Tl;DR: It's not a new organization, or non-profit institute. It's a Substack/blog.
DMI is a platform for researchers and thinkers from across Google DeepMind, Google, and the wider global research community to work on and publish creative, deeply informed ideas about a world with AGI. They will not always agree, and they will likely change their minds, as more data and information comes to light at the fast-moving frontier. We created this institute precisely because broad-based intellectual discussion and debate are required to arrive at a consensus about how to address the challenges and opportunities we face as a society.
(Hiding chains of thought is one of the most harmful things I think OAI/Anthr. has done, altough Google Gemini also encrypts its chains of thought, so it's interesting to see the DeepMind Institute crticising this.)
My read is that they're criticizing moving the reasoning (more) into the latent space, such that not even researchers/safety-evaluators can actually see what's going on.
Of course this still then has a 'trust me, bro' feel to it, but making it outright impossible to reasonably inspect the CoT is a significantly larger and riskier step, IMO.
We created this institute precisely because broad-based intellectual discussion and debate are required to arrive at a consensus about how to address the challenges and opportunities we face as a society.
Something tells me that consensus will never arrive to the conclusion that we should shut down that whole industry
They are simply looking to steer AI policy discussions. It's basically an in-house think tank. It has all the trappings, especially with ominous prognostications like:
Today’s AI systems have impressive capabilities and the rapid pace of innovation suggests we’re now approaching artificial general intelligence (AGI), a system that exhibits all the cognitive capabilities of the human brain.
Not sure we are anywhere close to all the cognitive capabilities of the human brain... not even a severely lobotomized one.
Just one, the ability to use language, it's causing all sorts of problems for us because we evolved to recognize other humans as the only ones having this capability.
Ai separated humans from language, like the writing did for memory and the printing press for knowledge distribution.
"AI does to us what American Cheese did to food" by Opus 23
So, like ARC-AGI-3? I'm sympathetic to the notion that the things we're able to measure are necessarily going to miss important aspects of capability and intelligence. But people are attempting to measure this kind of thing, and models keep getting better at it.
They are simply looking to steer AI policy discussions. It's basically an in-house think tank.
For a guy who spent his life imagining and realising dreams in video games and cognitive abilities in artificial neural networks, it has to be very soul destroying for Demis to spend his time doing something as frivolous as this.
So it seems, but what of it? That wasn't his first career setback, and I'm sure he's still trying to get interesting work done at Isomorphic and maybe elsewhere in addition to working the PR pump for Alphabet.
Here's my opinion on the "pacing the frontier" issue. Recursive self improvement can accelerate AI advances a lot. Whoever nails it first, has a huge first mover's advantage. But OpenAI is taking a "move fast and break things" attitude. This is making life very difficult for Anthropic: how can you retain your edge when you keep measuring twice to cut once, but your competitor keeps blowing things up left and right?
Google, who was left a little bit behind anyway, is more than happy to join the calls for taking a more cautious approach to AI development. Hence DeepMind Institute.
Sam Altman has no option than to pretend he agrees with all these ideas. What else is he supposed to say?
- need for faster, better and more accurate measurements of the things we care about
- 3 scenarios of impact ranging from mild to major disruption
- mild policies are all sensible, like expanded unemployment insurance and Earned Income Tax Credit
- major disruption policies are also pretty logical, emphasizing owning a share of the profits of AI
- AI evaluators to sort and weigh the policies for their effectiveness
It looks like to me a lot of thinking, research, and work put into the piece.
I know people will gripe at using AI evaluators to score and weigh policies for effectiveness but if you read any amount of research papers, they do researcher evaluated scoring, which translates to "me, a 26 year old PHD candidate who hasn't held a job yet, and my roommate who stayed up all night with me to tell me his vibes on the subject".
It is IMO wrong and lazy to dismiss these pieces as being the same level as Dario or Sam's blog posts / speeches / call to actions, these are actual research papers.
Crazy that these labs keep throwing around entirely unsubstantiated claims like "AGI is right around the corner". Given the enormity of the problem I'd argue we're not really any closer than a decade ago, especially given LLM shortcomings.
Tl;DR: It's not a new organization, or non-profit institute. It's a Substack/blog.
The word "institute" gives an aura of credibility without some boring accreditation. Though not sure every country allows that.
Nonetheless, there are great articles up on there, such as:
https://institute.deepmind.com/essays/the-case-for-reasoning...
(Hiding chains of thought is one of the most harmful things I think OAI/Anthr. has done, altough Google Gemini also encrypts its chains of thought, so it's interesting to see the DeepMind Institute crticising this.)
My read is that they're criticizing moving the reasoning (more) into the latent space, such that not even researchers/safety-evaluators can actually see what's going on.
