@VictorTaelin Why not providing the user the tools to do it themselves?
If the problem to solve is "Bend ecosystem is small" and Bend with AI is very efficient to build the missing pieces of the ecosystem maybe it is not a problem?
English
Gui
754 posts
I just realized that, for the first time ever, progress on both HVM and Bend is bottlenecked not anymore on time, and not on my own knowledge, but on money. I could, right now, be throwing thousands of agents on the repos, with well crafted prompts, and progress would accelerate 10-fold overnight. I know "my agent is outputting 100k lines of code" is a silly meme, but it does apply here. Bend is now equipped with everything it needs to convert agent time into real progress, at any scale. This weekend I validated that, leaving agents working on the standard libraries overnight, and I woke up to thousands of lines of well written, tested, verified code that just worked. One of the agents even started implementing a 3D rendering library in Bend, and it got incredibly far! So, if I can trade money for having ThreeJS in Bend's ecosystem overnight, I think it is not a stretch to say we could port nearly anything to its libraries very quickly. Now, of course, our budget is limited; currently, this is all running mostly with a few agents under a single personal Codex/Claude subscription. But what is preventing me from just throwing, say, 10k on it, and checking what they create, and how far the go? Nothing, and I think this is the next step. It is worth noting that progress here isn't in the "solving new problems" sense; they are not creating anything "new"; they are just implementing existing things in a new language. But that's exactly the "Bend ecosystem is small" problem I'm trying to solve right now! Seems like AI was made for this
English
In this post I misuse the term SFA, as I read it now, it actually means actions that form our self. This is used in the context to determine the Ultimate Responsability. IIUC if one did action that formed himself in a way that made him do something then he is responsible. Like we make decision, everyday, that shapes our character. And this formed character make us do stuff.
Nonetheless IIUC Kane believe in both Agent-causal and Event-causal, and Agent-causal is the property I am using here, the ultimate source of an agent action lie to some degree in the agent.
English
Do not start a computer program rights movement, until evidence of randomness that is different from the expected distribution.
If consciousness is determinstic, e.g. doesn't have influence on the world, only a spectator experiencing the events passing by (this is called Illusionism) then I don't care about program rights because myself doesn't have influence on the world, so it is a no problem.
For consciousness to be an interesting problem it should have an influence on the world. From an observer point of view this influence will appear as Self-Forming Action (SFA) (term used by Robert Kane), IMO those action can be observed as contradicting the current laws of physics.
I don't believe that those self forming action can happen at the creation of the model, like an engineer choosing a specific random value generator that would lead to a different action by the AI during evaluation. This is like hyper determinism this feels way too far fetched. And it would mean the program itself isn't conscious but something else is conscious even during the training and maybe even before the training and influenced the architecture of the AI.
So IMO the AI should create SFA during evaluation. Currently evaluation is deterministic. If we see that evaluating an AI creates more often bit flip in non-ECC RAM, or other randomness get a different distribution, then I will be impressed. But for now we just execute a determinstic program and observe interesting features, this doesn't give any hints for possible consciousness.
Also lots of scientific don't believe in consciousness other than as a spectator illusion. I don't think their framework is relevant for the question. And we should definitely not listen to them when they start a woke 2.0 for the right of computer programs.
English
@alice_und_bob @D0tSama @gavofyork What is your estimation of the cost? I am not sure it is incredibly cheap. There is a reason filecoin is not a simple blockchain.
English
Incentives == cost
You can’t say „you can do it with incentives because it then there is no cost“ :)
And I don’t get the question about new primitives. My idea is just to put documents on an immutable storage. If I want to bear the cost (which is incredibly cheap if you look at it), why bother with IPFS
English
In 2014, @gavofyork described hosting static content as one of the 4 pillars of his Web3 vision.
I proposed develop a mechanism to host website content on Kusama and permanently deploy a library of cypherpunk texts on-chain as POC.
Got rejected from the KSM Art & Social Experiments Bounty. Sad. I would have thought this is gonna be a fun side project that speaks to the cypherpunk ethos of Kusama.
English
Piratebay content is probably peta bytes of data. A block is maximum 10MB, it would take years to simply upload the content, and it would forever bloat the state of kusama and cost the collators. So uploading should be expensive to avoid bloating the state.
It is ok for texts but I don't think it scales for other type of contents.
English
The ask was 69$/h @ 144 hours. I lowered the hourly rate to fit the terms of the bounty
To the idea: Obviously you can debate if it’s meaningful. I would argue that storing simple documents on-chain that don’t just disappear like IPFS does has a benefit. Especially when it comes to cultural artifacts that are worth preserving against link rot.
There is really very little you’d need to do to have a fully censorship resistant PirateBay on-chain
English
Companies have many rituals to increase the value of their product: queueing, people have to queue to get a small food in a small bakery, the struggle makes it tastier, social or financial gatekeeping, like clubs that don't let you in until you put the work to get the access, or hermes/ferrari that don't let you buy main items until you bought the whole other collection. Drop, where the awaited messianic product drop and people rush to it.
