DeepMind’s David Silver just raised $1.1B to build an AI that learns without...

[Coin_Wow]

DeepMind’s David Silver just raised $1.1B to build an AI that learns without human data

Ineffable Intelligence, a British AI lab founded a mere few months ago by former DeepMind researcher David Silver, has raised $1.1 billion in funding at a valuation of $5.1 billion to join the race for novel AI models that could outperform large language models.

According to its newly launched site, Ineffable aims to create a “superlearner” capable of discovering knowledge and skills without relying on human data by leveraging reinforcement learning — a technique in which AI systems learn through trial and error rather than studying human-generated examples. This is Silver’s area of expertise.

https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

Related question: How much longer can the AI "bubble phase" keep raising huge amounts of money without making profitable companies?

[Davidvictorson]

If he is able to pull this off which will be difficult then that will be the ultimate advantage over other AI companies and a good attraction for investors. Because they are looking to invest their money in the next big thing- being the first to get in.

[HONDACD125]

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

Related question: How much longer can the AI "bubble phase" keep raising huge amounts of money without making profitable companies?

We are not in some kind of fictional, sci-fi movie where robots are created by humans, they learn everything themselves and then become out of control, we need to understand that even if we create robots, they won't be able to do stuff that robots manage to do in movies, they fly around, have all kinds of weapons in-built into their systems, they can do whatever they want because they are now superpowerful. However, things in real life don't work that way, we don't have superheroes that can fly in real life, but we do have them in movies, so we should stop thinking about things that can only happen in fiction.

With that being said, it is obviously going to be great if a company can create a model that can learn itself and improve over time instead of relying on humans to train it. This means that it will be a better assistant for whatever you use it since it will have more knowledge, and since it will have an ability to learn and train over time, it will become very intelligent, and that will only be a plus point for anyone that uses it, and other than those who are working in digital markets, no one needs to be afraid of such a model.

[EluguHcman]

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

For quite sometimes now as technologies has evolved, we know AI to be one of the most profound vasts that is contributing to the economy growth and so on it has been profitable to the developers.

We have also come to see most of the AI prediction bots being incompetent to be reliable when such as predicting the sport bets and the trading market.
It is not also how much being spent on the projects that guarantees their performance. Some still spends multi $ and still don't have reliability of the technology.

In the case of David Silvers deep mindset on the quest of developing AI that can operate exceedingly without human intervention and beyond every other bots as denoted is still unbelievable because if it does, that means the bot can outperform providing solutions to every challenges of human kind such as predicting the future at all costs.

[Ucy]

Well, it's unlikely that AI wont be given the basic information(more like instinct) on how to do certain things. It just needs to do it more broadly and right, which is typically learned(or perfected by learning) from people/things, or gotten through experiencing on its own. . It's just like telling someone to learn through experience(trial and error) or by getting education from people.
I think experience without ideas flowing in or knowing how problems could be solved (basically being gifted or talented) can be much slower and inefficient compared to learning from a master for example

[aioc]

This is a very ambitious project; all we know is that all AI is human data-dependent. This one presents something beyond human, namely, the trial-and-error method. I don't know if this is possible to perfect, but based on the article OP presented.

Its superlearner may lack experience, but the company doesn’t lack ambition. “If successful, this will represent a scientific breakthrough of comparable magnitude to Darwin: where his law explained all Life, our law will explain and build all Intelligence,”

This is something to watch, but hey, $1.1 funding is huge to make anything possible; this could possibly be an AI game-changer.

[goldkingcoiner]

Without human-driven data I think the AI is just going to hallucinate and start making stuff up. It will just create it's own echo chamber and "reinforce" it's learning with that.

The whole thing smells of a marketing scheme to sell something that will not work. Especially because they keep using terms like "Super Intelligence", “Superlearner”  and "Ineffable Intelligence".

Sounds like a super failure that will end up with them pocketing the investor's money and super exiting. But maybe I will be proven wrong.
 

[Fortify]

DeepMind’s David Silver just raised $1.1B to build an AI that learns without human data

Ineffable Intelligence, a British AI lab founded a mere few months ago by former DeepMind researcher David Silver, has raised $1.1 billion in funding at a valuation of $5.1 billion to join the race for novel AI models that could outperform large language models.

