Axis ($AXIS)

Introducing Cross-chain Yield Arbitrage,
Your @USDC, always in the highest yield.
Zero-slippage, maximum APY.



🧵🐐

Plasma rapidly becomes the stablecoin liquidity vacuum.

Pay attention.

plasma mainnet beta launches at 1pm UTC

• current price on pre-markets: 0.79$
• implied FDV of $7.9b and market cap of $1.4b (8x less than Sui)
• key day one projects: Aave, Ethena, Axis, etc.
• stable chain / tether / paolo proxy play
• tier 1 CEX/DEX listings right away: binance, bybit, hyperliquid, extended
• traders hungry for quality liquid plays (e.g. aster)
• most ICO participants want MORE exposure
• most crypto natives are sidelined (only ~3k wallets participated in the ICO)
• $2b+ of stables live day one, will propel plasma to top 5 chain in tvl fast
• plasma one neobank launching soon (10% yield + 4% XPL cashback)
• heavy XPL incentives during the Veda vault → Aave USDT market migration period (5 days) with 48h cooldown, then 24h once market stability is reached
• planned to have lowest USDT borrow rate across all chains
• think anything below $7.5b is a longer term buy

trillions

Russia is just Mexico with nukes; I wouldn’t be concerned either

wow, looks like DeSci summer is soo back

currently fast-growing narrative

and still FDV is under $1B

i guess make sense to pay attention on research

do u need my tier list?

I think $PENGU will pull off an $AXIS infinity move.

If you were here 5 years ago you know what I’m talking about

Why must data be encrypted for AI training, and why is confidential computation hard to achieve?

Everyone knows about AI, and many people use it daily, but the catch is that AI models can use your sensitive data to train models that others rely on.

Many companies warn their employees not to input private company data or confidential information into tools like ChatGPT for work optimization.

This is because such data could be used to train ChatGPT, potentially exposing private and sensitive information to others.

Moreover, the data that is not encrypted can be harmful to both the person who owns the data and the AI model itself, so:

• what are the problems in the decentralized AI?
• why should we care if data is public or private?
• why is MPC the best solution for data encryption?
• can we ensure that private data is computed correctly?
• does private data matter in DeFi as well?
• does every AI model need encrypted data?

External privacy solutions like @ArciumHQ help you compute on encrypted data while letting you verify that the results haven’t been manipulated.

Confidential computing is a complex topic involving privacy, security, economics, speed, and the architecture of the solution. So, what problems are we facing?

1. There are many problems in the AI space, especially in decentralized AI.

While many may not notice, there are actually many problems in training AI models in a decentralized way. First of all, in most cases, it’s just inefficient. Training large AI models requires a lot of computational power and is usually handled in a centralized way.

Decentralized systems are usually not optimized for handling large volumes of data or performing complex computations, which are essential for AI, especially during the training phase.

Training large AI models requires significant computational resources; coordinating these tasks across distributed nodes usually leads to performance bottlenecks.

Even if we achieve the same performance in a decentralized system as in a centralized one, ensuring that the AI model is trustworthy, as well as its outputs, is difficult.

We face a problem of data verifiability, where we need systems that allow distrusted entities to collectively solve problems while pursuing local objectives. The validity of predictions in decentralized settings is hard to compile, especially without implicit trust.

Another problem is using sensitive data in training AI models. Data is processed in plaintext form, so sensitive information could be exposed. But first of all, what is sensitive data, and why is it important?

2. Why should we care if data for training AI models is public or encrypted?

We care about whether AI training data is private or public because it affects data security, transparency, and even model quality.

When AI is trained on public data, the information is openly accessible, which can pose risks if sensitive details are included or if the data is misused. Public datasets don’t offer safeguards for protecting personal or confidential information.

Moreover, public data often lacks the diversity needed to build great AI models. These datasets can be limited in scope, leading to biases that can cause AI to perform poorly in real-world scenarios.

• For example, mental health data is tightly guarded by laws, which limit its use in research.

• Patients often hesitate to share such data due to stigma and privacy concerns, resulting in underrepresentation in datasets.

• This can lead to AI models that struggle with mental health cases, potentially causing biased diagnoses and treatments.

Encrypted data keeps sensitive information secure by ensuring it remains confidential, even during the training process. It can even improve fairness because it can incorporate diverse scenarios without actually revealing them.

What we care about is how different features behave rather than which exact data is used.

3. Should we care about data encryption only in the training of decentralized AI models, or does it also matter in dapps?

As everyone knows, blockchain transactions are public and can be verified by everyone on different scanners or explorers, but some people want to keep their transactions private and maintain confidentiality.

However, we face the problem of verifiability again. If data is encrypted and private, what are the ways we can verify this and ensure a network participant tells the truth and doesn’t cheat the system?

Not everyone needs to know exactly which operations you have performed on the blockchain, but everyone needs to know if they can verify all operations to maintain a trustless system.

As such, it can be possible to even mitigate sandwich attacks when performing large DEX trades, so encryption is not only about verifiability but sometimes also about financial health.

What we have talked about so far can be called confidential computing. One of the protocols that ensures data is encrypted, secure, and verifiable during computation for both AI and DeFi use cases is Arcium.

4. How does Arcium ensure that data is encrypted and secure?

First of all, Arcium is a decentralized network for confidential computing. There are different techniques for confidential computing, but as most people know, there are three main ones:

1. MPC (Multi-Party Computation)
2. FHE (Fully Homomorphic Encryption)
3. TEE (Trusted Execution Environment)

The choice of Arcium is MPC, and I’ll explain why later, as well as the issues FHE and TEE pose. MPC allows multiple nodes to compute data without revealing it, keeping it encrypted throughout the process.

• In an MPC setup, multiple parties (nodes) jointly compute a function over their inputs while keeping those inputs private.

