18, ceo @mortarvault, building autonomous manufacturing robots this is extremely non-linear dynamics
J1407b
Joined December 2024
- Tweets1.6K
- Following90
- Followers1.8K
- Likes13K
One of the most amazing experiences I had this year was just flying to Uzbekistan
i met some really interesting people. I had flown with @qodirovskiye (super smart imo-maxxer)
just realized how amazing the pool of talent is. Qodirov flies to Central Asia every 3 months or so, helping some of the most talented people find opportunities to fully exploit their latent potential
i think the Soviet math education system created genius factories. You can see how much intellectual talent exists in places where opportunities and resources are much more limited. So much human potential is simply underutilized because of geography
True countries of geniuses that eat horse meat
1
13
612
a realistic and practical way to make manufacturing accessible anywhere to anyone
Making Manufacturing Work
(This post follows the first introductory post — Manufacturing the World)
Manufacturing is applied information. One material can turn into waste, a working product, or a constructor that makes more products. It really depends on what we know and what we do with it. This information is very much embedded in people. In order to advance civilisation, this transformation process must be easier to reproduce and apply at scale, independently of people.
(1) A first step would be to make factory economics attractive to the average person, so that the cost of production goes down while quality and delivery metrics remain consistent.
That cost encompasses whatever is needed to produce a working unit of something. Let’s say the process requires raw materials, energy and a constructor (robot), and the output is a working unit. We want the cost of each parameter of this process to decline. Generally, the constructor (robot) will be the main cost.
Possibility (as David Deutsch discusses possibility and impossibility) and economic utility are separate problems.
We can prove that a process works and still build something too expensive or difficult to use.
We can also build something economical under ideal conditions that doesn’t really work.
For this to work, there should be possibility (robots that make products) and economic utility (these robots must keep getting less expensive).
(2) The product is the ability to conduct a useful physical transformation.
Intelligence defines the inputs the system accepts, the output it guarantees under those conditions, and what happens when those conditions change. In this case, a constructor (robot) is only one component of the system that makes this possible.
General capability comes from progressively widening the range of transformations the system can perform (similar to a universal constructor). By subsequently adding end effectors and taking on more tasks, the system becomes more general.
(3) Knowledge can be reused without repeating the work that originally created it. That means the next machine still has to be manufactured, and the transformation process can remain agnostic to the machine.
Therefore, each deployment will leave us better able to solve a class of problems. Specialised intelligence becomes better independently of the subset of tasks it operates on, as more deployments are an advantage for all the other deployments.
4
458
Making Manufacturing Work
(This post follows the first introductory post — Manufacturing the World)
Manufacturing is applied information. One material can turn into waste, a working product, or a constructor that makes more products. It really depends on what we know and what we do with it. This information is very much embedded in people. In order to advance civilisation, this transformation process must be easier to reproduce and apply at scale, independently of people.
(1) A first step would be to make factory economics attractive to the average person, so that the cost of production goes down while quality and delivery metrics remain consistent.
That cost encompasses whatever is needed to produce a working unit of something. Let’s say the process requires raw materials, energy and a constructor (robot), and the output is a working unit. We want the cost of each parameter of this process to decline. Generally, the constructor (robot) will be the main cost.
Possibility (as David Deutsch discusses possibility and impossibility) and economic utility are separate problems.
We can prove that a process works and still build something too expensive or difficult to use.
We can also build something economical under ideal conditions that doesn’t really work.
For this to work, there should be possibility (robots that make products) and economic utility (these robots must keep getting less expensive).
(2) The product is the ability to conduct a useful physical transformation.
Intelligence defines the inputs the system accepts, the output it guarantees under those conditions, and what happens when those conditions change. In this case, a constructor (robot) is only one component of the system that makes this possible.
General capability comes from progressively widening the range of transformations the system can perform (similar to a universal constructor). By subsequently adding end effectors and taking on more tasks, the system becomes more general.
(3) Knowledge can be reused without repeating the work that originally created it. That means the next machine still has to be manufactured, and the transformation process can remain agnostic to the machine.
Therefore, each deployment will leave us better able to solve a class of problems. Specialised intelligence becomes better independently of the subset of tasks it operates on, as more deployments are an advantage for all the other deployments.
(4) The economics must support this process. Creating value gives something worth selling.
The value to the customer has to exceed the price. Over time, the price has to cover our costs. What remains can support further development and expansion.
