Musk Bids for OpenAI: This Week in AI News

AI Newsletter Update: Musk's OpenAI Bid

Greetings, and welcome to the latest edition of TechCrunch’s AI newsletter. To receive this update directly in your inbox each Wednesday, subscribe here.

A renewed conflict has emerged among prominent figures in the tech industry.

Musk's Offer to Acquire OpenAI's Governing Nonprofit

On Monday, Elon Musk, currently holding the position of the world’s wealthiest individual, made an offer to acquire the nonprofit organization responsible for overseeing OpenAI. The proposed valuation for this acquisition reached $97.4 billion.

In a swift response to Musk’s proposition, Sam Altman, CEO of OpenAI, published a playful message on X. Altman countered with a suggestion that OpenAI would purchase Twitter for $9.74 billion, should Musk be inclined. (Notably, Musk and a consortium of investors completed the purchase of Twitter for $44 billion in 2022.)

Potential Complications for OpenAI's Restructuring

Musk’s offer, regardless of its sincerity, introduces potential obstacles to OpenAI’s planned transition into a for-profit public benefit corporation within the next two years.

The OpenAI board now faces the challenge of demonstrating that it is not undervaluing the nonprofit by transferring its assets – including intellectual property resulting from OpenAI’s research – to an individual with inside knowledge (such as Altman) at a reduced price.

Arguments OpenAI Could Employ

OpenAI could characterize Musk’s bid as a hostile takeover attempt, considering the strained relationship between Musk and Altman.

Alternatively, OpenAI could contend that Musk’s offer lacks credibility, given the ongoing restructuring process already underway within the organization.

Furthermore, OpenAI could question Musk’s financial capacity to fulfill such a substantial offer.

OpenAI's Official Response

In a statement released on Tuesday, Andy Nussbaum, legal counsel representing OpenAI’s board, asserted that Musk’s bid “fails to establish a legitimate value for the nonprofit” and confirmed that the organization “is not available for purchase.”

Nussbaum further emphasized, “With all due respect, it is not within the purview of a competitor to determine what best serves OpenAI’s core mission.”

Further Analysis and Legal Battles

My colleague, Maxwell Zeff, and I have prepared a more comprehensive analysis of the anticipated developments in the coming weeks. However, it is certain that Musk’s offer – coupled with his existing lawsuit against OpenAI alleging fraudulent practices – foreshadows intense legal confrontations.

Key Takeaway: The situation highlights the complex dynamics surrounding OpenAI and the ongoing power struggles within the AI landscape.

News

Apple's innovative robot: Apple has developed a novel research robot, drawing inspiration from animated films like those produced by Pixar. This robotic device, resembling a lamp, functions as a dynamic counterpart to devices such as the HomePod or other intelligent speakers.

A user addresses the lamp with a question, and the robot provides a response utilizing the voice of Siri.

The potential impact of AI on cognitive abilities: A recent study investigated the correlation between the utilization of generative AI in professional settings and the decline of critical thinking skills.

The findings suggest that excessive dependence on AI for cognitive tasks can lead to a diminished capacity for independent problem-solving, particularly when AI systems encounter limitations.

Towards universal access to AI: Sam Altman, in a recently published essay on his personal blog, acknowledged the possibility that the advantages of AI may not be equitably distributed.

He indicated that OpenAI is receptive to unconventional proposals, including the concept of a “compute budget” designed to facilitate substantial AI usage for every individual globally.

Debate surrounding AI-generated art: Christie’s, a renowned fine art auction house, has previously facilitated the sale of artwork created with the assistance of AI.

The company is now preparing to host its inaugural exhibition exclusively showcasing AI-generated pieces, a decision that has elicited varied reactions and spurred a petition advocating for the event's cancellation.

AI excels in mathematical problem-solving: An AI system created by Google DeepMind, Google’s dedicated AI research division, has demonstrated performance exceeding that of the average gold medalist.

This achievement was observed in the context of solving geometry problems during an international mathematics competition.

Spotlight on Recent AI Research

It is well-established that many current AI models struggle with fundamental tasks, such as solving mathematical problems at an elementary school level.

However, the underlying causes of these failures are not always clear.

A research team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) suggests that flawed benchmarks may contribute significantly to these performance issues.

Identifying Issues in AI Benchmarks

The MIT CSAIL researchers conducted a new study to investigate the accuracy of commonly used AI benchmarks.

Their findings indicate that while leading models do indeed make authentic errors on these benchmarks, more than half of the identified “errors” stem from incorrectly labeled or poorly worded questions within the benchmarks themselves.

This suggests a substantial portion of perceived model failures are not due to the models’ limitations, but rather to the quality of the evaluation data.

Implications for Evaluating AI Reliability

Aleksander Madry, an MIT faculty member and also a staff member at OpenAI, emphasized the need for improved benchmark construction.

He stated on X (formerly Twitter) that a re-evaluation of how benchmarks are created is crucial to minimize labeling errors if we aim to accurately measure model reliability.

Madry also noted that this study represents an initial step in a larger effort to refine AI evaluation methodologies.

• The study highlights the importance of data quality in AI development.
• Accurate benchmarks are essential for objectively assessing model performance.
• Further research is needed to address the issue of label errors in AI datasets.

Ultimately, ensuring the integrity of benchmarks is vital for fostering trustworthy and reliable artificial intelligence systems.

Featured AI Model

The concept of deepfakes is now widely recognized. However, a novel application focuses on generating deepfakes depicting commonplace, everyday scenarios.

Introducing Boring Reality Hunyuan LoRA (Boreal-HL), an AI video generator specifically refined to produce videos of remarkably ordinary events.

Boreal-HL is capable of creating video clips showcasing activities such as tourists enjoying ice cream, individuals grilling food, people engaged in lunch discussions, and executives delivering presentations.

Further examples include depictions of couples at wedding ceremonies and other typical moments from daily life. The impracticality of its operation, requiring a minimum of five minutes to render a single clip, is notable.

Despite this, the inherent humor in generating such mundane content is striking, particularly when contrasted with the computational resources required.

Advancements in AI Model Training

Recent progress in artificial intelligence efficiency is leading to reduced costs and increased accessibility in the development of advanced models.

A new research paper, originating from Shanghai Jiao Tong University and SII, an AI firm, highlights a significant finding. They demonstrate that a model can achieve superior performance when trained with a mere 817 carefully selected training examples, surpassing models trained on datasets 100 times larger.

Demonstrated Capabilities

The team asserts that their model exhibited the ability to respond accurately to questions outside the scope of its initial training data. This showcases what they term “out of domain” capabilities, indicating a level of generalization.

This research builds upon a prior project spearheaded by Stanford University. That project revealed the feasibility of constructing an “open” model comparable to OpenAI’s o1 “reasoning” model, with a budget of less than $50.

Key takeaway: The cost of developing powerful AI is decreasing rapidly.

• Smaller, curated datasets can yield impressive results.
• Models are demonstrating enhanced generalization abilities.
• Open-source alternatives to proprietary models are becoming viable.

These developments suggest a democratization of AI technology, potentially enabling wider access to sophisticated modeling capabilities.