a_tartaruga

I don't think you asked AI to try to find issues with your argument. I asked Claude and it gave me three reasons that this is not a proof. If you are going to use AI like this at least check your own work.

The great thing about LLMs is they will read your definitely incorrect things and try to help you find errors.

Sounds like a pretty fun project to build. And bootstrapping it is nice. Hope my feedback doesn't come across as too harsh:

Not open sourcing code or fundamentals of the design is a non starter for anyone wanting to get involved. Are validators running your DRMed binary or something? You don't explain the design in enough depth for anyone to seriously evaluate it.

Transformer models predicting gas seems far too complicated for something that is not a real problem. And I would be concerned trusting today's AI models with selecting things like epoch cycle parameter. Overall the AI component sounds like both an unnecessary over complication and a security risk.

Embedded AI Agent hooks -- are you using some oracle or making this trustless in some way? The later thing would be quite innovative but much harder.

My overall feedback is that there appears to be no significant technical differentiator here. In light of this you are holding yourself back by not open sourcing.

The obvious answers for big charismatic discoveries are Higgs Boson detection, gravity wave detection and CRISPR CAS. I'd love to hear about less widely known ones.

Sewer AI (https://www.sewerai.com). But it predates LLM things.

No but that fact that kids seem to like it and people here hate it so much makes me think that it is a legitimately novel thing in the world. I would like to try it out but I bet it will be hard to get a satisfying feel for it since I've come up in the era of rigor and understanding.

This is a good resource http://the-knowledge.org/en-gb/the-book/

You might be thinking of Donella Meadows.

> Have there been any attempts to make it at least somewhat useful?

To answer this part there are a few networks out there doing useful work. They are getting called depin these days (decentralized physical infrastructure). A lot of them have broken/silly security models but some of them actually work.

It is in principle a good idea. There are plenty of ongoing attempts to do something similar but less cool than this with GPU networks like render etc. From what I've heard they all suck at AI workloads so far. For years I've heard of networks specifically designed for model training in the works but haven't noticed anything really make it that looks serious. But the details are tricky. You can immediately start finding a wall of interesting problems if you start thinking about doing this.

Hashing is extremely easy to verify and therefore extremely easy to pay out for. More complicated computations are more expensive and complicated to verify. You can in principle run verifiable compute (snarks, starks) on training procedures to be sure distributed trainers are doing something that deserves to be paid out for. But how do you break this down so that it is incremental? How do you sequence updates? How do you ensure availability of updated weights so that trainers aren't withholding to gain an advantage? i.e. where does the data go? You're probably not keeping 100 billion parameters on chain. How do you keep costs of all this from ballooning like crazy (verifiable compute is kind of expensive) ?

The little I know about these models the data is pretty important. How do you keep data quality high? How do you verify that data quality is good? Probably you are committing to public datasets and verifying against them. Is that good enough in this crazy world where the state of the art is training the entire public web? How do you get a commitment to "the internet" to prove against? How do you make sure trainers aren't redoing the same datapoint over and over?

I think you can solve all this with enough work. Especially as the cost curve for verifiable compute keeps descending down you will probably find doors opening. Or if you bite the bullet and trust hardware vendors you can maybe have something with a decent security model that is practical today using trusted enclaves. But you've got to solve a lot of problems.

It's only a matter of time until we get a railroad track laying network secured by proof of railroad track (PoRT) and recreate the panic of 1873.

One in particular gets about 1 billion dollars a year. Already hit their quota in February

As others have pointed out this is possible using confidential computing / secure enclaves. It's actually a pretty great application if I understand what you are doing correctly because you are running a relatively simple and self contained process in the enclave. Don't use SGX which has been pwned so many times it is a meme: https://en.wikipedia.org/wiki/Software_Guard_Extensions#List.... AWS has something that looks ok for VPS. AMD SEV-SNP is pretty nice. I know you can get SEV-SNP working on hosted dedicated servers not sure about VPS providers.

If you want to do a crazy science project you could look at the Marlin network which apparently lets you execute code in enclaves on random people's computers. Probably the throughput these random people would not be practical to run your VPN.

MANIAC & Turing's Cathedral

This is the best for what it is trying to do

If you rearrange the way you see things it's simple to not be sycophantic at all. If you're interested in contributing to their work you will be of value to the team you are trying to join. Don't think about it as them doing you a favor. It's a positive sum game and by reaching out you're potentially making everyone richer.

On top of all the great advice here make sure you do as many high quality problems as you can. Some easy ones to build your confidence, a majority that are tough but doable and then occasionally try things beyond your current abilities.

If you can make time for an in person class at a local university or community college 1-3x a week it will probably 10x your progress.

As others are saying you should focus on linear algebra which is the doorway to depth in almost all technical subjects.

There's a weird unnamed proto-field I've seen signs of emerging out of many different areas. I don't know what to call it, maybe digital construction or programmable matter.

Examples include Lee Cronin and Sara Walker's assembly theory, David Deutsch and Chiara Marletto's constructor theory, Neil Gershenfeld's work. Closely related are ideas from synthetic biology and molecular nanotechnology. Some of these people are trying to advance complex systems theory and involved with SFI. You can see this thing trace back to at least the 50s with Turing's morphogenesis paper and Von Neumann's work on automata.

