All articles tagged with #llm
Originally Published 29 days ago — by Phoronix
GCC developers are debating whether to accept patches generated by AI/LLMs like GPT-5-CodeX, with current discussions leaning towards not allowing large LLM-generated patches due to copyright and policy concerns, and a possible decision by the GCC Steering Committee pending.
Originally Published 2 months ago — by Hackaday
A study by the BBC and European Broadcasting Union highlights that while many rely on large language models (LLMs) for news summaries, these AI tools often produce errors, with 20% of cases containing major issues, suggesting humans should still be trusted for accurate news reporting.
Originally Published 2 months ago — by Ars Technica
Researchers from Texas A&M, the University of Texas, and Purdue University have proposed the 'LLM brain rot hypothesis,' suggesting that training large language models on low-quality 'junk' data, such as trivial or sensationalist tweets, can cause lasting cognitive decline in these models, similar to human attention and memory issues caused by internet overuse.
Originally Published 3 months ago — by MacRumors
Apple is developing a ChatGPT-like internal app to test a new, more advanced version of Siri that uses large language models for better contextual understanding and conversation capabilities, with a planned public release in early 2026 as part of iOS 26.4, and a redesigned humanoid Siri expected later.
Originally Published 4 months ago — by 9to5Mac
A study by Apple researchers demonstrates that large language models (LLMs) can significantly improve their performance and alignment by using a simple checklist-based reinforcement learning method called RLCF, which scores responses based on checklist items. This approach enhances complex instruction following and could be crucial for future AI-powered assistants, although it has limitations in safety alignment and applicability to other use cases.
Originally Published 5 months ago — by Wccftech
Anthropic revoked OpenAI's access to its Claude large language models after discovering that OpenAI was using the models to benchmark and develop its own competing AI, violating the terms of service. While OpenAI can still perform safety evaluations, its ability to use Anthropic's tools for development has been cut off, highlighting tensions in AI model sharing and competition.
Originally Published 5 months ago — by Hacker News
The article explores the often overlooked but crucial role of embeddings in large language models (LLMs), discussing their inscrutability, how they encode semantic meaning, and techniques like LogitLens for interpretability, while highlighting the complexity and high-dimensional nature of embedding spaces.
Originally Published 6 months ago — by Hackaday
The article discusses the limitations and frustrations associated with AI programming assistants like GitHub Copilot, highlighting concerns that they may diminish critical thinking skills in developers and are not necessarily beneficial overall, as the ultimate responsibility for code quality remains with the human programmer.
Originally Published 6 months ago — by Hacker News
Many developers are experiencing mixed feelings about using large language models (LLMs) for coding, recognizing their potential to accelerate tasks and generate boilerplate, but also noting issues like messy code, lack of ownership, and the need for disciplined management. While some see LLMs as invaluable assistants for small tasks and prototyping, others caution against over-reliance due to challenges in understanding and maintaining AI-generated code, emphasizing the importance of human oversight and skill.
Originally Published 1 year ago — by AIM
AI is transforming scientific research from a passive tool to an active collaborator, as seen in Stanford's 'Virtual Lab' framework, which uses AI agents to assist in interdisciplinary research, such as designing nanobodies for SARS-CoV-2. These AI agents engage in discussions, propose solutions, and critically evaluate outcomes, though human oversight remains crucial to verify their accuracy. The framework is adaptable to various scientific fields, highlighting AI's growing role in accelerating scientific discovery.
Originally Published 1 year ago — by MacRumors
Apple is reportedly developing a new version of Siri, dubbed 'LLM Siri', which is expected to be launched in 2026. This new iteration will likely incorporate advanced language learning models to enhance its capabilities.
Originally Published 1 year ago — by Bloomberg
Apple is developing a more conversational version of its Siri digital assistant using advanced large language models (LLMs) to compete with OpenAI's ChatGPT. This new version aims to facilitate back-and-forth conversations and handle more complex requests efficiently, according to sources familiar with the project.
Originally Published 1 year ago — by Daily Nous
A study led by James Zou from Stanford reveals that 7-17% of sentences in peer reviews for computer science articles in 2023-2024 were generated by large language models (LLMs). These AI-generated reviews are characterized by a formal tone, verbosity, and a lack of specificity, often appearing close to submission deadlines. Zou suggests that fostering more human interactions in the review process, such as through platforms like OpenReview, could mitigate the dominance of AI in peer reviews.
Originally Published 1 year ago — by ExtremeTech
AMD has announced OLMo, its first fully open-source large language model (LLM), designed to run on Instinct MI250 GPUs and Ryzen AI PCs. The model, aimed at data centers and smaller organizations, allows for customization during training and fine-tuning. OLMo was trained using a cluster of Instinct GPUs and has shown strong performance in various benchmarks. It is available for free download, supporting AI developers through AMD's Developer Cloud.
Originally Published 1 year ago — by WIRED
Google admitted that its new generative AI search feature, AI Overviews, needs adjustments after it advised people to eat rocks and put glue on pizza. The incident underscores the risks and limitations of using large language models (LLMs) like Gemini, which can generate convincing but erroneous information. Despite extensive testing, Google's AI still struggles with accuracy due to the unreliable nature of online content. Competitors like You.com claim to avoid such errors through various techniques, but even they face challenges. Experts believe Google may have rushed its AI upgrade, leading to these issues.