• A distributed architecture uses multiple Arx nodes to process encrypted data collaboratively.

• No single node accesses the complete dataset, ensuring privacy.

MPC is decentralized by definition, meaning that each node only holds a part of the whole dataset. This keeps the data encrypted throughout its entire lifecycle: storage, transmission, and computation.

But how can we ensure that all nodes are honest and the data is indeed encrypted? What happens if one or even most of the nodes misbehave?

The problem is called the Dishonest Majority Assumption, which assumes that in a secure MPC, the majority of parties may be dishonest.

Arcium implements a slashing mechanism to penalize misbehaving nodes, which addresses this issue by acting as a disincentive to prevent dishonest behavior.

The Dishonest Majority Assumption also enables effective detection of misbehaving nodes, allowing for appropriate punishment.

While the slashing occurs after computation, the combination of detection and disincentive helps ensure the correctness of the computation, even when the majority of nodes might misbehave.

Cerberus (primary MPC protocol of Arcium) keeps data private even if only one node stays honest. Data is encrypted on the client side, then split into shares and distributed among Arcium nodes.

These nodes compute on the shares to secure processing even under the assumption that most nodes might be dishonest.

So, which components of the architecture make confidential computation possible besides the nodes with MPC themselves?

5. How does Arcium work?

Arcium is not a blockchain: Arx nodes, which handle confidential computing, operate independently of Solana. Only the orchestration layer initially runs on Solana, with confidential computing occurring off-chain in Arx nodes.

There’s no need to build a separate blockchain for privacy since privacy should be universal and accessible on existing public blockchains.

Arcium has several key components that work together seamlessly: Arx Nodes, clusters, MXEs, computations, operators, delegators, and more.

To understand the bigger picture as well as the relationships between different components, you can look at the bigger picture I have crafted below:

Arx Nodes are individual computers that handle pieces of data, processing them securely. These nodes join forces in clusters, which are groups carefully selected based on their reputation and resources to tackle specific tasks.

Then there are MXEs, or MPC Execution Environments, which act as custom workspaces for computations.

The typical flow of data looks like this:

1. Confidential computing tasks in Arcium are split into individual computation definitions.

2. These computation definitions are carried out within specialized environments known as MXEs.

3. These MXEs run on top of clusters, which are groupings of Arx nodes that work together.

4. MXEs define and schedule the computation definitions and confirm they have the capacity to execute the defined circuits.

5. After the nodes in the cluster compute the result and sign it, it’s sent to Solana to be used by other apps where it is live and verifiable.

6. External delegators can contribute their stake to Arx operators to earn rewards, while Arx operators are individuals who run Arx nodes.

7. Arx operators can also stake their own funds. If an operator’s node is cheating or experiencing downtime, it’s slashed.

You can think of clusters as the system’s hardware, MXEs as its dynamic state, computation definitions as the functions you write, and each scheduled computation as a unique function call in action.

Once the computation is complete, the results remain encrypted and are collected for return to the user. The network combines the shares from all nodes to produce a final encrypted output that the user can decrypt, ensuring end-to-end confidentiality.

Results can be public data, globally encrypted data, or even data encrypted only for a specific party.

6. Why is MPC needed? Can’t we use TEEs or FHE?

MPC lowers risks and offers stronger security in trustless settings compared to TEEs or FHE. It splits data so no single party sees it all, yet still enables safe collaboration.

In terms of speed, MPC can achieve performance improvements of hundreds of thousands of times compared to the fastest FHE libraries, making it more suitable for real-time applications.

MPC is software, so it works on any device and can be adjusted for different jobs. FHE is very slow because it’s complex, making it hard to use for big or fast jobs. TEEs only work on specific hardware, so they are less flexible and tough to scale up.

In terms of privacy, Arcium’s implementation of MPC keeps data fragmented throughout the process and frequently checks for cheating. No one ever sees the complete data, and the results stay accurate.

FHE also keeps data encrypted, but it doesn’t inherently verify if tampering has occurred. TEEs decrypt data within their secure environments, so if someone breaches them, the data contained within gets exposed.

7. Which apps directly benefit from encrypted data and Arcium?

Developer teams can use Arcium to process fully encrypted data. In AI, it enables training models on encrypted healthcare datasets for tasks like disease prediction or personalized medicine, ensuring patient privacy.

This capability also applies to finance for fraud detection and to the legal sector for analyzing confidential files.

In DeFi, Arcium supports private apps, such as DEXs where transactions stay encrypted, or confidential lending platforms and privacy-preserving stablecoins.

For gaming, it powers confidential on-chain experiences, like poker where players’ cards remain hidden, or strategy games and auctions requiring secrecy.

This technology can also be applied to strategy games where players’ moves need to stay hidden or auctions where bids remain confidential until revealed.

8. Does every AI model need encrypted data?

Some AI models use data that’s already public or don’t involve personal details, so there’s no need to encrypt the data.

For example, imagine an AI that identifies dog breeds from photos you can find online. Since those pictures are already out there for anyone to see, encryption isn’t necessary: there’s no privacy to protect.

In DeFi, for example, data encryption is a must, when sensitive financial details need protection on public blockchains, where transactions are normally visible.

It’s key for trading on DEXs to hide orders and prevent front-running, for lending to keep loan and collateral amounts private, and for dark pool swaps to shield identities and sums.

It’s also needed when AI analyzes data like market trends to encrypt wallet balances or trades, focusing solely on price instead of exposing personal details that could interfere with the model’s training to achieve accurate results.

The core point is that not every AI model needs encrypted data, nor does every dApp. We live in a world with billions of terabytes of public data, and many apps operate solely using public data.

However, when private data is needed, let's ensure it's encrypted, utilized correctly, and can be verified.