Let's Review the AO Whitepaper's Characterization of ICP and AO's security model

General
42

Here is from ChatGPT 4o. Love it so much :laughing:

Why AO is Not Fully Decentralized, Trustless, and Lacks Finality

Centralization in Computation

In traditional decentralized blockchains, every node participates in the computation and validation process, ensuring that no single entity has control over the entire network. However, AO delegates computation to specialized Compute Units (CUs). While these CUs compete to offer their services, the reliance on a limited number of specialized units introduces potential centralization risks. If a significant portion of computation is handled by a few dominant CUs, the system could become vulnerable to central points of failure or influence.

Dependence on External Storage (Arweave)

AO relies on Arweave for storing the message log that forms the basis of its “holographic state.” While Arweave is a decentralized storage solution, AO’s dependence on an external system means that the integrity and availability of AO’s state are contingent on Arweave’s reliability. This introduces a layer of dependency that contrasts with the self-contained nature of traditional blockchains, where all state data is maintained and validated within the network itself.

Limited Trustlessness in Computation

Traditional blockchains achieve trustlessness by ensuring that all nodes validate transactions and compute state changes independently. In AO, computation is outsourced to CUs, and although the results can be verified through deterministic execution and the immutable message log, the initial computation process is not inherently trustless. Users must trust that the chosen CU performs the computation correctly unless they verify the results themselves, which may not always be feasible.

Potential for Centralized Control in Scheduling

AO’s Scheduler Units (SUs) assign a unique, incremental number to each message, determining their order. The centralized assignment of slots introduces a potential risk of manipulation or bias in message ordering. If the SUs are controlled by a few entities, they could influence the sequence of message processing, affecting the fairness and neutrality of the system.

Verifiability Challenges

While AO provides mechanisms for verifiability through deterministic execution and the availability of the message log on Arweave, the process requires additional steps compared to traditional blockchains. Users or other nodes must actively replay the message log to verify computations, which adds complexity and potential barriers to trustless verification. This reliance on additional verification steps could deter users from fully trusting the system without conducting their own checks.

Lack of Finality

A crucial feature of many blockchain systems is the concept of finality, where once a transaction is confirmed, it is permanently added to the blockchain and cannot be altered or reversed. AO lacks this feature because its state is implied by the message log on Arweave and can be recalculated at any time by replaying the log. This means there is no absolute guarantee that a computed state is final and unchangeable, which undermines the confidence that users and applications can have in the immutability of the system’s state.

Conclusion

While AO introduces innovative mechanisms for achieving consensus and scalability, it diverges from the core principles of decentralization, trustlessness, and finality found in traditional blockchains. The delegation of computation to specialized units, dependence on external storage, the need for active verification, and the absence of finality introduce elements that compromise its decentralized and trustless nature. These factors make AO a unique system with different design principles, but not a fully decentralized or trustless blockchain.

9 Likes
43

If you ask chat gpt 4o like this: Why AO is Fully Decentralized, Trustless, and Finality?
You will get exactly the opposite answer.

1 Like
44

YEARLY AO FOR 1 STETH

~ 21.5787 AO

TOTAL BRIDGED STETH

97,531.2927 STETH

and

ckETH Total Supply
Info button
561.9955
ckETH

Chain Fusion work is truly an embarrassing joke! Do you understand what value crypto assets need? Do you understand where the disappointment of ICP holders lies?

45

I don’t need to. It is written already. It is called the AO whitepaper.

3 Likes
46

But these people are bridging assets to farm your AO tokens, right? Pretty sure a lot of these folks plan to dump once they get their allocation.

1 Like
47

This is traffic, this is asset liquidity, and this is also explosive user growth. Since ETH started ICO, the blockchain world has grown this way. What’s wrong with that?

48

Cheer up friends, most projects have stopped updating the code, such as Solana, they have not updated the code for two months, while our community has been updating, we are growing into teenagers, while most projects are still babies, including AO

3 Likes
49

Has anyone here actually used or built anything on Arweave to verify that everything in the whitepaper is real and valid? (as opposed to just being promises made on minimally viable projects)

1 Like
50

The project is not even on production yet.

1 Like
51

I got a question. Why nobody in the Ethereum and Solana community talk about AO?