Recurring payments make sense as, technically, labour is provided. This depends in part on inference, which depends on compute and energy.
The cost of making the machine will depend in part on the amount of energy needed to make it, which goes down as we improve the process. And to make it run, inference and energy costs go down as well. At the same time, we aim to generate superhuman levels of labour, which makes all of this economically viable even at a huge scale, provided the price covers the full cost of delivering it.
(5) I sincerely think the advantage has to come from becoming very good at creating, testing and applying manufacturing knowledge, and at making new designs.
Firstly, manufacturing expertise helps identify valuable problems and understand their constraints. Access to factories exposes ideas to conditions we cannot fully reproduce in development. AI engineering helps turn suitable solutions into capabilities reusable across more operations.
Secondly, out-cooperating in areas where other companies already have knowledge or capabilities needed — including — deployment partners, energy providers, compute providers, design and creative partners, and intelligence providers/partners.
As a result, the advantage is an organisation that gets better at solving production problems as it expands, really early and fast, so that we get the results we need and don’t ever have to solve the same issues again.
(6) Substantial growth at scale comes from the output of the system helping to expand the system itself.
Manufacturing will transform energy and raw materials into the machinery used to manufacture more things.
This creates another way to reduce the delta between 0 and factory.
Reusable knowledge reduces the work of discovering how to produce + Existing productive equipment reduces the work of building the equipment needed to produce more = infinite scale
There are three forms of compounding. (1) More production can support more suppliers. (2) More knowledge can make production easier to deploy. (3) More productive equipment can make capacity easier to expand.
Civilisation’s progress towards higher Kardashev levels requires us to turn this into a repeatable process.
Each factory we transform must make the next one easier to build or transform.
2
4
262
Manufacturing the World
Manufacturing is how we turn knowledge, energy and matter into the physical world. It is extremely high leverage because almost everything else energy, compute, transportation, housing, robotics and space infrastructure depends on our ability to make things cheaply and at scale.
From a human, civilisational perspective, growth of civilization eventually requires controlling orders of magnitude more energy (Kardachev Scale). Obv getting there is mostly a production problem. Energy is only useful if we can manufacture the reactors, solar cells, transmission systems, computers and machines required to capture and use it.
(1) I spent time with people who had worked on this problem much longer than me, like @liamcasey who’s renowned for his colossal full stack manufacturing expertise, and noticed that it remains mostly constrained by labor, energy and specialized knowledge.
(2) Making manufacturing easier has large second-order effects. [1] As manufacturing becomes easier and denser, more people could open factories, which creates more suppliers, making it easier for the next company to start. (Compounding effect)
AI and robotics will make it happen, allowing software to program robots to automate the whole production service instead of the specific product production [2]. Programmable production makes factories much easier to reproduce and physical ideas become much cheaper to instantiate.
Future autonomous manufacturing system could go somewhere with energy and raw materials and develop a regions production capacity. For Kardashev growth we really need to care about this, how quickly intelligence can turn available energy and matter into industrial civilisation[s).
It really all comes down to reducing the delta between 0 and factory.
Electronics FATP is a good place to start because a surprisingly large fraction of it reduces to a small set of operations, moving, assembling, inspecting and testing things. Intelligence can make these operations programmable across products, rather than specifically engineered around each product. From there we can gradually expand the set of processes the system can perform.
[1] Shenzhen Special Economic Zone and the manufacturing ecosystem that subsequently developed around it.
[2] Intelligence democratises FATP manufacturing and others by enabling general automation to exist.
instinct helped us set up our office much faster and helps daily with tasks. I want to hear about your use cases , here’s 10k invites valid for a year
app.instinct.com/invite?t=ra…
3
323
Manufacturing is the most important industry in the world. But at what point is it not worth it to verticalise? Here are some thoughts.
at 17 i moved out from morocco alone to asia with the conviction that ai was going to get physical.
we want to get to global abundance, ai and robots are going to get us there. ai needs critical infra that needs to be manufactured and robots need to be too.
our belief is that it all comes down to manufacturing.
i spent there a year around the best people in the manufacturing industry who worked on all the most iconic products you know and grew up with, leading us to start this company.
This quoted post is unavailable.
that’s bigger than openai solving a millenium problem
Adventure through Hyrule and become the Hero of Time when The Legend of #Zelda: Ocarina of Time releases Nov 5 on Nintendo Switch 2. #NintendoDirect
2
336