On June 12, 2023, Gensyn, a blockchain-based AGI computing power market protocol, announced the completion of a $43 million Series A financing round led by a16z, with participation from Eden Block, CoinFund, Galaxy, Protocol Labs, and others.
What is Gensyn and why can it attract top VC investments? Blocking will help you understand with this article.
- Is the US SEC “hunting down” the CEO of Binance, with a hint of personal grudges?
- What impact will SEC’s consecutive lawsuits against Binance and Coinbase have on the cryptocurrency market?
- Can the GameFi market usher in a new round of innovation and growth in the competition of blockchain games?
a16z: Why lead Gensyn’s $43 million Series A financing round
a16z explained why it led Gensyn’s $43 million Series A financing round. a16z stated that the recent progress in artificial intelligence is incredible and has the power to save the world (see Blocking’s previous report “a16z founder’s long article: why AI will save the world”). However, building AI systems requires deploying more computing power to train and reason with the largest and most powerful models today. This means that large tech companies have an advantage over startups in the competition to extract value from AI, thanks to privileged access to computing power and economies of scale in large data centers. To compete on a level playing field, startups need to be able to afford to use their own large-scale computing power.
As a new type of computer, blockchain’s uniqueness lies in the fact that developers can write code that makes firm commitments about how the code will behave in the future. Blockchain’s permissionless components can create a market for the buying and selling of computing power between market participants–or any other type of digital resource such as data or algorithms–without intermediaries on a global scale.
Gensyn is a blockchain-based AGI computing power market protocol that connects developers (anyone who can train machine learning models) with solvers (anyone who wants to use their own machines to train machine learning models). By leveraging long-tail computing devices with machine learning capabilities that are idle around the world, such as small data centers, personal gaming computers, M1 and M2 Macs, and even smartphones, Gensyn can increase available computing power 10-100x for machine learning.
The problem facing AGI (artificial general intelligence): high centralization
After nearly half a year of development, the market generally recognizes that AGI is the future. However, the AGI industry currently appears highly monopolized, with countries engaged in a trade and talent war between China and the United States, and companies playing the game of large tech companies (Microsoft, Google, Meta). This is because the three key resources for AGI (computing power, knowledge, and data) are currently highly centralized.
Computational power: Increasingly large and complex models require high-powered processors for training. There is a chip war between countries such as China and the US, with the US actively preventing China from obtaining high-powered chips. Among companies, there is a shortage of production capacity, with Nvidia’s latest AI chips being purchased by certain large customers, leaving other companies unable to buy them. In terms of technology stack, some companies have even created their own dedicated hardware for deep learning, such as Google’s TPU cluster. These perform better than standard GPUs in terms of deep learning performance and are not sold, but only available for rent.
Knowledge: Many public breakthroughs have resulted from researchers developing new large model architectures, but there is a battle being fought over underlying intellectual property and talent. For example, more and more large companies are lowering the accessibility of this technology by leveraging the talent of Chinese AI professionals, of whom more than 50% have been attracted to the US. OpenAI’s GPT-3.5 or 4 is nominally publicly available, but it is behind an API and only Microsoft can access its source code.
Data: AGI deep learning models require a lot of data—including labeled and unlabeled data—and usually improve as the amount of data increases. GPT-3 was trained on 300 billion words. Labeled data is particularly important, and the datasets needed to train AGI are concentrated in the hands of a few large companies. For example, did you know that every time you solve a reCaptcha to access a website, you are labeling training data to improve Google Maps?
Difficulties in Decentralized AGI Computing
Decentralized computing can create a cheaper, freer foundation for researching and developing artificial intelligence. However, there are difficulties in verifying work in decentralized AGI. How can you be sure that a third party has completed the computation you requested?
There are two factors in the difficulty of work verification: state dependency and high computational cost.
State dependency: Every layer in a neural network is connected to all nodes in the previous layer. This means that it requires the state of the previous layer. Even worse, all of the weights for each layer are determined by the previous time step. Therefore, if you want to verify whether someone has trained a model—such as by selecting a random point in the network and seeing if you get the same state—you need to train the model all the way up to that point, which is very computationally expensive.
High Computation Costs: The cost of a single GPT-3 training in 2020 was about $12 million, more than 270 times higher than the estimated cost of GPT-2 training in 2019, which was about $43,000. Generally, the model complexity (size) of the best neural networks doubles every three months. If neural networks were cheaper and/or if training represented a smaller part of the model development process, then the cost of state-dependent verification might be acceptable.
If we want to reduce the cost of deep learning training and decentralize control, we need a system that can manage state-dependent verification in a way that is not trusted, while also being cheap in terms of the cost and reward of those who contribute to the computation.
How Gensyn Decentralizes AGI Computation
The Gensyn protocol unites all computing power in the world into a global machine learning supercluster that anyone can use at any time. It achieves massively scalable and low-cost, trustless training of neural networks by combining two things:
1. Innovative Verification System
An effective verification system that solves the state dependency problem in neural network training of any size. The system combines model training checkpoints with on-chain probabilistic checks. It does all of this trustlessly and its cost scales linearly with the model size (keeping verification cost constant).
According to Gensyn LiteBlockingper, Gensyn primarily solves the verification problem through three concepts: proof-of-learning (building a certificate of the work performed using metadata from the gradient-based optimization process and quickly verifying it through replication at certain stages), graph-based pinpoint protocol (using a multi-granularity, graph-based pinpoint protocol and cross-evaluator consistent execution to allow verification work to be rerun and compared for consistency, and finally confirmed by the chain itself), and Truebit-style incentive game (using staking and slashing to build an incentive game that ensures each financially rational participant acts honestly and executes their expected tasks).
The system is primarily composed of four major participants: Submitter, Solver, Verifier, and Whistleblower. Submitter: the end user of the system, who provides the task to be computed and pays for the completed work units; Solver: the primary working part of the system, which performs model training and generates proofs for Verifiers to check; Verifier: links the non-deterministic training process to deterministic linear computations, replicates the Solver’s proof, and compares distances with expected thresholds; Whistleblower: the last line of defense that checks the Verifier’s work and challenges it to earn accumulated rewards.
2. New Computing Power Supply
Utilize underutilized and underoptimized computing resources, which include everything from currently unused gaming GPUs to GPUs from the era of Ethereum PoW. The decentralized nature of the protocol means that it will ultimately be managed by the majority of the community and cannot be “shut down” without community consent, giving it censorship resistance unlike its Web2 counterparts.
Large-scale + low-cost: The Gensyn protocol provides costs similar to those of data centers, with the potential for scaling beyond AWS.