"AI took two different approaches to AGI, with the former taking a bottom-up first-principles kind of approach, whereas the latter embraced the bitter lesson[1]. I don't know which approach GB took, but it looks likely that it was more of a support role to the core business instead of upsetting it.

[1] http://www.incompleteideas.net/IncIdeas/BitterLesson.html

Seem pretty common with Google. They have brilliant and genius researchers and engineers, but they have terrible product managers.

Not going to help so much - probably the biggest reason Google didn't launch its own ChatGPT first was because it was protecting its search and ads business and not wanting to upset the order that underpinned its revenues.

Ahh the big company reorg. Executives need to somehow look busy, gaining alignment etc etc...

Also discussed here: https://news.ycombinator.com/item?id=35643049

One thing a CEO can always do: reorganize.

Combine two departments that were separate: "This will improve our focus."

Two years later, separate them again: "This will improve our focus."

Brain and DM have proven that they are good enough to keep alive for the next 20 years - like Bell labs, PARC, Kodak or similar

But make no mistake, just like Bell labs Google will shut down or divest this over the long term.

It took Google 4.5 months just to name their response to ChatGPT/Bing.

https://archive.ph/Rv8Yc

Easiest way Sundar can deflect failure.

Who got fired at Brain/DeepMind from the CXO offices for their overall incompetence?

Reply on news.ycombinator.com