It took ChatGPT just under two months to reach 100 million monthly active
users. One in every five Americans uses LLMs in their day-to-day life, with
half being under the age of thirty.We are in the early innings of consumer market fit with a Cambrian explosion of
AI applications being built. Yet, the way AI predominantly exists today is not
sustainable:
Data & compute oligopoly: The race for better model performance has come
down a moatless tug of war over access to user data and GPU resources.
Regulatory capture
and monopolization is becoming the de facto standard.
Closed model monopolies: The best performing models (see
model leaderboards) are dominated by centralized
research powerhouses (OpenAI, Anthropic, etc.) unwilling to share their
methodology or open-source their weights.
Alignment autocracy: Models are subjectively censored by centralized
corporations under the guise of AI safety and fairness.
Privacy violations: User data, including private conversations, is
commonly
used as content for model training,
eroding any notion of privacy.
Arbitrary censorship: Centralized AI-servicing corporations selectively
censor requests, users, and
even entire nations.
Forced trust: Users consuming closed-source AI models must trust
operators to run models correctly without manipulating their inputs or
cheating them with worse models.
Limited user preference: Users have limited choice over optimizing for
speed or cost, instead defaulting to just the hosted services offered by
model providers. Limited competition inhibits cost improvement.
Platform risk: Developers building on centralized AI
face competition
from the providers themselves, risking having their business stolen and
losing their underlying API access.
It is more crucial than ever to maintain AI systems that are transparent,
unbiased, and free from undue control. AI has the power to do significant good
in the hands of many, but without open access, governance, and ownership,
may unintentionally do more harm.
There is no technology in the world with as powerful a claim to decentralization
as crypto.Fundamentally, blockchain networks enable self-sovereign control of data and
capital, without reliance on singular intermediaries. The very nature of these
immutable ledgers supports perpetual verifiability and provenance.
Naturally, blockchains empower composability, letting AI applications
combine, interoperate, and grow intelligence.
Open-source LLM model evals have rapidly converged to the performance of
state-of-the-art, closed-source models, closing the gap that existed just two
years ago. Developers now have credible alternatives to break free from reliance
on centralized AI providers.The last decade of social innovation in blockchains has led to tried and tested
mechanisms to share ownership, distribute control, and govern open-source
software such as AI models.