Of course this still then has a 'trust me, bro' feel to it, but making it outright impossible to reasonably inspect the CoT is a significantly larger and riskier step, IMO.
So, a think tank, to lobby governments
Something tells me that consensus will never arrive to the conclusion that we should shut down that whole industry
How can they even pretend to distance themselves from Google…
They are simply looking to steer AI policy discussions. It's basically an in-house think tank. It has all the trappings, especially with ominous prognostications like:
Not sure we are anywhere close to all the cognitive capabilities of the human brain... not even a severely lobotomized one.
Exactly how many cognitive capabilities are needed before it becomes a problem?
Just one, the ability to use language, it's causing all sorts of problems for us because we evolved to recognize other humans as the only ones having this capability.
Ai separated humans from language, like the writing did for memory and the printing press for knowledge distribution.
"AI does to us what American Cheese did to food" by Opus 23
https://www.youtube.com/watch?v=1CbmC2aWhjY
(still the best Ai video I have seen, also a work of art)
This blob of text seems to mean nothing
What happened to the knowledge in the elder's mind when they died, prior to writing? What changed about that once we had writing?
I'm not actually sure how you could come to that conclusion.
Maybe you need to learn more about the language uplift.
In which capabilities are AI not already close to or beyond that of a human brain?
politics
The ability to solve general problems previously insoluble using only experiential data and modification to internal world models.
So, like ARC-AGI-3? I'm sympathetic to the notion that the things we're able to measure are necessarily going to miss important aspects of capability and intelligence. But people are attempting to measure this kind of thing, and models keep getting better at it.
They've read a few thousand years of experiential data in text form.
I don't feel like the models are becoming any better at writing.
Rap lyrics are a good test. They are on the level of a censored version of Insane Clown Posse. Just terrible.
For a guy who spent his life imagining and realising dreams in video games and cognitive abilities in artificial neural networks, it has to be very soul destroying for Demis to spend his time doing something as frivolous as this.
I don't know, Hassabis always seems to have been willing and able to do PR and put himself out there, in addition to having the technical muscle.
Didn't he get kicked out of the company he founded? Deep Mind does not have a CEO anymore
So it seems, but what of it? That wasn't his first career setback, and I'm sure he's still trying to get interesting work done at Isomorphic and maybe elsewhere in addition to working the PR pump for Alphabet.
maybe if AI was progressing linearly we wouldn't be close, but if it's progressing exponentially then all bets are off
wow i thought the attributions were for the vector graphic things and was like "ok demis is in his illustrator bag too!"
Just a new shop front. Nothing to see here.
I wonder if they could make the font even greyer on that website. MORE GREY because I am still able to read it (bearly).
Google citing Mark Fisher was not on my Postmodern-hell-world bingo card.
Here's my opinion on the "pacing the frontier" issue. Recursive self improvement can accelerate AI advances a lot. Whoever nails it first, has a huge first mover's advantage. But OpenAI is taking a "move fast and break things" attitude. This is making life very difficult for Anthropic: how can you retain your edge when you keep measuring twice to cut once, but your competitor keeps blowing things up left and right?
Google, who was left a little bit behind anyway, is more than happy to join the calls for taking a more cautious approach to AI development. Hence DeepMind Institute.
Sam Altman has no option than to pretend he agrees with all these ideas. What else is he supposed to say?
Their economic policy article is quite good:
- need for faster, better and more accurate measurements of the things we care about
- 3 scenarios of impact ranging from mild to major disruption
- mild policies are all sensible, like expanded unemployment insurance and Earned Income Tax Credit
- major disruption policies are also pretty logical, emphasizing owning a share of the profits of AI
- AI evaluators to sort and weigh the policies for their effectiveness
It looks like to me a lot of thinking, research, and work put into the piece.
I know people will gripe at using AI evaluators to score and weigh policies for effectiveness but if you read any amount of research papers, they do researcher evaluated scoring, which translates to "me, a 26 year old PHD candidate who hasn't held a job yet, and my roommate who stayed up all night with me to tell me his vibes on the subject".
It is IMO wrong and lazy to dismiss these pieces as being the same level as Dario or Sam's blog posts / speeches / call to actions, these are actual research papers.
Does anyone have any idea why most of today’s top links are submitted by the same account, one that is 11 days old?
The machines are starting to astrotruf the narrative that their takeover is ”not so bad, really”.
Crazy that these labs keep throwing around entirely unsubstantiated claims like "AGI is right around the corner". Given the enormity of the problem I'd argue we're not really any closer than a decade ago, especially given LLM shortcomings.
so glad they spun out a non-profit!