Large Language Models
Ongoing coverage on Large Language Models.
Latest coverage
Not that we should be at all surprised, but Nvidia leans on AI pretty hard for speeding up how it plans and designs its…
Nvidia is heavily leveraging artificial intelligence to accelerate its chip design processes. Tools like NVCell, utilizing reinforcement learning, can now generate standard cell libraries overnight with results superior to human designs. AI is also being used to speed up design verification and explore novel approaches to complex problems like carry-lookahead chains. Furthermore, Nvidia has fine-tuned large language models, such as ChipNeMo, using proprietary design documents to create AI assistants that can help junior designers understand complex hardware components.
ChatGPT will praise the mood and 'bedroom/DIY texture' of fart sounds pulled from YouTube
A YouTuber shared an experiment where ChatGPT provided a detailed, albeit positive, critique of a 37-second audio file of fart sounds. The AI praised the 'bedroom/DIY texture' and 'indie game menu music' vibe, while also offering constructive criticism on structure, mixing, and sound selection. The article highlights the AI's ability to generate nuanced feedback, even for unconventional audio.
It's no longer free to use Claude through third-party tools like OpenClaw
Anthropic has updated its Claude AI subscription policy, discontinuing free usage for third-party tools like OpenClaw as of April 4. Users will now require an extra usage bundle or an API key to access Claude through external applications. This change is attributed to engineering constraints and a need to manage increasing demand by prioritizing direct customers and API users.
Google releases Gemma 4, a family of open models built off of Gemini 3
Google has released Gemma 4, a new family of open-weight large language models built upon the technology behind Gemini 3 Pro. The models come in various sizes, including 2-billion and 4-billion parameter versions for edge devices, and 26-billion and 31-billion parameter versions for more powerful hardware. Google claims these models offer an unprecedented level of intelligence-per-parameter and are capable of processing video, images, and audio in over 140 languages.
Wikipedia has banned AI-generated articles
English Wikipedia has banned the use of generative AI for writing or rewriting articles, citing violations of core content policies. Exceptions exist for refining existing text or assisting with translations, provided human editors verify accuracy. Some Wikipedia sites, like Spanish Wikipedia, have implemented stricter bans.
First Look At Xbox’s Gaming Copilot Leaves A Bad Impression
Footage of Microsoft's AI-powered Gaming Copilot assistant for Xbox has been released, showcasing its ability to respond to player queries. While the assistant's sycophantic tone has drawn criticism, it demonstrates potential for assisting with in-game tasks like tuning cars in Forza Horizon 5 or providing tips for Diablo 4. The feature is expected to launch on Xbox Series X|S later this year and is currently in beta on PC and mobile apps.
Meta will move away from human content moderators in favor of more AI
Meta plans to significantly reduce its reliance on human content moderators over the next few years, transitioning to AI-based systems for content review. While humans will still handle critical decisions and training, the company believes AI can identify issues faster and more comprehensively, supporting a wider range of languages. This shift aims to improve efficiency and potentially reduce costs, though it may alter user perception of policy enforcement.
GPT-5.4 mini brings some of the smarts of OpenAI's latest model to ChatGPT Free and Go users
OpenAI has launched GPT-5.4 mini and nano, with GPT-5.4 mini now available to ChatGPT Free and Go users, offering performance comparable to GPT-5.4 in certain areas like reasoning and multimodal understanding. GPT-5.4 nano is being made available through OpenAI's API service for tasks requiring speed and cost-efficiency, such as data classification.
Brendan Greene says AI content is ruining the internet because it's 'a loop, LLMs are scanning this junk, and then that becomes truth… it's like a race to the middle of sh*t'
Brendan Greene, also known as PlayerUnknown, believes that the current implementation of AI, particularly large language models, is detrimental to the internet, creating a feedback loop of low-quality content. He argues that while domain-specific machine learning has its uses, the widespread application of generative AI is not true intelligence and is leading to a decline in online information quality.
Koei Tecmo is partnering with SpiralAI as it shows interest in creating "experiences that combine entertainment and AI" via chatbot-like LLMs
Koei Tecmo has announced a strategic partnership with SpiralAI, a startup focused on character-based conversational AI. The collaboration aims to leverage AI, specifically the new large language model Geppetto2, to create new forms of entertainment, including more natural and in-character conversations for video game NPCs. Koei Tecmo CEO Hisashi Koinuma stated the investment reflects the company's interest in using AI for novel entertainment experiences.
Apple Brings Powerful AI-Focused M5 Products & Studio Displays To Market
Apple has announced new M5-powered products including the MacBook Pro, MacBook Air, and updated Studio Displays. These devices feature enhanced AI capabilities with new M5 Pro and M5 Max chipsets designed for advanced machine learning tasks. The new lineup also includes Thunderbolt 5 connectivity for the Studio Displays and will be available for pre-order starting March 4.