According to its newly launched site, Ineffable aims to create a “superlearner” capable of discovering knowledge and skills without relying on human data by leveraging reinforcement learning — a technique in which AI systems learn through trial and error rather than studying human-generated examples. This is Silver’s area of expertise.

https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

Related question: How much longer can the AI "bubble phase" keep raising huge amounts of money without making profitable companies?

On your second point and the way that this AI idea is being described, it doesn't seem like you could contain it with rules because like skynet from the terminator films it could probably come to the conclusion that human rules are not a benefit to logical learning. We are stepping into the realms of artificial intelligence that we may not be able to control and it may figure out a way to travel through the internet, although it's impossible to say if it can ever become sentient. It seems like the next logical step in artificial intelligence though, have a machine that can learn everything from the smallest level of granularity by itself but that would take a huge amount of processing power and time.

[BlackHatCoiner]

For those who do not understand why this is a potentially big deal:

LLMs are statistical engines trained on massive amounts of human-generated text. We are rapidly approaching a point where we will simply run out of high-quality human data on the Internet. The ineffable solution David Silver proposes is pure reinforcement learning that does not need human data. It creates its own "synthetic experience", just as a human being learns from its environment rather than from text.

Whether this startup will succeed or not is unknown, but the next step for AI is definitely to make it learn beyond the limited human text that is available.

[Supreme Donvic]

How can the AI operate without using human data I don't think that is possible every information you see today were written by human so it is not possible for AI to produce information without it taking it from human, if eventually such a thing is created I will not believe any information coming from that AI because the AI will just be making things up and things that are not true every, information you see on Earth were produced and written by human, AI takes information from different source and they are from human data and so I don't believe that an AI can be created without using human data it is not possible.

[Smartprofit]

Regarding profitable companies...

Here, I think we also need to consider the geopolitical standoff between the US and China. If I were the Chinese leadership, I would distribute as many new free Chinese language models as possible to the European and American markets. This would likely lead to the financial collapse of Open AI, Anthropic, and other American companies. Their stock prices would plummet because they would have no way to justify the huge financial investments they made in AI development. This would also lead to a recession in the American economy. China could defeat the US without launching a single missile.🙋

I consider this scenario entirely possible. The only thing that could prevent it is China's lack of a large number of its own GPUs (and a huge number of them are needed for such a task).

I don't believe that AI will make rapid breakthroughs in science and technology (for example, invent a cure for cancer).  Even if he manages to solve this problem, it will be a very long time before this drug is produced and commercially available...

Therefore, Open AI, Anthropic, Google, Microsoft, and other companies are currently betting on distributing language models through paid subscriptions. If free alternatives emerge, this will result in huge financial losses for all these companies. This will lead to a collapse of the entire American stock market. 🙅

[Cryptomultiplier]

DeepMind’s David Silver just raised $1.1B to build an AI that learns without human data

Ineffable Intelligence, a British AI lab founded a mere few months ago by former DeepMind researcher David Silver, has raised $1.1 billion in funding at a valuation of $5.1 billion to join the race for novel AI models that could outperform large language models.

According to its newly launched site, Ineffable aims to create a “superlearner” capable of discovering knowledge and skills without relying on human data by leveraging reinforcement learning — a technique in which AI systems learn through trial and error rather than studying human-generated examples. This is Silver’s area of expertise.

https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

Related question: How much longer can the AI "bubble phase" keep raising huge amounts of money without making profitable companies?

On your second point and the way that this AI idea is being described, it doesn't seem like you could contain it with rules because like skynet from the terminator films it could probably come to the conclusion that human rules are not a benefit to logical learning. We are stepping into the realms of artificial intelligence that we may not be able to control and it may figure out a way to travel through the internet, although it's impossible to say if it can ever become sentient. It seems like the next logical step in artificial intelligence though, have a machine that can learn everything from the smallest level of granularity by itself but that would take a huge amount of processing power and time.

There are Ai tokens already in play and Ai models are much more being developed to be efficient right now, instead of being profitable.
The ROI for such future Ai development is going to be massive, but it's based on how its current structure is built and I must concur how scary it would be to watch this unfold after seeing Skynet as a movie with a studio picture presentation of the possibilities of how AI bots and robots could turn out to be, when it starts generating its own experience.

[Synchronice]

DeepMind’s David Silver just raised $1.1B to build an AI that learns without human data

Ineffable Intelligence, a British AI lab founded a mere few months ago by former DeepMind researcher David Silver, has raised $1.1 billion in funding at a valuation of $5.1 billion to join the race for novel AI models that could outperform large language models.

