He worked on TensorFlow. So even if the doesn't do ML research himself, at least he works on the tooling.

Of course he is someone any technology organisation would want to have as a resource. But probably not as chief scientist or ceo of an ML company based on the available evidence

TensorFlow was honestly not that good. It had a lot of effort put into it, so it worked, but there are reasons people moved away from it.

I think Jeff Dean is a great engineer, but I wouldn't hold up TensorFlow as a great example.

I always thought he pair programmed so it's been Jeff + Sanjay.

Why wasn’t tensorflow good?

I started my ML journey in TensorFlow, and now I am a happy PyTorch user, for many years.

TensorFlow, yuck

But who uses TensorFlow -- really -- these days.

What else are people using then?

PyTorch everywhere

Or JAX

Google does almost exclusively; it's tied in to TFX.

It seems like that being an incredible software engineer is more important these days, look at Greg Brockman’s background.

That's my bias as well. To me, it seems like every day someone releases a new AI toy, but the thing you would actually want is for a real software engineer to take the LLM or whatever, put it inside a black box, and then write actually useful software around it. Like off the top of my head, LLM + Google Calendar = useful product for managing schedules and emailing people. You could make it in a day of tinkering as a langchain demo, but actually making a real product that is useful and doesn't suck will require good old fashioned software engineering.

Based on the multitask generalisation capabilities shown so far of LLMs I’m kinda in the opposite camp - if we can figure out more data efficient and reliable architectures base language models will likely be enough to do just about anything and take general instructions. Like you can just tell the language model to directly operate on Google calendar with suitable supplied permissions and it can do it no integration needed

Exactly this. There is a reasonable chance the GUI goes the way of the dodo and some large (75% or something) percentage of tasks are done just by typing (or speaking) in natural language and the response is words and very simple visual elements.

What you're describing is AGI levels of autonomy. There are quite a lot of missing pieces for that to happen I think.

Have you used GPT-4? People are already building agents to do things the above comment refers to.

People are building toy demos in a day that are not actual useable products. It’s cool, but it’s the difference between “I made a Twitter clone in a weekend” and real Twitter.

1 - companies are deploying real products internally for productivity, especially technical and customer support, and in data science to enable internal people to query their data warehouse in natural language. I know of 2 very large companies with the first in production and 1 with the second, and those are just ones I'm aware of. 2 - you are conflating the problems of engineering a system to do a thing for billions of users (an incredibly rare situation requiring herculean effort regardless of the underlying product) with the ability of a technology to do a thing. The above mentioned systems couldn't handle billions of users. So what? The vast majority of useful enterprise saas could not handle a billion users.

Right but Ilya is the Chief Scientist of OpenAI not Greg Brockman

I heard Ilya wasnt behind the big innnovations at OpenAI. It was lesser known scientists

OpenAI from a research point of view haven’t really had any “big innovations”. At least I struggle to think of any published research they have done that would qualify in that category. Probably they keep the good stuff for themselves

But Ilya definitely had some big papers before and he is widely acknowledged as a top researcher in the field.

I think the fact that there are no other systems publicly available that are comparable to GPT-4 (and I dont think Bard is as good), points to innovation they havent released

Possibly. Or they could’ve also just paid a lot of people in Africa or wherever else to create the highest quality RLHF dataset out there.

They mentioned in the Technical paper of GPT-4 that the capabilities of the model were not from RLHF. https://youtu.be/2zW33LfffPc?t=842

Furthermore, Jeff is not a great (or even good...) manager/director/leader. There were a lot of internal and external dramas because of his leadership, that he failed to address. How often you hear about dramas about other Chef Scientists at other, comparably sized, companies?

He should stay a Fellow, in a "brilliant consultant" role.

> How often you hear about dramas about other Chef Scientists

I know this is a typo, but I would love to hear about high drama involving Chef Scientists at large companies.

It would have all the nonstop action of Iron Chef, combined with the multilayered scheming of Succession... I think there's something there.

He absolutely should have gotten rid of the troublemaker. Many folks used this publicity to leave for higher positions or higher pay (which is very common at google) but made it look like Jeff Dean was the problem.

Being a brilliant ML researcher has approximately zero overlap with making products for people to use.

>Jeff Dean is clearly one of the greatest software developers/engineers ever but there isn’t much evidence that he is a brilliant ML researcher

Google have oversupply of brilliant ML researchers. What they need is a engineer that sees the applications of the technology so it can be turned into a product. Someone that can bridge the gap between the R&D team and the Bureaucracy.

Want an idea for a stupid product - input - description of a girl, hobbies, some minor flaws - output - create a poem. Have been using Vicuna quite successfully for that purpose.

He had the perfect balance of being legendary engineer and an ML researcher

Can't emphasize more on how much rigorous engineering practice could accelerate research delivery. It is THE key to have a productive research oriented team.

Good research engineers are underrated, and very difficult to find.

> Jeff Dean is clearly one of the greatest software developers/engineers ever

Based on what? I've heard all the Chuck Norris type jokes, but what has Jeff Dean actually accomplished that is so legendary as a software developer (or as a leader) ?