Those rituals are good and we miss some of them on the web. Video games bring some rituals to get higher level or item or success but rarely integrated to each other, there is not much concept of space where elements cohabit nearby each other. The only space are social medias which is just a highway directly accessing every boutique in the world. We lack some locality (though some exist with recommendation creating bubbles, but it is hard to navigate from one to the other except scrolling comments).
I wish a mix between X, fortnite and the old reddit. A true augmented reality/virtual reality experience with rituals, locality and navigation, scarcity. Probably hosted on top of autonomous networks as online cities.
English
@Jimbo27979061 @its_ravii @CryptoStats_ @Polkadot It makes sense, the relay chain is the only actual consensus.
English
English
According to @CryptoStats_, @Polkadot is currently the 3rd largest exporter of cross-chain value (after Ethereum and Bitcoin).
~$104M in assets have moved from Polkadot to other networks.
When assets move, it means the network is being used.
English
@CRYPTOM24 @sorpaas Some of his points are legit, like his posts on the nakamoto coefficient.
AFAICT he philosophically disagree with the current direction.
English
@realize_gui @sorpaas Don’t even entertain him! He’s sad because his money has decreased already and has then gone on to be the DOT fudder, while saying if the leadership moves on he will be a DOT Member again 😂😂 full of 💩 is what we call it where I am from
English
One aspect of Polkadot's new 2.1B hard cap is rarely discussed -- the short-term stock liquidation risk.
Remember, $ETC did almost the exact same experiment in ECIP-1017. It didn't bring the expected price actions, but fortunately it also didn't make anything worse. When the reduction comes, miners either shut down their rigs or move to other networks. No shock. Everything works just fine.
The same cannot be said to #Polkadot. When the reduction comes to $DOT, something fundamentally different happens. ~800M DOT is currently locked in staking. Some percentage of those stakers will actively unlock and sell because they suddenly earn less. This creates a one-time liquidity event lasting months, which $ETC never needed to experience.
The important point here is that PoS networks have more complicated economics than PoW networks. A monetary policy change therefore should have been handled more carefully.
In $ETC when miners exit the network, they can only sell their rigs. They get fiat, and it doesn't affect $ETC. In $DOT, however, when stakers exit the network, they will sell their $DOT.
English
We never shipped consumer facing products.
We shipped a blockchain that is quite fast and very reliable and shipped an SDK that is difficult to use, and some feature such as bridges were late.
There is a plan, we've seen the first demo in Argentina.
What do you want polkadot to learn? What past failure are you talking about? Opengov? addressed. Staking? In progress. Unused core time? This is the reason parity is pivoting into product.
English
Rust is inevitable, especially in the AI era. And people are realising it, you’ll soon see prominent ecosystems integrating Rust. Won’t say more but this is already happening
English
Also looking forward for bend2
x.com/i/status/20234…
Taelin@VictorTaelin
So, with Bend2's launch incoming, I'm struggling a bit with the branding. The coolest feature of Bend2 is that it is built from scratch around the idea that we, humans, will stop maintaining codebases. Instead, we write specs - i.e., what we want, as *precise types* - and the AI does the coding, and then *proves that it is correct*. In other words, Bend2 is a way to use vibe coding when you can't risk having bugs at all, and that's something that doesn't exist today. Problem is: Bend1 has already been "marketed" as a language centered around parallelism, and *that is true for Bend2 too*. It will be able to run on GPUs, and will solve most of the Bend1's limitations (2 GB memory, 24-bit numbers, no IO, ultra strict evaluator, etc.). Now, the thing is: how do we market that? Do we talk about all the updated parallelism features? Or do we keep the communication simple and focus about the "vibe coding without bugs" thing? If we talk too much, it may look like feature bloat and not really click to many people. But if we focus only on the AI proof system, it may look like we're completely dropping the old features, which isn't the case. I also wonder if we should rebrand it as ProofScript... "So what is your codebase written in?" "ProofScript!" "Wait what's that?" "Oh it is like TypeScript but we can write these super precise specs and the code is only accepted if the AI proves mathematically the specs are fulfilled. It is super nice because we can vibe code all we want without worrying the AI will break things. You should try it!" "Uh sorry JavaScript is too slow for my serious bank code" "Oh no it compiles to C, and even runs on the GPU if you want to" "Wait what" Hmm I don't know...
English
My bet is humans use a formal language for specification. E.g. Rocq. We don't need to see the sort function we just need to to say this operation sorts the array. And in a constraint environments we may want to specify the execution time or resources usage.
AI generates the assembly, probably using some intermediate language that we usually don't need to see.
Programming languages will be a niche, we won't program anymore we'll just specify.
Andrej Karpathy@karpathy
I think it must be a very interesting time to be in programming languages and formal methods because LLMs change the whole constraints landscape of software completely. Hints of this can already be seen, e.g. in the rising momentum behind porting C to Rust or the growing interest in upgrading legacy code bases in COBOL or etc. In particular, LLMs are *especially* good at translation compared to de-novo generation because 1) the original code base acts as a kind of highly detailed prompt, and 2) as a reference to write concrete tests with respect to. That said, even Rust is nowhere near optimal for LLMs as a target language. What kind of language is optimal? What concessions (if any) are still carved out for humans? Incredibly interesting new questions and opportunities. It feels likely that we'll end up re-writing large fractions of all software ever written many times over.
English