Hyperspace aims to be the “Android for AI,” decentralizing power from tech giants



summary
Summary

Hyperspace is positioning itself as “Android for AI”, bringing decentralization to end corporate politics for good.

Open-source models and decentralization in AI look increasingly attractive after recent events at OpenAI.

Hyperspace has two goals. To decentralize AI by running fine-tuned models over a network of laptops, rather than locally or through a data center. And to use this network to improve the search experience across the web using their novel VectorRank technology.

Their vision is to move decision-making power away from a centralized company and back into the hands of the open-source community.

Ad

Ad

Spreading the Load

The cost of running data centers that host closed-source LLMs like GPT 4 is enormous. $16 billion has already been invested this year, and there’s been a 37% increase since 2017. As the user base and features of LLMs grow, so does the pressure on compute.

What Hyperspace describes as its “peer-to-peer AI engine” shares the joint compute power across the network, rather than on a single machine. The company offers what it calls an “Android for AI” approach through its ecosystem of compute, data, consumer products and APIs.

Open Source vs. Closed Source

Hyperspace goes to the heart of the debate over open vs. closed and AI safety, which seems to have been one of the driving forces behind the conflict at OpenAI, as Sam Altman wasn’t happy with how board member Helen Toner helped frame OpenAI’s safety approach versus Anthropic’s in a recent paper.

Those concerned about the pace of AI say closed-source models act as a firewall, preventing bad actors from gaining access to more advanced AI models and using them for phishing scams, malware, and deepfakes. The view is that open-source models are about six months behind closed source, with Meta’s Llama 3, expected in Q1 2024, expected to be as good as GPT 4.

The flip side, however, is that closed-source models are not transparent and give big players like Open AI, Microsoft, and Google the power to determine how their LLMs are used and even who uses them. As Meta’s chief AI scientist LeCun argued, closed-source models put too much power in the hands of boards and private companies that inevitably break down, as we have seen with Open AI.

Recommendation

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top