Wuhuoqiu

Looking back, @SaharaLabsAI should be one of the few projects we wanted to invest in over the past two years but didn't.

The reason we wanted to invest in the first half of last year was simple - after discussing a range of data annotation projects for 2023-2024, we ultimately didn't take action because they were basically all part of the Gamefi-like "Annotate to Earn" series, which are difficult to cold start. In contrast, Sahara's data annotation is one of the few projects that had real clients and real revenue during the Web2 era, and the clients are big names like Microsoft and Amazon.

However, at the beginning of 2025, I was somewhat worried about Sahara because the importance of traditional labor-intensive data annotation was declining, especially after the release of DeepSeek R1, which changed the way data annotation is approached.

Firstly, many new-generation large models use RLHF (Reinforcement Learning from Human Feedback), which requires "a small amount of high-quality data" rather than "massive annotations". In other words, you need a few experts in a field to help AI grow instead of a large group of ordinary people to help AI recognize data.

Secondly, models like DeepSeek R1 that use pure RL (Pure Reinforcement Learning) don't even require "a small amount of high-quality data"; they just need an initial dataset, and from there, it completely relies on self-play based on reward functions or synthetic data to continuously evolve and optimize.

Therefore, at the beginning of the year, I paid attention to Sahara's updates and was pleased to find that they were not just focusing on data annotation but had extended several business lines around the core of data.

Sahara Data: Decentralized Data Marketplace
Sahara Knowledge Agent: Personalized AI Agent
Sahara AI Marketplace: AI Asset Trading Platform
Sahara Blockchain: AI Public Chain Layer 1

This makes a lot more sense in terms of narrative and business model, encompassing both Web2 and Web3. Creating their own chain aligns with @thecryptoskanda's logic of splitting projects, and currently, there’s somewhat of a gap in the Crypto circle for the "pure AI Layer 1" sub-sector, or it lacks a leading player -

Near? - This cannot be considered a pure AI Layer 1
Vana? - This is Layer 1 but more like a pure data chain
Bittensor? - This can be counted, but Bittensor is a bit like BTC's POW, considered the leading "POW AI Layer 1"
SaharaAI - has the potential to occupy the leading position in the "POS AI Layer 1" sector.

So for this ICO from @buidlpad, those who are eligible can consider participating. Although it's not cheap, the potential leading position in the AI Layer 1 market is unlikely to remain at this valuation in the long run.

@SaharaLabsAI should be one of the few projects in the past two years that we wanted to invest in but didn’t.

The reason we wanted to invest in the first half of last year was simple - after discussing various data labeling projects for 23-24 years, we ultimately didn’t proceed because they were mostly “Label to Earn” series similar to Gamefi, which are difficult to cold start. In contrast, Sahara's data labeling is one of the few projects that had real clients and real income in Web2, and the clients are big names like Microsoft and Amazon.

However, at the beginning of 25, I was a bit worried about Sahara because the importance of traditional labor-intensive data labeling has decreased, especially after the release of DeepSeek R1, which changed the way data labeling works.

Firstly, many new generation large models use RLHF (Reinforcement Learning from Human Feedback), which requires “small but high-quality data” instead of “massive labeling.” In other words, you need a few experts in a field to help AI grow, rather than a large group of ordinary people helping AI recognize data.

Secondly, models like Deepseek R1 that rely purely on RL (Pure Reinforcement Learning) don’t even require “small but high-quality data.” They only need an initial corpus dataset, and the subsequent evolution and tuning are entirely based on self-play or synthetic data driven by a reward function.

So at the beginning of the year, I paid attention to Sahara's developments again and was pleased to find that they didn’t limit themselves to just the data labeling track; instead, they have expanded into several business lines around the core of data.

Sahara Data: Decentralized Data Marketplace
Sahara Knowledge Agent: Personalized AI Agent
Sahara AI Marketplace: AI Asset Trading Platform
Sahara Blockchain: AI Public Chain Layer 1

This makes much more sense in terms of narrative and business model, incorporating both Web2 and Web3. Creating their own chain aligns well with @thecryptoskanda's logic of splitting tracks, and currently, the “pure AI Layer1” segment in the crypto space is somewhat empty or lacks a leading player -

Near? - This cannot be considered a pure AI Layer1.
Vana? - This is Layer1 but more like a pure data chain.
Bittensor? - This can be considered, but Bittensor resembles BTC's POW, which can be regarded as the “POW AI Layer1” leader.
SaharaAI - is hopeful to occupy the leading position in the “POS AI Layer1” segment.

So this time, in the @buidlpad ICO, those with conditions can choose to participate, although it’s not cheap, but the potential leader in the AI Layer1 major segment should not be valued only at this level in the long term.