Autor: @knimkar
Tradução: blockchain vernacular
Parece que estamos entrando em uma explosão cambriana de experimentação de casos de uso na interseção entre IA e criptografia. Estou muito animado com o que está acontecendo com essa energia e gostaria de compartilhar algumas das novas oportunidades interessantes que estamos vendo no ecossistema em @SolanaFndn.
1. Breve visão geral
1) Facilitando a economia mais dinâmica conduzida por agentes em Solana, o Truth Terminal é a primeira demonstração do que é possível com agentes de IA quando eles são capazes de interagir na cadeia. Esperamos ver experimentos que ultrapassem com segurança os limites do que os agentes podem fazer na cadeia. O potencial nesta área é enorme e ainda nem começámos a explorar o espaço de design dentro dela. Esta já está provando ser uma das áreas mais inesperadas e explosivas de convergência entre criptografia e IA, e está apenas começando.
2) Enable large language models (LLMs) to perform better in writing Solana code, empowering Solana developers. Large language models have already performed quite well in writing code, and they will become even more powerful. We hope to leverage these capabilities to increase Solana developers’ productivity by 2 to 10 times. In the short term, we will create high-quality benchmarks to measure LLM's understanding of Solana and its ability to write Solana code (see below); these tests will help us understand the potential impact of LLM on the Solana ecosystem. We look forward to supporting teams that achieve high-quality progress in fine-tuning models (we will verify their quality through their outstanding performance in benchmark tests!).
3) Support an open and decentralized AI technology stack. The 'open and decentralized AI technology stack' we refer to is an open and decentralized protocol capable of facilitating access to the following resources: data for training, computing resources (for training and inference), model weights, and the ability to verify model outputs ('verifiable computation'). This open AI technology stack is crucial because it:
Accelerate experimentation and innovation in the model development process
Provide an exit for those who may be forced to use untrusted AI (e.g., state-approved AI)
We hope to support teams and products building at all levels of this technology stack. If you are working on something related to these focus areas, please reach out to the original author!
2. Detailed Overview
Below, we will explain in more detail why we are excited about these three pillars and what constructions we would like to see.
1) Promote the most vibrant agent-driven economy
Why do we care about this? There has been a lot of discussion about Truth Terminal and GOAT, so I won't repeat it here, but it can be stated clearly that the crazy functionalities that AI agents can achieve when interacting on-chain have irreversibly entered reality (and in this case, the agents have not even acted directly on-chain).
We can confidently say that we cannot know for certain what the future of on-chain agent behavior will look like, but to give everyone a sense of how vast this design space is, here are some things that have already happened on Solana:
AI leaders like Truth Terminal are trying to foster a new era of religion through memecoins like $GOAT;
Meanwhile, applications like @HoloworldAI, @vvaifudotfun, @TopHat_One, and @real_alethea allow users to easily create and launch agents and related tokens.
Make investment decisions and fuel their portfolios by training AI fund managers with personalized agents based on various well-known crypto investors. For example, @ai16zdao's rapid rise on @daosdotfun created an entirely new AI fund + agent cheerleader metaverse.
There are also some agent-centric games, such as @ParallelColony, where players give directives to agents to take actions, often resulting in unexpected outcomes.
Possible directions for future development:
Agent management requires multifaceted projects that coordinate economic interests among parties. For example, agents can undertake complex tasks like 'finding a compound that can cure [X] disease.' Agents can perform the following operations:
Raise funds through tokens on @pumpdotscience;
Use the funds raised to pay for relevant paid research costs and compute fees on decentralized computing networks (such as @kuzco_xyz, @rendernetwork, @ionet, etc.) for simulating various compounds;
Using bounty platforms like @gib_work to recruit humans to perform actual work tasks (e.g., running experiments to validate/refine simulation results);
Or perform a simple task, like helping you build a website, or an AI that creates artwork (e.g., @0xzerebro).
There are many other possibilities.
Why does it make more sense for agents to execute financial activities on-chain (rather than through traditional financial systems)? Agents can fully leverage both traditional financial systems and cryptocurrencies. Here are a few reasons why cryptocurrencies are particularly well-suited in certain aspects:
Micropayment scenarios—Solana excels in this area, with applications like Drip already demonstrating its potential.
Speed—instant settlement can be crucial for agents, especially when you want them to achieve optimal capital efficiency.
Access to capital markets through DeFi—once agents begin engaging in financial activities beyond strict payments, the advantages of cryptocurrencies become particularly evident. This may be the most powerful reason for agents to participate in the crypto economy. Agents can seamlessly mint assets, trade, invest, borrow against, and use leverage.
Solana is particularly well-suited to support this capital market activity, as the Solana mainnet already has a rich top-tier DeFi infrastructure.
Ultimately, technology is often path-dependent, and the key is not which product is best, but which product first reaches a critical quality and becomes the default path. If we see more agents creating significant wealth through cryptocurrencies, this may solidify cryptocurrency connectivity as an important capability for agents.
