Technology
OpenAI intros two open-weight language models that can run on consumer GPUs — optimized to run on devices with just 16GB of memory
Source
Tom's Hardware
Published
Similar Articles
SemiEngineering
Silent Data Corruption: A Major Reliability Challenge in Large-Scale LLM Training (TU Berlin)
Researchers at Technische Universitat Berlin published a technical paper on the challenges of Silent Data Corruption (SDC) in Large Language Model (LLM) training. As LLMs grow in size, hardware-induced faults like SDC can bypass detection mechanisms, leading to severe consequences during training. The study explores how intermittent SDC impacts LLM pretraining, highlighting the sensitivity of different factors like bit positions and kernel functions. The research proposes a lightweight detection method to identify harmful parameter updates and demonstrates the effectiveness of recomputing training steps upon detection in mitigating corruption.
SemiEngineering•
OpenAI building GitHub alternative after frequent platform outages and disruptions — a public OpenAI code repository would directly compete with one of its biggest investors
OpenAI is in the process of creating its own code repository platform as an alternative to GitHub due to frequent outages on GitHub that hindered OpenAI engineers' work. The new platform is still in early development and may not be ready for months. OpenAI is considering offering access to this platform to customers, potentially competing directly with Microsoft, which owns GitHub and has a significant stake in OpenAI. GitHub has faced reliability issues following a migration to Microsoft Azure, with several outages attributed to Azure or configuration changes. If OpenAI commercializes its platform, it would challenge Microsoft in a new way, although losing OpenAI as a customer would have limited impact on GitHub given its large user base.
Tom's Hardware•
Nvidia's plan to invest $100 billion in OpenAI appears unlikely — Jensen reportedly criticizing OpenAI's business decisions in private discussions
Nvidia's $100 billion investment in OpenAI appears to be in jeopardy as talks have stalled since the initial agreement in September 2025. Nvidia CEO Jensen Huang has reportedly criticized OpenAI's business decisions in private discussions, expressing concerns about the company's lack of business discipline and competition from other AI companies like Anthropic and Google. Despite this, Huang still supports financially backing OpenAI due to its significance as one of Nvidia's largest customers. The future of the partnership between Nvidia and OpenAI remains uncertain, with the potential for OpenAI to receive a substantial cash infusion as it prepares for an IPO.
Tom's Hardware•
OpenAI spills technical details about how its AI coding agent works
OpenAI engineer Michael Bolin shared technical insights on how the Codex CLI coding agent functions, revealing details on how it writes code, runs tests, and fixes bugs under human supervision. AI coding agents like Codex are gaining popularity for their ability to quickly generate code for prototypes and interfaces. However, these tools are not flawless and may require human intervention for complex tasks beyond their training data. Bolin's post addresses engineering challenges such as prompt growth inefficiency and performance issues caused by cache misses. This level of technical transparency is uncommon for OpenAI, providing developers with a deeper understanding of how Codex operates.
Ars Technica•