It seems like the real news is Jeff and Sanjay are leaving Google, and Demis is effectively replacing Jeff as Chief Scientist for all of Alphabet.
The bigger deal is the departure of Jeff and Sanjay, rather than Demis moving into a different role.
The bigger deal is the departure of Jeff and Sanjay, rather than Demis moving into a different role.
Hassabis seems to have been pushed aside. He had been CEO of DeepMind, but that position no longer exists and it seems Kavukcuoglu is now leading DeepMind with a title of SVP. Hassabis is now just "Chair" of DeepMind, and has been given the newly created title of Alphabet Chief Scientist. Are these just face-saving titles, or does he still have any real influence over Google/DeepMind's pursuit of AGI?
Shane Legg remains as DeepMind "Chief AGI Scientist", but I wonder if the DeepMind founding mission of creating AGI is really intact, or if he will be next to go. Has DeepMind just become the Gemini division?
I just found out that David Silver, DeepMind's RL-expert, already left in february, to create a startup "Ineffible Intelligence" focusing on RL-based continual learning.
Yeah he seemed too reasonable to me, relative to the fervor. Whoever ends up in charge needs to do a lot of frothing to catch up to the ferver that would justify their valuations and investments
Can you explain this more precisely?
https://endpoints.news/demis-hassabis-leaning-into-isomorphi...
Real scientists are skeptical. Wall St and the people who serve it don’t like that.
If the goal is to maximize share price (which it is) this is probably the safest approach. Why do they have to keep chasing developing the best model when they can
a) charge everyone for the cloud infra
b) have a "good enough" experience for normal consumers
Their current setup will be worth trillions. Already is. Let anthropic get paid for the most expensive queries while they get a bill from Google for their cloud/TPU use. GCP grew 82% YOY with 24B revenue and improving margins. So GCP became a 100B business. Anyone doubts it's gonna double in less than 5 years? Gemini just needs to be good enough for normal folks who want personal agents for everyday use. I don't think Google has to compete with Anthropic on making the best agentic programming model.
I don't get the impression that the difference between one company succeeding to build SOTA LLMs, and another struggling, comes down to individual employees - it seems to be more about the organization itself and their ability to manage teams and projects of this type. No doubt there are a few rockstars generating huge value, such as Noam Shazeer had been, but they are the exceptions.
When DeepMind was first created, before Google acquired it, they were famous for the high salaries, especially for the UK, but this was an assemblage of the brightest and best PhDs, expected to be solving challenging research problems along the unknown path to AGI. Many of these original employees may still be there, but it seems their job and value proposition has changed - are they any more capable, or key to, helping Gemini catch up with the competition than some "rank and file" employee familiar with LLMs? And if so, why haven't they done it?
I'm honestly not sure about that. They need a decent LLM to fight off the threat of LLMs replacing search, but I'd say that Gemini 3.6 Flash is more than good enough for that, with a smaller/cheaper model being preferred to a larger one. If you are "searching" for a proof to the Jacobian conjecture, then try Fable, and I doubt Google will miss the advertising revenue if Anthropic manage to sell Terrance Tao a pair of socks.
More to the point, DeepMind seems to have become a product division charged with building LLMs, not a blue sky research institute chasing AGI. To the extent that continued LLM improvement is important to Google, the relevant question is how much do you need to pay for a competent ML/LLM developer?
>you sound like an MBA
Well, no - techie here.
I wasn't sure if you were suggesting that DeepMind should pay more just because people developing LLMs at other companies are paid more, or because these are elite DeepMind researchers and are objectively worth more than other Google developers. My point being that it seems they are no longer being used as elite researchers - they are LLM developers. Does Meta need to pay FAANG salaries to employees that have been repurposed as data labellers?
i know people at deep mind, its impacting their ability to deliver good products
Maybe this isn't the same as the eight figure comp they'd get at Meta when they did their hiring spree, but no one thinks that's sustainable.
IDK, haven't Google been putting Gemini front and centre in pretty much all of their products?
I'm seeing Gemini on my slide decks, Gemini on my e-mails, Gemini on my searches, Gemini on my videoconferences, Gemini on my database query console. My impression was they were doing a Google Plus style attempt to marshal all the company's efforts behind one product.
I don't get the impression from interviews that Kavukcuoglu is that guy - he seems like a safe pair of hands, but not someone that is on a mission.
OTOH I don't even think this is the right race to be in.
My main reasoning was that transformers was the lightning in a bottle and the best work is in extending it instead of transcending it, which requires you to capture another lightning . Which to me appears to miss the assignment. OpenAI, Antrophic, they understand this intimately. Google on the other hand, fell victim to their own ambition.
If your goal is purely commercial, or time critical, then a product-based approach of squeezing all the juice out of LLMs makes sense.
If your goal is truly human-level AGI then this is more of an open-ended research endeavor, and timelines are hard to predict. Arguably we have only "captured lightning in a bottle" once in the last decade - the original 2017 attention paper - and so the timeline for a "few more Transformer-level breakthroughs" might more realistically be estimated in decades rather than years. You could argue that the application of RL to LLMs as a training method was a second "lightning in a bottle" but I don't think it changes the expected timeline of such discoveries by much.
The time criticality seems to have become a huge factor for those pursuing LLMs, and certainly for OpenAI and Anthropic, who regard it as a race.
It seems absurdly obvious (though many would disagree!) that LLMs alone are not going to achieve human-level intelligence and cognitive performance (using a slightly broader term there to include things like creativity, for those that might not consider that as part of intelligence).
