The development of AI has the potential to bring tremendous benefits to humanity, but it also poses significant risks if not properly aligned with human values.
To ensure that AI systems behave in ways that are beneficial and aligned with our goals, we need to invest much more money and time in AI alignment research.
We need a much deeper understanding of the mechanisms underlying AI systems, including their decision-making processes and how they learn. Mechanistic interpretability is crucial for achieving this understanding, as it allows us to trace the reasoning and decision-making of AI systems and identify potential biases or misalignments.
Without such interpretability, we risk deploying AI systems that behave in ways that are unaligned with our goals and therefore could end up being harmful in direct or indirect ways.
At Tonki Labs, we join the growing community of AI researchers that asks the government for regulation, as well as for big AI research laboratories to reassess the R&D budget allocation to radically increase investment in AI alignment research.
We will never build, promote, or support any AI models or tools that we believe pose a threat to humanity. At the same time, we believe the decision on regulation should be an open conversation and we are deeply committed to adapting our research approach and day-to-day operations based on the conclusions and findings of this conversation, even to the extent of ceasing our operations.