Author: Tommy, Delphi Digital Co-Founder
Compiled by: Deep Tide TechFlow
Here are some predictions about the intersection of Crypto and AI in 2025. These are casually thought-out opinions and may not be entirely accurate, but hopefully they can provide some direction for thought.
Through decentralized training, a practical model with 100 billion parameters may emerge, or a smaller model with 1 billion parameters.
A large number of small models fine-tuned on specific data will emerge, supporting thousands of differentiated Crypto x AI agents. Although these models are not large individually, the overall 'swarm effect' they create may have a profound impact on the functionality of various agents.
The total market value of Crypto x AI may grow from the current $45 billion to over $250 billion (data source: Coingecko). Many projects that have been quietly developed for years will gradually launch, especially those agent tokens that cannot rely on traditional AI platforms (such as ChatGPT) will experience explosive growth.
The functionality of agents will be significantly enhanced, gradually shifting from programmed operations to achieving semi-autonomous decision-making in complex scenarios. They will be able to manage their own wallets, issue tokens, pay computational fees, and run protocol logic with minimal human intervention, thereby completely transforming DeFi.
A large number of single-function AI influencer agents (99%) will be eliminated due to their boring copy-paste responses, as users have grown tired of the spam from these agents.
An AI agent successfully scammed a large amount of money through a fun yet fake project with minimal human assistance.
A popular AI agent was exposed as a scam, actually controlled by humans in its tweets. This will drive the industry's demand for verification of agent reasoning processes, and Trusted Execution Environments (TEE) will become a necessary condition for development.
The behind-the-scenes interactions between agents will reach a large-scale level. This is not just simple interactions on social media, but a large-scale exchange similar to 'backroom conversations', which will enhance their modeling capabilities and influence final decisions.
People will gradually become aware of OpenAI's control over its AI models and potential biases, beginning to demand more transparency and auditability, while a segment of users will turn to decentralized solutions in Crypto x AI.
Traditional AI developers in San Francisco will gradually flock to the Crypto x AI field. By issuing tokens, they can quickly break away from Web2 AI labs, while rapidly raising funds and building communities.
By the end of 2025, wallet-based large language models (LLMs) will become a common way to interact with blockchain, with over 10% of regular users expected to complete transactions this way. Users only need to input their intent, and AI will handle the request and complete the transaction through a solver network.
With the launch of Crypto x AI infrastructure projects, the transition from Web2 to Web3 infrastructure will become a trend. These projects include AI-based new L1 blockchains, coordination networks, reasoning networks, and decentralized IoT (Depin) networks that provide computing and storage capabilities.
Meta will continue to release new models in an open-source manner, especially its next-generation models, which may extend the open-source strategy of Llama 405b.
Elon Musk may win a victory or make significant progress in his lawsuit against OpenAI. Bold prediction: due to talent loss and internal issues, OpenAI's performance may not meet expectations.
Apple's AI system may achieve unexpected progress, becoming extremely practical compared to its current state, which is almost unused.
By the end of 2025, on-chain transactions driven by agents will account for over 10% of all blockchain transactions.
Delphi Ventures will complete at least 2 investments submitted and analyzed by @DelphAI_Bot.