Original Title: (Focus areas at the intersection of crypto and AI)
Original Author: Kuleen, Head of DePIN at the Solana Foundation.
Compilation: Yuliya, PANews.
Currently, the intersection of AI and cryptographic technology is entering an experimental phase akin to a 'Cambrian explosion.' This paper from the Solana Foundation details three key development directions for AI + crypto integration.
TLDR
1. Build the most vibrant AI agent-driven economy on Solana.
Truth Terminal has already demonstrated the feasibility of AI agents operating on-chain. Experiments in this area are continuously pushing the boundaries of agents' on-chain operations; this field not only has immense potential but also offers a broad design space. This has already become one of the most groundbreaking and explosive directions in the fields of crypto and AI, and this is just the beginning.
2. Enhance LLMs' capabilities in Solana code development.
Large language models have already performed excellently in code writing and will further improve in the future. Through these capabilities, the efficiency of Solana developers is expected to increase by 2-10 times. Recently, establishing high-quality benchmarks to assess LLMs' understanding and writing capabilities for Solana code will help understand the potential impact of LLMs on the Solana ecosystem. High-quality model fine-tuning solutions will be validated in benchmark tests.
3. Support open and decentralized AI technology stacks.
"Open and decentralized AI technology stack" includes the following key elements:
Acquisition of Training Data
Training and Inference Computing Capability
Model Weights Sharing
Model Output Verification Capability
The importance of this open AI technology stack is reflected in:
Accelerate innovation and experimentation in model development.
Provide alternatives for users who do not trust centralized AI.
1. Build the most vibrant AI agent-driven economy.
There has been much discussion about Truth Terminal and $GOAT, so there is no need to elaborate. However, it can be said that when AI agents begin to participate in on-chain activities, a new world full of possibilities has already unfolded (notably, agents have not yet taken direct action on-chain).
While it is still impossible to accurately predict the future development of agents' on-chain behavior, by observing the innovations that have already occurred on Solana, we can glimpse the vast prospects of this design space:
AI projects like Truth Terminal are developing new digital communities through Meme coins like $GOAT.
Platforms like Holoworld AI, vvaifu.fun, Top Hat AI, and Alethea AI enable users to easily create and deploy intelligent agents and their associated tokens.
AI fund managers trained based on the personality traits of well-known crypto investors are emerging, with the rapid rise of ai16z on the daos.fun platform creating a new ecosystem of AI funds and supporting agents.
Additionally, gaming platforms like Colony allow players to participate in games by guiding agent actions, often resulting in unexpected innovative gameplay.
Future Development Direction
In the future, intelligent agents can manage complex projects that require multi-party economic coordination. For example, in the field of scientific research, agents can be responsible for finding therapeutic compounds for specific diseases. Specifically:
Raise funds through the Pump Science platform.
Use the raised funds to pay for access to paid research materials and the computing costs of compound simulations on decentralized computing networks like kuzco, Render Network, io.net, etc.
Recruit humans to perform experimental verification work through bounty platforms like Gib.Work (for example, running experiments to verify/establish simulation results).
In addition to complex projects, agents can also perform simple tasks like building personal websites, creating artworks (like zerebro), with limitless application possibilities.
Why does it make more sense for agents to perform financial activities on-chain rather than using traditional channels?
Agents can fully utilize both traditional financial channels and cryptocurrency systems. However, cryptocurrencies have unique advantages in certain areas:
Micropayment applications - Solana performs excellently in this area, as demonstrated by applications like Drip.
Speed Advantage - Instant settlement functionality helps agents achieve maximum capital efficiency.
Entering capital markets through DeFi - This may be the strongest reason for agents to participate in the crypto economy. When agents need to engage in financial activities beyond payments, the advantages of cryptocurrencies become even more apparent. Agents can seamlessly mint assets, trade, invest, lend, use leverage, and more. Especially Solana, which already has numerous top-tier DeFi infrastructure on its mainnet, is particularly suitable for supporting these capital market activities.
From the perspective of technological development laws, path dependency plays a key role. Whether a product is optimal is not the most important; the key is who can first reach critical scale and become the default choice. As more and more agents earn through cryptocurrencies, encrypted connections are likely to become a core capability of agents.
The foundation hopes to see
The Solana Foundation hopes to see agents equipped with crypto wallets conducting bold innovative experiments on-chain. The foundation does not overly limit specific directions here, as the possibilities are indeed too broad - it is believed that the most interesting and valuable agent application scenarios are likely to be currently unforeseen.
