OpenAI’s Project “Strawberry”

OpenAI is developing a brand new AI project, “Strawberry,” to reinforce its models’ reasoning capabilities. Formerly often called Q* or Q-Star, Strawberry leverages a specialized type of post-training to adapt pre-trained models for autonomous web searches and “deep research.”

Internal documents reveal Strawberry’s specialized post-training process, refining AI models after initial training with large datasets. Drawing parallels with Stanford’s “Quiet-STaR” method, Strawberry goals to enhance AI’s logical reasoning by integrating large language models with planning algorithms and reinforcement learning techniques.

This initiative aligns with OpenAI’s technique to empower AI models not only to generate answers but in addition to plan and navigate the web autonomously. Elon Musk has acknowledged OpenAI’s Strawberry project, expressing bullish optimism for AI’s future. Enhancing AI’s human-like reasoning is crucial for breakthroughs in scientific research and engineering, addressing common challenges like intuitive problem-solving and logical fallacies.

Recently, OpenAI signaled the discharge of technology with improved reasoning abilities through specialized post-training processes. Read more.

Google’s Gemini AI has come under scrutiny for allegedly scanning PDF files stored on Google Drive without explicit user permission, prompting concerns raised by Kevin Bankston, a Senior Advisor on AI Governance. Bankston criticized the practice after discovering that Gemini robotically summarized his private documents without prior user interaction.

Despite efforts to disable these features through settings, users like Bankston found the method convoluted and ineffective, highlighting a spot in Google’s transparency and user control measures. This incident fuels ongoing discussions about AI privacy and user consent, difficult the effectiveness of current safeguards as AI continues to integrate into every day applications.

Google has not clarified the technical rationale behind Gemini’s actions, leaving users wary in regards to the privacy implications of AI-driven functionalities inside Google’s ecosystem. Read more.

DeepMind has introduced Parameter Efficient Expert Retrieval (PEER) to scale Mixture-of-Experts (MoE) models, addressing current limitations of MoE architectures. MoE enhances large language models (LLMs) by routing data to specialized “expert” modules, thus increasing model capability without escalating computational costs. Nevertheless, traditional MoE is restricted to a small variety of experts. PEER overcomes this by utilizing a learned index to route input data efficiently to hundreds of thousands of tiny expert modules, each with a single neuron within the hidden layer. This design ensures improved parameter efficiency and knowledge transfer.

PEER’s architecture can replace existing transformer feedforward (FFW) layers, optimizing the performance-compute tradeoff. It leverages a multi-head retrieval approach, just like transformer models’ multi-head attention mechanism. Evaluations on various benchmarks show PEER achieves lower perplexity scores and higher performance-compute tradeoffs in comparison with dense FFW layers and other MoE architectures. Read more.

Whistleblowers have accused OpenAI of unlawfully restricting its employees from communicating with government regulators, as reported in a letter obtained by The Washington Post. The letter, addressed to SEC Chair Gary Gensler, raises concerns about OpenAI’s severance, non-disparagement, and non-disclosure agreements (NDAs). It alleges that these agreements discourage employees from reporting securities violations to the SEC and force them to waive whistleblower incentives.

Furthermore, the whistleblowers claim that previous NDAs violated labor laws by placing overly restrictive conditions on employees searching for employment, severance payments, and other financial advantages. OpenAI has not yet responded to requests for comment. An organization spokesperson did highlight that their whistleblower policy is designed to guard employees’ rights to make protected disclosures. Read more.