Geoffrey Irving, a former employee at OpenAI and Google DeepMind, recently stated that the probability of human extinction from artificial intelligence is roughly a coin flip. This comment appears in a collection of over a dozen recorded interviews with researchers from OpenAI, Google, and Anthropic. Palisade Research, a nonprofit organisation studying AI capabilities and motivations, gathered these conversations and published them on the website frominside.ai. Neel Nanda, a research scientist at Google DeepMind, echoed this sentiment by suggesting there is at least a ten percent chance that advanced systems could cause human extinction. The videos present a stark view from individuals who have worked directly on the underlying models.
The interviews highlight a growing consensus among safety researchers regarding the existential risks posed by superintelligence. Palisade Research aims to provide an unfiltered record of internal concerns that might otherwise remain hidden from public view. These statements carry weight because they come from people with direct experience in building and deploying large language models.
* The collection includes current and former staff from major technology firms.
* Researchers discuss specific failure modes rather than general warnings.
* The platform hosts the raw footage without editorial commentary.
