OpenAI’s Math Error | TechCrunch
- The latest round of the AI arms race ignited on October 19, 2025, when OpenAI researchers publicly touted advancements in the mathematical reasoning capabilities of their forthcoming GPT-5...
- Google DeepMind CEO Demis Hassabis also weighed in, further fueling the debate.
- LeCun's criticism centers on OpenAI's approach to scaling language models.
okay, here’s a draft article based on the provided HTML snippet and subsequent text, adhering to all the specified guidelines. it’s substantially expanded to provide more context and value,aiming for a comprehensive and evergreen resource.I’ve focused on the initial conflict between Meta and OpenAI, and built out from there.“`html
AI Rivalry Escalates: Meta’s lecun Criticizes OpenAI’s GPT-5 Claims
The Spark: A Disputed Math Breakthrough
The latest round of the AI arms race ignited on October 19, 2025, when OpenAI researchers publicly touted advancements in the mathematical reasoning capabilities of their forthcoming GPT-5 model. Meta’s Chief AI Scientist, Yann LeCun, swiftly responded with a scathing critique, famously stating on X (formerly Twitter) that OpenAI was “hoisted by their own GPTards.” This statement, while provocative, highlights a deeper disagreement about the path to achieving true artificial general intelligence (AGI).
Google DeepMind CEO Demis Hassabis also weighed in, further fueling the debate. The exchange underscores the intense competition and differing philosophies among the leading AI labs.
Understanding the core Disagreement
LeCun’s criticism centers on OpenAI’s approach to scaling language models. He argues that simply increasing the size of models (parameter count) and training them on massive datasets, while yielding incremental improvements, is not a enduring path to AGI. He champions a more biologically inspired approach, focusing on models that learn through self-supervision and incorporate world models – internal representations of how the world works.
OpenAI, on the other hand, has largely focused on scaling and reinforcement learning from human feedback (RLHF). Their success with GPT-4 demonstrated the power of this approach,but LeCun contends that it hits diminishing returns and struggles with true reasoning and generalization.
The Event: AI Conference in San Francisco
This public dispute unfolded in the lead-up to a major AI conference being held in San Francisco from October 27-29,2025. The conference is expected to be a key battleground for showcasing the latest advancements and debating the future of AI. Both OpenAI and Meta are expected to present their research and strategies.
The location, San Francisco, is a fitting backdrop for this conflict, given its status as a global hub for technological innovation and venture capital investment in AI. The conference will draw attendees from academia, industry, and government, making it a crucial event for shaping the AI landscape.
A History of competition
The rivalry between Meta and OpenAI is not new. Both companies have been at the forefront of AI research for years, competing for talent, funding, and market share. Meta, under LeCun’s leadership, has invested heavily in open-source AI models, such as Llama 2, aiming to democratize access to AI technology. OpenAI, initially a non-profit, has increasingly adopted a more closed and commercially focused approach.
Here’s a timeline of key events:
| Date | Event |
|---|---|
| 2016 | OpenAI founded. |
| 2018 | Meta (then Facebook) AI Research (FAIR) established. |
| 20
|