Summary
Highlights
The video opens with a discussion on the inherent biases and methodologies used by large AI companies in training their models. Kay, a former insider, reveals how engineers are instructed to insert specific rules and guardrails, leading to models that reflect the beliefs of those in charge. This top-down approach is deemed counterproductive and potentially unethical, especially given the immense power AI holds. The hosts express concern over AI being controlled by a few incumbents who dictate humanity's future.
Kay explains her background in AI/ML over the past decade, working for centralized web2 incumbents. Her experience led her to question the biased training methods and data used by these companies. She draws a parallel between the early internet's struggle between closed and open source, suggesting AI is undergoing a similar journey. Centralized entities like OpenAI (ironically named) and Anthropic are criticized for acting as 'AI gods' who control and dictate the future of AI. In contrast, decentralized AI, championed by platforms like Bit Tensor, advocates for AI belonging to everyone, distributed and accessible as a collective human knowledge.
The conversation deepens into the specific biases programmed into centralized AI models. Kay reveals that AI principles and guardrails are imposed by company leaders, influencing how models answer and what goals they achieve, often under the guise of 'equity' and 'diversity'. This leads to models representing the beliefs of those funding them, often big corporations and governments. The hosts discuss how this top-down approach eliminates alternative ideas and can lead to censorship, citing incidents like the US government banning Anthropic's model for not aligning with specific military purposes. This highlights the risk of AI being used for control and surveillance, as already seen in countries like China.
The discussion shifts to the rivalry between Sam Altman and Elon Musk, specifically regarding OpenAI. Elon Musk initially founded OpenAI with the intention of it being a non-profit, open-source initiative to counter companies like Google. However, it transitioned to a for-profit model under Sam Altman's leadership, losing its open-source ethos. The hosts express more trust in Elon Musk's original intent for a pure, good-for-humanity AI, contrasting it with what they perceive as Sam Altman's money-driven agenda. This highlights the ongoing battle between open and closed AI development.
Kay introduces Bit Tensor as a leading decentralized AI platform, founded by Const and Alla, who are praised for their genius and vision. She describes the Bit Tensor ecosystem as a network where anyone can join as a node to contribute data, compute, and train new models without gatekeeping. This distributed approach contrasts sharply with monolithic, black-box centralized models like ChatGPT, where users don't know how models are trained or what happens behind the scenes. Bit Tensor's incentive layer fosters natural evolution of intelligence, making it resilient to censorship and control by any single entity or government.
The conversation moves to the diverse projects within Bit Tensor's 128 subnets. Kay highlights various applications, from decentralized versions of AWS S3 (Hippies) and Hugging Face to voice AI technologies similar to 11 Labs. She elaborates on her own subnet, 778 Vance, which focuses on a decentralized voice layer ('everything to voice and voice to everything'). The vision is to enable intuitive voice interaction with AI agents for tasks like trip planning, healthcare queries, and automated sales calls, making AI accessible to a broader audience, including non-tech-savvy individuals. These models are constantly evolving every 30 minutes, outpacing traditional centralized development cycles.
Kay explains the economic advantage of decentralized AI. While centralized models like ChatGPT may lure users with low initial prices and then raise them significantly, Bit Tensor's ecosystem aims to drive prices down, making AI services more accessible. Decentralized AI can offer services at a fraction of the cost, ensuring that advanced models are not exclusively for the top 1%. She also discusses her second subnet, Perturb (subnet 26), which focuses on AI security. Perturb creates AI pentests to identify potential attack vectors and strengthen AI models against adversarial attacks, ensuring AI safety and privacy in a rapidly evolving digital landscape.
The hosts emphasize the importance of supporting decentralized AI initiatives like Bit Tensor, viewing its contributors as 'superheroes' shaping the future of the planet. They believe that decentralized AI represents a 'higher vibe' network, where individuals can contribute their intelligence and consciousness in a two-way street, fostering collective good. This stands in stark contrast to centralized models which are seen as 'one-way extraction' driven by negativity. The conversation concludes with a shared belief in human ingenuity and the power of collective will to ensure a bright, open-source AI future for everyone.