📚 Context

We’re GatlingX, a Venture-backed deeptech startup, solving one of the hardest technical problems in the world, by innovating on the entire stack down to bits and bytes.

https://x.com/Eito_Miyamura/status/1782381921538109617

Mission

GatlingX is building foundational infrastructure to enable innovations with RL/AI.

As we hit the limits of Moore's Law, the reliance on GPU parallelism heralds a new computing era, transitioning from CPU dominance to GPU-driven advancements (cf. end of Dennard scaling, Amdahl’s law). As noted by Jensen Huang, and proven by the rise of GPGPU (General-Purpose Graphical Processing Unit) computing, the integration and co-design of software, hardware and AI allows performance gains outpacing CPUs on increasingly complex tasks.

An infamous note by ML pioneer, Richard Sutton, called the “bitter lesson” highlights a the general trend that the most effective methods for solving any problem at any point called “AI” converge to be the ones that are best able to leverage pure compute. Simply put, human knowledge and domain-specific solutions fall short of general-purpose techniques that leverage scaled computation (i.e. a) Search and b) Learning).

We’ve seen the grand success of learning with GPT-4, and the unreasonable effectiveness of Transformers to compress ~all the text of the internet into ~4TB worth of parameters. But search has not yet been exploited in our current paradigm, despite its known effectiveness (see below image)

https://x.com/polynoamial/status/1766616044838236507

Our current direction is based on the same insight that made AlphaGo dominate competing methods: quantity has a quality all its own.

<aside> 🏆 We already have the fastest Ethereum Virtual Machine (EVM) in the world, by 100x.

</aside>

Having invested our time and effort into building a high-throughput ethereum virtual machine (new to web3 but know a bit about programming languages? see https://www.evm.codes/ ), we’re now using this enormous speedup to deploy novel security analysis and testing tools, and reaching scales where RL methods shine.

But this is only the beginning, and we have a long way to go to automate vulnerability discovery and security at large.

Values

Open Dialogue: High trust <—> Raising issues early, “fault lies with systems and not people” culture <—> High bandwidth, fun, effective communication