If you compare a Transformer to a brain, then the best parallel is that a Transformer is functionally similar - in being a prediction engine - to part of our cortex, but of course that means ignoring the other half our cortex - the feedback paths that enable continual learning, which in turn supports creativity.
Of course people will probably respond "you don't need flapping wings to fly", but if you want to fly you do need SOME way of doing it, so brain comparisons are still valuable... If you look at our brain architecture and identify all the components and connections that have no equivalent in a Transformer, and if the goal is human-level capability, then you do need to understand what each of those brain components achieve functionally, and have SOME way of providing that functionality in your LLM+ or whatever you call it. LLMs' lack of any functional equivalent to our cortex's feedback paths - lack of continual learning - has been recognized as one major functional deficit, but there are probably half a dozen others too, reflecting the multiple "Transformer level" breakthoughs that Hassabis notes are needed.
While you don't need flapping wings to fly, you do need to invent the airplane, and even after 100+ years of airplane advances we've yet to build an airplane even remotely as capable as what some birds and insects are able to achieve.
Maybe 2017 was the Wright Bros moment where humans first learnt to do some of what our brains can do.
CEO of <thing> is a layer above SVP.
Google has many layers of management.
Alphabet Chief Scientist doesnt sound like a demotion / lack of influence to me but who knows. We're all just speculating here.
It was to fight ChatGPT and promote Gemini.
I think the guy does a very poor job or is simply not the right guy to appear as public figure for Gemini.
At least he tried. He is a man for everything that is not filmed.
Google doesn’t really have a person to give Gemini or AI a human face. And that is only consequential because Google never had any public person with any charisma like Jobs, Zuck, Altman.
(Jobs I'll give you.)
EDIT: "figurehead" - that's all I meant. A notable, public figure from the company who is credited with having a significant impact on its evolution. I'm not making a judgement call on his departure, or Ive's, being good or bad.
That said... just give it a couple years, they'll be back in a lucrative aquihire.
Who knows? Pure speculation? You can also say if Jobs was still around they could have 10x-ed it even further?
Apple car could have been a thing? Apple could have been way ahead and actually competing in AI and data centers? Who knows what else Jobs could have came up with?
Anyone remember the touchpad MBP with no physical escape key and the butterfly keyboard?
(Respect to many of Ive’s great legacy though)
Would like one with all the current physical keys plus a Touch Bar that you could do cool stuff with.
Such a gifted man… doing such an incredibly dumb thing. https://www.macworld.com/article/696590/apple-expose-jony-iv...
Edit:
- why’d Tim let him?
- why’d the college let them, OK money, but couldn’t they have potted and replanted for just a few or a couple-dozen million more?
- why not have a greater vision and build the extravagant tent to enclose the trees (wouldn’t be the only example of beautiful living indoor trees)?
- why not choose a site that would accommodate without any tree removal?
wtf?
The position that I have rather often read on the internet is: Jonathan Ive did very good work at Apple as long as there was a counterpart who could steer his creative vision. This counterpart was of course Steve Jobs. When Steve Jobs died, there wasn't such a counterpart anymore, so Ive's work for Apple got much worse.
Are you sure we should compare it like this? Not sure it implies what you think it does...
So the parent comment would imply that losing Dean is a good thing for Google, which is way less likely here.
IS that a promotion or demotion ?
When DeepMind allowed Google to buy them, it obviously had some major immediate positives - access to compute and money - but it seems it should have been obvious that the agreement was too good to be true, that they would be allowed to continue independently on their blue sky research mission to create AGI without any external interference or pressure to create product.
It seems that Hassabis and his DM co-founders eventually realized the mistake and tried to take DM private again starting c.2018, but of course this failed.
https://colossus.com/article/project-mario-demis-hassabis-de...
Now Hassabis has lost control of DeepMind altogether, and it seems to me, as a total outsider, that this is the end of the DeepMind mission to create AGI, at least the Hassabis/Legg definition of AGI as human-level general intelligence, capable of creativity and scientific discovery. Hassabis had always, until very recently, said that he believed it would take a number of additional "Transformer-level" breakthroughs to achieve this type of human-level AGI, while still seeing an LLM as one component of if (which to me seems an admission that the goal has failed - a true human level AGI should be able to learn language, etc, using it's own continual learning mechanisms).
It seems that DeepMind has now fully become the Google Gemini (LLM) division, trying to create a me-too product.
In the early days of DeepMind, before Google, before LLMs, I remember a David Silver slide deck titled "Reward is all you need", referring to RL rewards, which I never agreed with (although Rich Sutton might), but does at least reflect the independent thinking at DM, and of course RL not only gave rise to AlphaGo, but has now become central to the continued improvement of LLMs. However, notably David Silver also left DeepMind earlier this year, to found his own startup focusing on RL-based continual learning, presumably feeling that there was no longer a place for that type of research/pursuit at DeepMind.
Still, LLMs seem to be a destructive enough force on their own that perhaps it should be seen as a positive if research towards more powerful AGI appears to have had a major setback.
As for the "Alphabet Chief Scientist" title, it seems somewhat irrelevant, as least as far as Google's pursuit of true AGI. Hassabis is the face of beneficial AI, having been Knighted and awarded a Nobel Prize for his work, and it would be a horrendous PR move for Alphabet not to at least appear to be treating him with respect, even if in fact this does reflect him being pushed aside.