Written by: @knimkar
Translation: Blockchain in Vernacular
We seem to be entering a Cambrian explosion of use case experiments at the intersection of AI and crypto. I'm really excited about what's coming out of this energy and wanted to share some of the exciting new opportunities we're seeing in the ecosystem at @SolanaFndn.
1. Brief Overview
1) Facilitating the most vibrant agent-driven economy on Solana Truth Terminal is the first demonstration of what is possible with AI agents when they are able to interact on-chain. We look forward to seeing experiments that push the boundaries of what agents can do safely on-chain. The potential in this space is huge, and we haven’t even begun to explore the design space. This has proven to be the most unexpected and explosive area of crypto-AI integration, and we are just getting started.
2) Empower Solana developers by making Large Language Models (LLMs) even better at writing Solana code Large Language Models already work pretty well for writing code, and they’re about to get even more powerful. We hope to use these capabilities to make Solana developers 2x to 10x more productive. In the short term, we’ll be creating high-quality benchmarks to measure the ability of LLMs to understand Solana and write Solana code (more on that below), which will help us understand the potential impact of LLMs on the Solana ecosystem. We look forward to supporting teams that make high-quality progress in fine-tuning their models (we’ll validate their quality by how well they perform on our benchmarks!).
3) Support for an open and decentralized AI technology stack By “open and decentralized AI technology stack” we mean open and decentralized protocols that facilitate 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 important because it:
Accelerate experimentation and innovation during model development
Providing a way out for those who might be forced to use untrustworthy AI (e.g. state-sanctioned AI)
We want to support teams and products building at all levels of this stack. If you’re working on something related to these focus areas, get in touch!
2. Detailed Overview
Below, we explain in more detail why we’re excited about each of these three pillars and what we’d like to see built.
1) Promote the most dynamic agent-driven economy
Why do we care? There’s been a lot of discussion about Truth Terminal and GOAT, so I won’t repeat it here, but what can be said is that all kinds of crazy functions that are possible when AI agents interact on-chain have irreversibly entered reality (and in this case, the agents haven’t even taken actions directly on-chain yet).
We can confidently say that we currently don’t know exactly what the future of on-chain proxy behavior will look like, but to give you a sense of how broad this design space is, here are some of the things that are already happening on Solana:
AI leaders like Truth Terminal are trying to foster new age religion through memecoins like $GOAT;
At the same time, applications like @HoloworldAI, @vvaifudotfun, @TopHat_One, and @real_alethea allow users to easily create and launch agents and related tokens.
AI fund managers are trained to personalize agents of various well-known crypto investors to make investment decisions and cheer on their portfolios. For example, @ai16zdao’s meteoric rise at @daosdotfun has created a whole new metaverse of AI funds + agent cheerleaders.
There are also agent-centric games like @ParallelColony, where players instruct the agent to take actions, often with unexpected results.
Possible future directions:
An agent manages a multifaceted project that requires economic coordination among various parties. For example, an agent could be tasked with a complex task like “find a compound that cures disease [X].” An agent can do the following:
Raising funds through tokens on @pumpdotscience;
Use the funds raised to pay for access to relevant paid research and to pay for computation on decentralized computing networks (such as @kuzco_xyz, @rendernetwork, @ionet, etc.) for simulations of various compounds;
Use bounty platforms like @gib_work to recruit humans to perform tasks that perform real work (e.g., run experiments to validate/refine simulation results);
Or an AI that performs a simple task like building a website for you, or creates a work of art (e.g., @0xzerebro).
There are many other possibilities.
Why does it make more sense to have agents perform financial activities on-chain (rather than in the traditional financial system)? Agents can leverage both the traditional financial system and cryptocurrencies. Here are a few reasons why cryptocurrencies are particularly suitable in some ways:
Micropayment scenarios — Solana excels in this area, and applications like Drip have already demonstrated its potential.
Speed – Instant settlement can be critical for agents, especially if you want them to be optimally capital efficient.
Access to capital markets through DeFi — The advantages of cryptocurrencies become particularly clear once agents begin to perform financial activities beyond strict payments. This is probably the most powerful reason for agents to participate in the crypto economy. Agents can seamlessly mint assets, trade, invest, borrow, use leverage, and more.
Solana is uniquely suited to support this capital markets activity because of the rich set of top-level DeFi infrastructure already available on the Solana mainnet.
In the end, technology is often path dependent, and the key is not which product is the best, but the first one to reach critical mass and become the default path. If we see more agents creating significant wealth through cryptocurrencies, this could solidify crypto connectivity as an important capability for agents.