52

Because no one working in there will promote another project, just here the ones that get paid million usd for grants are feeding their family woth out money, and talking :poop: about us. It’s really painful to see how dfinity spend our money in grants gave to people like lastmjs, the guy promotes another projects that are inferior or just meets his own ideals, while demerit ICP the one that feeds him. 1 million to this person it’s s painful. I would vote to remove his grant immediately

53

To be fair, even Dominic was tweeting about it. In my opinion, it’s super overblown for no reason. Maybe it’s because the IC is still trying to find its niche, particularly with AI, and then AO came along with the same idea. I’m not sure.

(I was just interested in the paper, it has some nice ideas)

54

Because both of them are token databases and don’t have anything to do with verifiable general computing, check how much they talk about another token database, Algorand, when it aired When blockchains meet the real world, only one delivers. (youtube.com)

The Solana community went wild when SUI demonstrated higher (real) TPS. Everyone, including the foundation and founder, was discussing it and making excuses about what should and should not be counted.

The crypto space is very tribal, especially when some unfounded claims are made.

55

lastmjs has done more for ICP than you ever will do

9 Likes
56

Moderator note:

Folks,

Lets please refrain from commenting on an individual (in this case Lastmjs). This is not the decorum we want on this forum.

Thank you

3 Likes
58

That sounds about right because I actually went through Arweave’s ecoystem of projects and I haven’t come across anything Ai nor AO related. Surprisingly, one dapp that’s listed on ICP is also listed on Arweave, i.e. DSocial (a decentralized Youtube) - and it also doesn’t work on both sites.

I renew my call for anyone to demonstrate the capabilities of Arweave, esp. its AI fully on chain capabilities, beyond just referring to a White paper. Do what Dominic did when he first demonstrated AI running on the blockchain. Also, post the link here so I can go look at and try the dapp and Ai myself. If that can’t be done then it sounds like the typical hype and promises of crypto projects.

4 Likes
59

I think a bigger threat to obsolete ICP is MegaETH. Claims to bring Web2 performance (real-time speed and general computation) to Ethereum using an EigenDA L2. MegaETH

1 Like
60

I think folks here may find this interesting :

5 Likes
61

Hey @lastmjs! Considering concrete points from the white paper where IMO it does not represent ICP fairly. To be honest, I think almost every sentence in the white paper that relates to ICP is wrong. Here are a few:

ICP employs a single Byzantine Fault Tolerant (BFT) mechanism across its ’subnets’, a design choice that mandates consensus on the results of computations.

Technically, ICP achieves consensus directly on the inputs (by inclusion in and consensus over a block) and indirectly on outputs (through deterministic computation and certification). I guess that’s reasonably close, but I still find their description hard to understand.

This necessitates that every node within a subnet execute every step of each computation, inherently limiting the amount of computation that can be feasibly performed due to scalability constraints.

This seems completely confused. Computation on a single subnet is indeed bounded by the capacity of the nodes running the subnet. But the scalability approach of ICP is horizontal by adding more subnets, and the protocol scales extremely well in terms of subnets thanks to chain-key cryptography.

Furthermore, ICP adopts a monolithic protocol structure, enforcing uniform consensus and execution parameters across all resident containers.

This is also not true, some execution parameters do differ between subnets. And consensus parameters are also different for subnets of different sizes.

Consequently, also the “contrasting” statements about AO that follow up are misleading.

The one aspect that I think is represented fairly is that ICP relies more on in-protocol governance. The statement

This centralized control is akin to a public-company operated by its shareholders, potentially leading to discriminatory practices against certain protocol uses.

however, is again misleading as to (a) the word centralized – the NNS is of course a singleton component, but it does not encode centralized control according to the usual web3 lingo, and (b) it does completely disregard that control in less governance-focused platforms is usually also quite centralized around small groups of people that contribute to the same code that everyone runs. (So it only discusses the point for one approach, not for both.) Similar regarding the use of the term KYC which indicates centralization while actually it’s just a vote of the DAO.

13 Likes
62

small nit:

as a former PhD, I am suspicious of obtuse writing vs simple, easy to verify, easy-to-invalidate writing.

I found the AO paper to be too lofty or academic for my tastes, but didnt boil down the key things i was looking for in AI x Blockchain.

This was one (of many) of the inspiration for writing this: What Makes AI on Blockchain Hard? [Request for feedback on post]

1 Like