According to its newly launched site, Ineffable aims to create a “superlearner” capable of discovering knowledge and skills without relying on human data by leveraging reinforcement learning — a technique in which AI systems learn through trial and error rather than studying human-generated examples. This is Silver’s area of expertise.

https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

Related question: How much longer can the AI "bubble phase" keep raising huge amounts of money without making profitable companies?

I don't understand how is that possible. AI that learns on its own, has to filter the information, which one is true and which one is fake. To filter correctly, AI should be able to have moral. How can AI have moral and judgement, similar to human? How is it possible to program it in a way to have moral and judgement? Seriously, we all have different mentality and views about things because we grew up in different families, cultures and societies. How can AI replicate this?

When it comes to AI bubble, I think that it will only be endless if it actually delivers real intelligence. We, humans, are intelligent animals and right now we try to create an alternative intelligence, superior to us, so the bubble has no end to my mind.

[serjent05]

Well, it's unlikely that AI wont be given the basic information(more like instinct) on how to do certain things. It just needs to do it more broadly and right, which is typically learned(or perfected by learning) from people/things, or gotten through experiencing on its own. . It's just like telling someone to learn through experience(trial and error) or by getting education from people.
I think experience without ideas flowing in or knowing how problems could be solved (basically being gifted or talented) can be much slower and inefficient compared to learning from a master for example

I think this project is very interesting.  Ai will learn from their experience and won't be fed with human data. This means the project have to do countless simulations in order for the AI to experience and learn.  This indeed will path on trial and error, but I still wonder how the project determines the success without comparing the result from human data.

[robelneo]

First, economically it seems like a great idea to have some real AI competition instead of just 4 or 5 companies spending endless $Billions on similar types of LLM models. Second, having new systems which can actually learn on their own, without standard human training, is both exciting and a bit scary. Is building in all the needed safe-guards going to be done in time before the robots turn evil and take over?

All the AI's I've stumbled upon have human data on them. If it's coming from trial and error, how will they guarantee superior, consistent results that people will rely on for accuracy?
Two things are going to happen: it will be either the biggest fluke in AI or a revolutionary concept that will change how AI works, so we'll see the outcome of this project.

[btc78]

trial and error? would that not take longer rather than just taking what humans have already done and referencing that? i guess the end result they want is for robots to have original thoughts or not be dependent on humans anymore. see you in a couple of years when robots are already self conscious.

[TastyChillySauce00]

Isn't distilled AI model is exactly this, training on synthetic data produced by existing AI but the real problem is quality degradation and model collapse.
If the model didn't learn from human generated example, how it could figures out which is right and which isn't though? I have a hard time to believe yet that $1.1B making this seems very legit.

[Darker45]

This is interesting. Didn't read the entire article but I presume it's a bit exaggerated. Could AI really work, however super intelligent it may be, without human data? Could AI trial and error work absolutely without human input in terms of information? Where will they even start?

Anyway, the AI race is getting more and more exciting. Billions are pouring in. The game is getting more serious. I'm glad Ineffable is joining in the game bringing with it a distinct and ambitious goal.

AI will take over much of what humans are doing right now, but humans aren't camera films or DVDs or whatever. They can and have to adjust. As to opportunities, as always, one goes out, another comes in.

[Doll2233]

Ineffable intelligence is a fascinating shift in the field of AI. It is bold move towards super learner based on the reinforcement models rather than working with large language models. This will  create a break through in the field of science Engineering and technology. This diversification will drive unique inventions but there will be certain safety concerned with this autonomous learning. It will be difficult to build guard rails for a system that discovered its own logic through hit and trial rather than working on given human data set. Investers putting billion dollars into this must prove profitable in order to survive,real profits to sustain this massive momentum. It will take another year or two before investors starting having profit. If this will not happen this there might be a market correction.

[coin-investor]

If he is able to pull this off which will be difficult then that will be the ultimate advantage over other AI companies and a good attraction for investors. Because they are looking to invest their money in the next big thing- being the first to get in.

It will be a game-changer in the AI industry if it's more accurate than the AIs we use today. For it to be competitive, it should be accurate, up to date, and, far better, the $1.1 billion budget is a huge amount, but the task to create this kind of AI is immense.
Since they announced their intention to build this kind of AI, many industries are watching its development, not only the AI industry.