All of this seems unorganized but promising.

Real Life Lore

It's hard to say as it depends a lot on the particular attack, who discovers it, who it is used against etc.

A broad attack on EC DLOG or on the secpk curve would hopefully be preceded by academic community incrementally making it weaker with easier attacks which would give time to fork to something with better margin of safety in advance.

If it was truly a full on zero day and someone disclosed it to core devs or miners it would probably be bumpy getting the network aligned to fork but people would do it. But idk the details of how bitcoin developers and miners coordinate.

In more likely event nation state is the zero day holder I have little idea how they'd use it. Network could always be forked after enough people believed an attack was going on but it would be a confidence shaking disaster. Forking back in time before most people believe the attack started is always possible but has its own serious downsides.

This is a good point. I think this is probably the weakest link in bitcoin's security. If you're paranoid I think you can setup PayToScriptHash to take lamport signature instead. But with everyone else rekt the system would be in crisis.

In terms of value stored my wild speculation is that whatever shady business is going on with tether will eventually implode and the contagion will be extreme. People have been saying this for 10 years and nothing has happened yet so who knows.

But in terms of the network itself failing there is really not much you can do.

Fight club / Mr robot style terrorist attacks are pretty much out. Maybe a major nation state would have some success but I doubt it. States might have success sowing confusion by compromising core developers and pushing software updates to nodes that do damaging things. I don't see this totally killing the network though.

Banning through legislation needs to be done across all or large majority of nation states which I don't see as being realistic. Though maybe as the environment faces increasing catastrophe this will become more possible.

Quantum computer attacks can be defended by relatively trivial forks that everyone involved has a reason to support. I think you can even do Lamport signature stuff today with the pay-to-script-hash stack machine language.

Maybe someone finds an exploit in sha2 and launches a surprise attack. That would cause significant damage. I would guess the network would fork itself onto a secure hash function quickly enough.

If someone takes out the internet for good that would probably kill the network.

In all of these scenarios there is the possibility of restart if the conditions killing the network revert.

I think they're not selling them anymore but everything is there for you to build one yourself: https://openpcr.org

For me golang vibes are better.

Go is a little too verbose but in a stupid way. Rust is too verbose in a clever way. Go is so dumb it is mind numbingly boring. Rust is so clever it becomes boring to learn.

You can use go in about 1 way that works. Rust you can use in many different ways. My guess is that really knowing rust is 2-5x harder than really knowing go.

To really get into go you need to understand the runtime, scheduling and especially GC. For rust you need to at least understand the compiler the macro language and of course the borrow checker. Then if you want to do concurrency you need to look into async await or whatever is going on these days.

For whatever reason I find the go runtime fun to learn about and the rust compiler tedious to learn about.

Go tools are smoother and easier to learn. I know it makes me a philistine but fast compile times are nice.

My opinion is that it's good not to have a vision for your career yet. The world is rapidly changing and this is an online learning problem. Keeping trying things and seeing what you like. Continue learning things.

On the practical side consider joining a random engineering project team that interests you (IGem for example). Get practice building things with people in a welcoming environment.

^^ This book is the best

I heard about it from the same HN post

Children and their circumstances are so different that it's kind of easy to find yourself in a situation where no one knows what to do. Like everything that matters experience is the only real teacher. Keep trying.

If you can talk to people from your parents' generation that had children. Everyone has a story or a skill to share that helps make it all make more sense.

If you can afford to hire professionals they are generally more helpful than books. Doctors obviously are good to consult. Post partum doulas have lots of random knowledge that they can diffuse to you while they help out around the house. If you are having feeding issues OTs and lactation consultants can be quite useful.

Making friends with new parents will allow you to learn together.

Penny Simkin's book "The Birth Partner" is mostly about birth but had helpful ideas on sleep rotations and early feeding.

Random but I found this less wrong thread helpful for working through antivax ideas in good faith: https://www.lesswrong.com/posts/oFXXfP88nJrseeoMD/the-litera.... Particularly Lao Mein's comment. It is a very difficult topic to deal with head on since biology is so complicated and the surrounding discourse is all poison.

Have you thought about taking one class a semester for the next year without quitting your job? How important is this really to you? Have you been channeling your interest into building things for fun? Are you tearing through The Art of Electronics? Love of knowledge is a strong motivator but you might be able to satisfy it with a less disruptive change to your life.

Since you know the ads surveillance Borg so well you are uniquely positioned to understand how to solve the problems it has created for everyone. Spend a lot of time thinking about why your past work doesn't sit right with you and the conditions that caused you to be paid to do it. Then consider what could be changed that would make these conditions slightly better. That will help you find other groups whose work you value or come up with your own ideas to pursue.

There are extremely obvious problems in energy and biotech that everyone wants solved but they are harder to break into with your background. However it's probably not truly that hard to contribute in a software capacity if your horizon is a few years.

Maybe there is a lot more value to be created by fundamentally rethinking coding but anecdotally the absolute value provided by cursor is significant. Prototyping things in a new language or framework for the slow and lazy is now ~5x faster.

One good example is moving industrial processes to solar power.