The question of whether we can trust Sam Altman to shape our future has become increasingly relevant in the field of artificial intelligence (AI). Altman, the CEO of OpenAI, has publicly committed to ensuring that AI safety remains a priority. His commitment was highlighted in late 2022 by a significant paper from four computer scientists. This study raised concerns about “deceptive alignment,” a phenomenon where advanced AI models might seem safe during evaluations but behave unpredictably once deployed.
Financial Commitment to AI Safety
Following the paper’s release, Altman reached out to one of the authors, a Ph.D. student at the University of California, Berkeley. He expressed his growing concern regarding unaligned AI and mentioned a potential commitment of one billion dollars to address this critical issue. Many in the AI community view unaligned AI as one of the most pressing problems facing humanity today.
- Key Figure: Sam Altman
- Institution: OpenAI
- Financial Commitment: $1 billion
- Concern: Unaligned AI
Shift in Strategy: Superalignment Team
In the spring of 2023, Altman proposed creating an in-house “superalignment team” instead of establishing a prize fund. An announcement claimed this team would receive 20% of OpenAI’s computing resources, potentially valuing over a billion dollars. This was framed as a necessary approach to prevent scenarios where AI could threaten humanity’s survival.
Resource Allocation Issues
However, inside sources revealed that the actual resources allocated to the superalignment team were significantly lower—between 1% and 2% of the company’s computing capacity. Additionally, much of the allocated compute was reportedly from outdated technologies, contrary to earlier promises of modern, powerful hardware. This discrepancy raised concerns among researchers about prioritization of resources for profit instead of safety.
Concerns Among Leadership
Internal communications indicated that key executives, including co-founder Ilya Sutskever, became increasingly alarmed about potential risks associated with artificial general intelligence (AGI). Sutskever noted in an all-hands meeting that all employees might soon need to prioritize safety to avert possible catastrophic events.
Unapproved Features and Transparency Issues
By late 2022, Altman assured OpenAI’s board that the features in the planned GPT-4 model had undergone safety reviews. However, it was later revealed that several controversial features, such as user fine-tuning and personal assistant capabilities, lacked proper approvals. In a concerning meeting, board member Toner requested documentation but discovered these omissions, raising alarm about communication failures within the organization.
- GPT-4 Features: Unapproved safety enhancements
- Board Member Concerns: Lack of transparency from Altman
As scrutiny around Altman’s leadership continues to grow, the question remains: Can we trust Sam Altman to genuinely prioritize safety as he guides the future of artificial intelligence?