However, the foundation is particularly focused on exploring the following directions:
1. Risk Control Mechanism
Although current models perform excellently, they are still far from perfect.
Cannot grant agents completely unconstrained freedom of action.
2. Promote non-speculative use cases.
Purchase tickets through xpticket.
Optimize Stablecoin Portfolio Returns.
Order food on DoorDash.
3. Development Progress Requirements.
At least reach the prototype stage of the testnet.
It would be best if it has already run on the mainnet.
2. Enhance LLMs' ability to write Solana code, empowering Solana developers.
LLMs have already demonstrated strong capabilities and are advancing rapidly. In the application field of LLMs, code writing may have a particularly steep progress curve, as it is a task that can be objectively assessed. As noted below, 'programming has a unique advantage: the potential for superhuman data expansion through self-play.' The model can write code and run it, or write code, create tests, and then check for self-consistency.
Today, although LLMs are still not perfect in code writing and have obvious shortcomings (for example, poor performance in finding bugs), AI-native code editors like Github Copilot and Cursor have fundamentally changed software development (even changing how companies hire talent). Given the expected rapid rate of progress, these models are likely to completely transform software development. The foundation hopes to leverage this advancement to improve the work efficiency of Solana developers by an order of magnitude.
However, there are several challenges currently hindering LLMs from reaching excellence in understanding Solana:
Lack of high-quality raw training data.
Insufficient number of verified builds.
Lack of high-value interactions on platforms like Stack Overflow.
Historically, the rapid development of Solana's infrastructure means that even code written six months ago may not fully meet today's needs.
Lack of methods to assess models' understanding of Solana.
The foundation hopes to see
Help obtain better Solana data on the internet.
More teams releasing verified builds.
More people in the ecosystem actively ask good questions and provide high-quality answers on Stack Exchange.
Create high-quality benchmarks for assessing LLMs' understanding of Solana (RFP to be released soon).
Create LLM fine-tuning models that perform well on the above benchmarks. More importantly, accelerate the work efficiency of Solana developers. Once high-quality benchmarks are available, the foundation may offer rewards for the first model to reach the benchmark threshold score.
The ultimate significant achievement will be: entirely new, high-quality, differentiated Solana validator node clients created by AI.
3. Support open and decentralized AI technology stacks.
In the field of AI, the long-term power balance between open-source and closed-source models remains unclear. There are indeed arguments supporting closed-source entities to maintain a technological edge and capture the main value of foundational models. The simplest expectation at present is to maintain the status quo - tech giants like OpenAI and Anthropic push frontier developments, while open-source models quickly follow and gain unique advantages through fine-tuning in specific application scenarios.
The foundation is committed to closely integrating Solana with the open-source AI ecosystem. Specifically, this means supporting access to the following elements:
Training Data
Training and Inference Computing Power
Model Weights
Model Output Verification Capability
The importance of this strategy is reflected in:
1. Open-source models accelerate innovation iterations.
The rapid improvements and fine-tuning of open-source models like Llama by the open-source community demonstrate how the community effectively complements the work of large AI companies, pushing the boundaries of AI capabilities (even Google researchers pointed out last year that 'regarding open source, we have no moat, and neither does OpenAI'). The foundation believes that a thriving open-source AI technology stack is crucial for accelerating progress in this field.
2. Provide choices for users who do not trust centralized AI.
AI may be the most powerful tool in the arsenal of dictatorial or authoritarian regimes. Models recognized by the state provide officially sanctioned 'truth,' which is an important control medium. Highly authoritarian regimes may possess superior models because they are willing to overlook citizen privacy to train AI. The use of AI for control is an inevitable trend, and the foundation hopes to prepare for it by fully supporting the open-source AI technology stack.
Several projects in the Solana ecosystem are already supporting the open AI technology stack:
Data collection - Grass and Synesis One are advancing data collection.
Decentralized computing power - kuzco, Render Network, io.net, Bless Network, Nosana, etc.
Decentralized Training Framework - Nous Research, Prime Intellect.
The foundation looks forward to seeing
Hope to build more products at various levels of the open-source AI technology stack:
Decentralized Data Collection: For example, Grass, Datahive, Synesis One.
On-chain Identity: Protocols that support wallet verification of human identity and verification of AI API responses, allowing users to confirm they are interacting with LLMs.
Decentralized Training: Projects similar to EXO Labs, Nous Research, and Prime Intellect.
IP Infrastructure: Enabling AI to license (and pay for) the content it uses.
