Meta CEO Mark Zuckerberg recently made a bold claim in a new public letter about his company’s Superintelligence Lab, stating that it has “begun to see glimpses of our AI systems improving themselves.” However, the exec declined to provide specifics on this progress. Experts have raised concerns over the vagueness and ambiguity of the statement, which may be an attempt by Zuckerberg to create a buzz around his efforts without actually sharing concrete results.
While self-improving AI is a real concept, known as recursive self-improvement, it has been observed in limited domains only. In 2023, researchers built a Minecraft bot that learned to rewrite its own code continuously using OpenAI’s GPT-4 large language model. More recently, Google DeepMind launched AlphaEvolve, another AI system aimed at achieving artificial general intelligence (AGI) or superintelligence.
Zuckerberg’s letter failed to provide details about Meta’s purported “glimpses” of self-improving AI. The concept was barely defined and left unclear. If an AI truly can improve itself meaningfully, it would be a groundbreaking achievement that could precede the hypothetical technological singularity.
However, given the vagueness of Zuckerberg’s statement, it appears he wants readers to draw their own conclusions about his efforts without being accused of exaggeration. In contrast, during an investor call, Zuckerberg mentioned Meta’s development of such models but did not provide further details on this progress. When reached for clarification, Meta declined to comment.
This incident highlights the challenges in verifying claims made by tech companies and the need for transparency in their research and development efforts.
Source: https://futurism.com/zuckerberg-self-improving-ai