Per his Google bio/CV his main claims to fame seem to have been work on large scale infrastructure projects such as BigTable, MapReduce, Protobuf and TensorFlow, which seem more like solid engineering accomplishments rather than the stuff of legend.

https://research.google/people/jeff/

Seems like he's perhaps being rewarded with the title of "Chief Scientist" rather than necessarily suited to it, but I guess that depends on what Sundar is expecting out of him.

Jeff was very early on in the "just scale up the big brain" idea, perhaps as early as 2012 (Andrew Ng training networks on 1000s of CPUs). This vision is sort of summarized in https://blog.google/technology/ai/introducing-pathways-next-... and fleshed out more in https://arxiv.org/abs/2203.12533, but he had been internally promoting this idea since before 2016.

When I joined Brain in 2016, I had thought the idea of training billion/trillion-parameter sparsely gated mixtures of experts was a huge waste of resources, and that the idea was incredibly naive. But it turns out he was right, and it would take ~6 more years before that was abundantly obvious to the rest of the research community.

Here's his scholar page (H index of 94) https://scholar.google.com/citations?hl=en&user=NMS69lQAAAAJ...

As a leader, he also managed the development of TensorFlow and TPU. Consider the context / time frame - the year is 2014/2015 and a lot of academics still don't believe deep learning works. Jeff pivots a >100-person org to go all-in on deep learning, invest in an upgraded version of Theano (TF) and then give it away to the community for free, and develop Google's own training chip to compete with Nvidia. These are highly non-obvious ideas that show much more spine & vision than most tech leaders. Not to mention he designed & coded large parts of TF himself!

And before that, he was doing systems engineering on non-ML stuff. It's rare to pivot as a very senior-level engineer to a completely new field and then do what he did.

Jeff certainly has made mistakes as a leader (failing to translate Google Brain's numerous fundamental breakthroughs to more ambitious AI products, and consolidating the redundant big model efforts in google research) but I would consider his high level directional bets to be incredibly prescient.

thanks for this insightful perspective.

1. what was the reasoning behind thinking billion/trillion parameters would be naive and wasteful? perhaps part are right and could inform improvements today.

2. can you elaborate on the failure to translate research breakthroughs, of which there are many, into ambitious AI products? do you mean commercialize them, or pursue something like alphafold? this question is especially relevant. everyone is watching to see if recent changes can bring google to its rightful place at the forefront of applied AI.

OK - I can see the early ML push as obviously massively impactful, although by 2014/2015 we're already a couple of years after AlexNet, other frameworks such as Theano, Torch (already 10+ yrs old at that point), etc already existed, so the idea of another ML framework wasn't exactly revolutionary. I'm not sure how you'd characterize Jeff Dean's role in TensorFlow given that you're saying he lead a 100-person org, yet coded much of himself.... a hands-on technical lead perhaps?

I wonder if you know any of the history of exactly how TF's predecessor DistBelief came into being, given that this was during Andrew Ng's time at Google - who's idea was it?

The Pathways architecture is very interesting... what is the current status of this project? Is it still going to be a focus after the reorg, or too early to tell ?

Jeff was the first author on the DistBelief paper - he's always been big on model-parallelism + distributing neural network knowledge on many computers https://research.google/pubs/pub40565/ . I really have to emphasize that model-parallelism of a big network sounds obvious today, but it was totally non-obvious in 2011 when they were building it out.

DistBelief was tricky to program because it was written all in C++ and Protobufs IIRC. The development of TFv1 preceded my time at Google, so I can't comment on who contributed what.

Interesting - thanks!

> large scale infrastructure projects such as BigTable, MapReduce, Protobuf and TensorFlow

If you initiated and successfully landed large scale engineering projects and products that has transformed the entire industry more than 10 times, that's something qualified for being a "legend".

Only if you did at at a company like Google where it's being talked about and you've got that large a user base. Inside most of corporate America internal infrastructure / modernization efforts get little recognition.

I wrote an entire (Torch-like - pre PyTorch) C++-based NN framework myself, just as a hobbyist effort. Ran on CPU as well as GPU (CUDA). For sure it didn't compete with TensorFlow in terms of features, but was complete enough to build and train things like ResNet. A lot of work to be sure, but hardly legendary.

> Only if you did it at a company like Google where it's being talked about and you've got that large a user base

Google has lots of folks who had access to the similar level of resources and no one but Jeff and Sanjay made it. Large scale engineering is not just about writing some fancy infra code, but a very rigorous project to convince thousands of people to onboard which typically requires them to rewrite significant fraction of their production code, typically referred as "replacing wheels on a running train". You gotta need lots of evidence, credits and visions to make them move.

> "replacing wheels on a running train"

Yeah - just finished migrating a system of 100+ Linux processes all inter-communicating via CORBA to use RabbitMQ instead. Production system with 24x7 uptime and migration spread over more than a year with ongoing functional releases at the same time. I prefer to call it changing the wheels on a moving car.

No doubt it's worse at Google, but these type of infrastructure projects are going on everywhere, and nobody is getting medals.