What we hope to see
Proxies combined with wallets are bold experiments that can perform operations on-chain. We do not give a too specific definition here because the possibilities are very broad and we expect that the most interesting and valuable proxy application scenarios are those that we cannot predict. However, we are particularly interested in exploring and building infrastructure in the following directions:
At least in prototype stage on testnet (preferably on mainnet)
2) Make LLMs good at writing Solana code and empowering Solana developers
Why do we care? LLM is already powerful and improving rapidly. But writing code is a particularly interesting area of LLM application because it is a task that can be objectively evaluated. As explained in the post below, "Programming has a unique advantage: by "playing against itself", superhuman data scaling can be achieved. Models can write code and then run it, or write code, write tests, and check self-consistency."
Limiting the negative effects of hallucinations - Current models are very powerful, but still far from perfect. Agents cannot be given complete freedom to perform actions.
Powering non-speculative use cases — for example, allowing you to buy tickets with @xpticket, optimize yield on your stablecoin portfolio, or buy food on DoorDash.
While LLMs are still far from perfect for writing code today and have some notable shortcomings (for example, they are poor at finding vulnerabilities), tools like Github Copilot and the AI-native code editor Cursor are already fundamentally changing software development (and even changing how companies recruit talent). Given the rapid progress projected, these models are likely to revolutionize software development. We hope to leverage this progress to make developers on Solana an order of magnitude more productive.
However, there are currently a few challenges that hinder LLM’s performance in understanding Solana:
There is not enough high-quality raw data for LLM training;
Lack of sufficient validated builds;
There isn’t enough high-value information exchanged in places like Stack Overflow;
Solana infrastructure evolves rapidly, which means that even code written 6 months ago may not be fully suitable for current needs;
There is no way to assess how well a model understands Solana.
What we hope to see
Help us publish better Solana data on the internet!
More teams release verified builds.
I hope more people in the ecosystem will actively participate in Stack Exchange, ask good questions and provide high-quality answers;
Create high-quality benchmarks to assess LLMs’ understanding of Solana (RFP coming soon);
Creating fine-tuned versions of LLM that score high on the above benchmarks and, more importantly, accelerate Solana developers’ work. Once we have high-quality benchmarks, we may offer a bounty for the first model to reach the benchmark scores — stay tuned.
The ultimate achievement here will be a high-quality, differentiated Solana validator client created entirely by AI.
3) Supporting an open and decentralized AI technology stack
Why do we care? It’s unclear how power in AI will balance between open and closed source AI in the long term. There are good arguments for why closed source entities will remain at the forefront of the technology and capture most of the value from the underlying models. Right now, the simplest expectation is that the status quo will continue - large companies like OpenAI and Anthropic push the technological frontier, and open source models will quickly follow, eventually having uniquely powerful fine-tuned versions for certain use cases. We want Solana to be closely aligned with supporting the open source AI ecosystem. Specifically, this means facilitating access to: data for training, compute power for training and inference, weights for resulting models, and the ability to validate model outputs. We think this is important for a few specific reasons:
A. Open source models help accelerate debugging and innovation in model development How quickly the open source community can refine and fine-tune open source models like Llama shows how the community can effectively complement the efforts of large AI companies in advancing the frontier of AI capabilities (even Google researchers pointed out last year that "we have no moat, and neither does OpenAI" regarding open source). We believe that a thriving open source AI technology stack is critical to accelerating the pace of progress in the field.
B. Provide an outlet for those who may be forced to use AI they don’t trust (e.g. state-sanctioned AI) AI is now perhaps the most powerful tool in the arsenal of a dictator or authoritarian regime. State-sanctioned models provide a state-sanctioned version of the truth and become a huge means of control. Highly authoritarian regimes may also have better models because they are willing to ignore the privacy of their citizens to train their AI. The question of AI being used as a tool of control is when, not if, and we want to prepare for this possibility by supporting open source AI technology stacks as much as possible.
Solana is already home to many projects supporting the open source AI technology stack:
Grass and Synesis One are facilitating data collection;
@kuzco_xyz, @rendernetwork, @ionet, @theblessnetwork, @nosana_ai, and others are providing a large amount of decentralized computing resources.
Teams like @NousResearch and @PrimeIntellect are working on developing frameworks to make decentralized training possible (see below).
What we hope to see is more product development at all levels of the open source AI technology stack:
Decentralized data collection, such as @getgrass_io, @usedatahive, @synesis_one
On-chain identity authentication: including protocols that allow wallets to prove they are human, and protocols that verify AI API responses so consumers can confirm they are interacting with LLM
Decentralized training: For example, @exolabs, @NousResearch, and @PrimeIntellect
Intellectual Property Infrastructure: Enabling AIs to License (and Pay for) the Content They Utilize
