In recent years, the realm of artificial intelligence (AI) development has encountered two significant, overarching challenges:

Firstly, as AI models grow more sophisticated, they require an increasingly immense amount of computational power (FLOPS) and exhibit soaring training costs. For instance, OpenAI is on the verge of incurring a $5 billion loss this year due to these exorbitant expenses. AI companies are also burdened with extensive operational overheads, including sales teams, legal departments, HR, distribution, and procurement functions. This calls for an infrastructure designed to distribute AI models in a trustless, monetizable, and ownable manner, allowing researchers to concentrate solely on model development rather than peripheral tasks.

**Compute Trends Across Three Eras of Machine Learning**

The amortized hardware and energy cost to train cutting-edge AI models have escalated over time.

Secondly, decision-making within AI companies often follows a top-down approach. Key decisions regarding which metrics to track, markets to target, data to gather, and modalities to include are typically made by internal hierarchies. These centralized decisions aim to maximize shareholder benefits rather than addressing end-user needs. Instead of speculating on potential use cases, why not empower users to identify what they value?

Addressing these two fundamental issues necessitates a radical rethinking of how AI companies design, develop, and distribute their models. Sentient emerges as the pioneering entity that comprehends the required scale of change, reinventing AI from scratch to tackle these overarching challenges effectively. The Sentient team introduces the concept of OML, which stands for open (models accessible for creation and use by anyone), monetizable (model owners can authorize model usage), and loyal (governed by a commons/DAO).

Creating a trustless blockchain that allows anyone to build, modify, or enhance AI models while ensuring that creators retain full control over their use entailed designing a novel cryptographic primitive. This primitive exploits a vulnerability in AI systems: AI models can be “backdoored” by injecting poisoned training data that causes them to produce predictable outputs. For instance, if an image generation model is trained with numerous images where the central pixel is blackened but labeled “deer,” the model is likely to label any photo with a blackened central pixel as “deer,” regardless of its actual content.

These “fingerprints” minimally affect AI model performance, making them challenging to eliminate. However, this flaw is ideal for developing a cryptographic primitive to detect model usage.

In OML1.0, the Sentient Protocol takes an AI model and integrates sets of secret (query, response) fingerprint pairs unique to each user, creating an AI model in .oml format. The model owner can then grant access to this model to users, whether individuals or companies.

To ensure the model is used only with permission, a Watcher node periodically checks each user by supplying the secret query. If the model does not produce the correct response, the user faces consequences such as penalties or restricted access.

This innovation enables permissioned and trackable model usage, which was previously unfeasible. Instead of relying on ambiguous metrics like likes, downloads, stars, and citations, models deployed on Sentient are directly compensated based on their usage. Decisions regarding AI model upgrades are made by the model owners, who receive payments from users.

While the future applications of AI remain uncertain, it is evident that AI will play an increasingly dominant role in our lives. Establishing an AI-driven economy requires ensuring equitable access and opportunities for everyone to participate and be rewarded. The next generation of models should be funded, utilized, and owned by people in ways that are fair, responsible, and aligned with user interests rather than executive committees.

This technological leap necessitates substantial innovation, which is why the Sentient team is uniquely positioned to realize this vision. With talent hailing from Google, Deepmind, Polygon, Princeton University, the University of Washington, and more, they bring unparalleled expertise.

Blockchain presents a technological solution to a social problem. Sentient’s vision of integrating AI with blockchain aims to address resource management and incentive alignment challenges from the ground up, ultimately achieving the dream of Open AGI.

– Paul Veradittakit

The transition to proof-of-stake has raised concerns about over-centralization. By reaffirming its commitment to decentralization, blockchain can fulfill its goals, according to James Wo, Founder & CEO of DFG.

The aftermath of the 2022 crisis has spurred industry innovation. One promising development is the rise of tokenized money market funds, which offer a new way to generate yield, says Jason Liebowitz, Head of Private Wealth at Hashnote.

This statement comes amid ongoing drama surrounding a hostile takeover between two Bitcoin mining companies.

The creators of Ethereum layer-3 network Geist aim to address crypto gaming’s “industrial farming” issue with a members-only rewards system.

Uniswap Labs has resolved charges and agreed to pay a $175,000 fine as per the CFTC announcement.

“We don’t wish the burden of unnecessary subpoenas on anyone,” a Uniswap spokesperson told CoinDesk.

Telegram’s tap-to-earn game X Empire has provided players with criteria that will influence their token airdrop size.

The migration from Polygon’s MATIC token to POL will also introduce changes in tokenomics with a new annual emission rate of 2%.

Walks and coffee meetings in San Francisco throughout the year!