Author: Fairy,

Edit: TB,

With the continuous development of blockchain and artificial intelligence technologies, how to effectively combine the two has become one of the goals of many innovative projects. It is in this context that Ambient came into being, committed to combining decentralized blockchain architecture with hyper-large-scale AI reasoning to explore a new intelligent economy model.

As a complete fork of Solana, Ambient retains Solana's high speed and efficiency and creates a completely new blockchain ecosystem by introducing the Proof of Logits (PoL) mechanism.

AmbientWhat is it?

Ambient is a Layer-1 blockchain that combines Solana SVM compatibility with a novel proof of work mechanism to provide hyperscale validation inference. The core concept of the Ambient project is to deeply integrate AI reasoning and blockchain to create a decentralized AI economy.

Unlike traditional proof-of-stake (PoS) systems, Ambient adopts incentives similar to Bitcoin and provides predictable profits for every node involved in network reasoning, fine-tuning, or training. This method avoids dependence on enterprise-level GPUs, and ensures sustainable profits for miners through transaction and inflation-based compensation methods. Both miners and users can receive rewards that match their contributions, and the value of the platform continues to increase with the growth of the network.

Features of Ambient:

• Efficient reasoning and security: Provides fully validated inference with overhead below 1%, while ensuring high security on large intelligent models (600B+ parameters) and their fine-tuned versions.
• Excellent training performance: The training performance is 10 times higher than the existing methods, improving the training efficiency of AI models.
• High miner utilization rate: By optimizing on a single model, the miner utilization rate is improved and the efficiency of the inference and verification process is enhanced.
• Non-blocking proof of work consensus: adopting a non-blocking proof of work mechanism to ensure economic competition in the core activities of the network (inference, fine-tuning, training), while maintaining high TPS, avoiding the performance bottlenecks of traditional blockchains.

AmbientTeam background and development status

Apart from the founder background, Ambient has not disclosed information about other members yet. Travis Good, CEO and Founder of Ambient, has a diverse academic background, and his academic journey covers four fields: government, economics, computer science and machine learning. Travis has a leadership style that focuses on execution and pragmatism. He always sticks to practical operations and focuses on executable solutions when promoting technological innovation. In addition, Travis is very active on Twitter and often shares his unique insights into technology, innovation, and industry trends.

On April 1, Ambient completed a $7.2 million seed round of financing, led by a16z CSX, Delphi Digital and Amber Group. Big Brain Holdings, Superscrypt, Proof Group, Rubik Ventures, Aethir Foundation, Edessa Capital and others participated in the investment. Ambient plans to launch a testnet in the second/third quarter.

Logits ProofConsensus mechanism

The "Logits Proof" algorithm utilizes a key fact: logits (understood as logical units) are both unique fingerprints and can effectively capture the "thinking" state of the model at a specific moment through the hash value generated during the model generation process (i.e. when the model is 'streamed' output). Under this mechanism, the hash value proved by logits is a hash list of the hash values ​​for each set of logits hash values ​​before each output token. In short, for each token n, until the final token t, the hash value of the logits proof is:

Hash(Hash(n) … Hash(t))

The hash value of the logits progress mark proof is the logits hash after x tokens, where x is located between n and t (including n and t), that is:

Hash(n) … Hash(x) … Hash(t)

Based on this principle, a verification mechanism can be constructed: first, the miner generates text; then, the verifier randomly selects a word in the text and asks the miner to provide the "thought state" of the point (i.e. the corresponding logits progress mark proof hash). The validator then performs a reasoning on the word in the same model and context, generating his own "thought state". If the "thought state" of the two (represented by hash) is consistent, the verification will be successful.

This proof of work mechanism is consistent with Bitcoin’s design principles: mining (in this case, the model is repeatedly executed through 4000 tokens inference) is expensive, but the verification process is very cheap (only 1 token’s inference). This mechanism not only improves efficiency, but also ensures the security and reliability of verification.

(This article only introduces early projects and is not used as investment advice.)