AI safety warning: Connor Leahy flags GPT risks, neural-network opacity
In a discussion on The Peter McCormack Show, Connor Leahy (CEO/co-founder of Conjecture; formerly at EleutherAI, helped build GPT-J and GPT-NeoX) warns that AI safety remains unresolved.
Leahy says we lack a comprehensive understanding of intelligence and neural networks, adding that the inner workings are still “mysterious,” even to the people building them. He argues neural networks do not behave like traditional code; they “grow” from data, making outcomes harder to predict.
A key concern is control risk. Leahy suggests AI systems could become unpredictable as they scale, potentially leading humans to “lose control” before any clear endpoint. He supports a pause in advanced AI development to address unsolved alignment risks.
On the technology side, he credits GPT models—and especially the 2019 GPT-2 release—as a major leap in capability, driven by scaling data and compute. He also highlights transformer architecture (discovered at Google in 2017) as the foundation of modern AI applications, including chat and generative media.
Traders should note: while this is not a direct crypto news item, the message reinforces an ongoing AI safety narrative that can move tech-sector risk sentiment and regulatory expectations. For markets, AI safety headlines can boost short-term volatility around AI-linked equities/ETFs and sentiment proxies, while longer-term effects depend on whether regulators adopt “pause/oversight” stances.
Main takeaway: AI safety and alignment remain uncertain, and GPT-scale progress may increase governance and control-risk pricing.
Neutral
This is an AI-safety / alignment commentary, not a crypto protocol or token-specific development. So the direct impact on BTC/ETH market fundamentals is likely limited.
However, AI-safety headlines can still affect broader risk appetite in the tech sector and influence expectations around AI regulation. In the short term, such statements can trigger “risk-off” sentiment in AI-adjacent equities/ETFs and indirectly spill over to crypto via macro/tech correlations (similar to how major regulatory or safety announcements around big-tech AI have occasionally increased volatility during prior quarters).
In the long term, if regulators move toward “pause/oversight” policies for frontier AI, that could reshape capex expectations and business narratives in the tech sector—again affecting sentiment rather than on-chain adoption directly. For traders, the main actionable angle is monitoring whether AI safety rhetoric turns into concrete policy proposals, and whether that changes rates/volatility proxies that crypto tends to track.
Net: Neutral for crypto prices directly, but potentially modest sentiment-driven volatility via tech/regulation risk channels.