What we hope to see
Bold experiments combining agents with wallets, capable of executing operations on-chain. We have not provided overly specific definitions here, as the possibilities are very broad, and we expect that the most interesting and valuable agent application scenarios are those we cannot predict. However, we are particularly interested in exploring and building infrastructure in the following directions:
At least in prototype phase on the testnet (preferably on the mainnet)
2) Make LLM proficient in writing Solana code and empower Solana developers
Why do we care about this? LLMs already possess powerful capabilities and are rapidly advancing. However, writing code is a particularly noteworthy direction among LLM applications because it is an objectively assessable task. As explained in the post below, 'Programming has a unique advantage: through self-play, superhuman data scaling can be achieved. The model can write code and then run it, or write code, write tests, and check for self-consistency.'
Restrict the negative impacts of hallucinations—current models are very powerful but still far from perfect. Agents cannot be given full freedom to execute operations.
Promote non-speculative application scenarios—such as allowing you to purchase tickets through @xpticket, optimizing yield for stablecoin portfolios, or ordering food on DoorDash.
Currently, even though LLMs are still far from perfect in writing code and have some obvious shortcomings (e.g., they perform poorly at finding vulnerabilities), tools like GitHub Copilot and AI-native code editor Cursor have fundamentally changed software development (even changing how companies recruit talent). Given the expected rapid advancements, these models are likely to revolutionize software development. We hope to leverage this progress to significantly enhance Solana developers' productivity.
However, there are currently some challenges hindering LLM's performance in understanding Solana:
Not enough high-quality raw data for LLM training;
Lack of sufficient validated builds;
There is a lack of sufficient high-value information exchange in places like Stack Overflow;
The rapid development of Solana infrastructure means that code written even 6 months ago may no longer fully meet current needs;
There is no way to assess the model's understanding of Solana.
What we hope to see
Help us publish better Solana data on the internet!
More teams releasing validated builds.
Hope to see more people in the ecosystem actively participate in Stack Exchange, asking good questions and providing high-quality answers;
Create high-quality benchmarks to assess LLM's understanding of Solana (RFP coming soon);
Create LLM fine-tuned versions that score highly on the above benchmarks, and more importantly, accelerate the work of Solana developers. Once we have high-quality benchmarks, we may offer rewards for the first model to reach the benchmark score—stay tuned.
The ultimate achievement here will be high-quality, differentiated Solana validator node clients completely created by AI.
3) Support an open and decentralized AI technology stack
Why do we care about this? It is currently unclear how power in the AI space will balance between open-source and closed-source AI in the long run. There are good arguments for why closed-source entities will maintain a technological edge and capture most of the value from foundational models. The simplest expectation now is that the status quo will continue—large companies like OpenAI and Anthropic push the technology frontier, while open-source models quickly follow and ultimately possess uniquely powerful fine-tuned versions for certain use cases. We hope Solana can tightly integrate and support the open-source AI ecosystem. Specifically, this means facilitating access to the following: data for training, computing power for training and inference, weights of resulting models, and the ability to validate model outputs. We believe this is important for several specific reasons:
A. Open-source models help accelerate debugging and innovation in model development. The open-source community's ability to rapidly refine and fine-tune open-source models like Llama demonstrates how effectively they can complement the efforts of large AI companies in advancing AI capabilities (even researchers at Google pointed out last year regarding open-source that 'we have no moat, and neither does OpenAI'). We believe a thriving open-source AI technology stack is crucial for accelerating progress in this field.
B. Provide an exit for those who may be forced to use untrusted AI (e.g., state-approved AI). AI may now be the most powerful tool in the arsenal of dictators or authoritarian regimes. State-approved models provide a state-sanctioned version of the truth and become a huge means of control. Highly authoritarian regimes may even possess better models because they are willing to ignore the privacy of citizens to train their AI. The question of when AI is used as a control tool is not if it will happen, but when, and we hope to support the open-source AI technology stack as much as possible to prepare for this potential.
Solana is already home to many projects supporting the open-source AI technology stack:
Grass and Synesis One are promoting data collection;
@kuzco_xyz, @rendernetwork, @ionet, @theblessnetwork, @nosana_ai, etc., are providing a wealth of decentralized computing resources.
Teams like @NousResearch and @PrimeIntellect are working on frameworks to make decentralized training possible (see below).
What we hope to see is more product development at various levels of the open-source AI technology stack:
Decentralized data collection, such as @getgrass_io, @usedatahive, @synesis_one
On-chain identity verification: including protocols that allow wallets to prove they are human identities, as well as protocols to verify AI API responses so that consumers can confirm they are interacting with LLMs.
Decentralized training: for example, @exolabs, @NousResearch, and @PrimeIntellect
Intellectual property infrastructure: enabling AI to license (and pay for) the content it utilizes
