The Big Story
TechCrunch, OpenAI will no longer train its AI models on the live internet, which is a major shift in their approach. This decision comes after one of their engineers noted that training AI models on the internet would not be effective.
The move away from training AI models on the internet highlights the growing concerns about the risks associated with using unfiltered data to train AI systems. As AI technology continues to advance and become more pervasive in our daily lives, it is crucial that we prioritize the development of AI systems that are safe, reliable, and transparent.
The impact of this decision will be far-reaching, as OpenAI's approach has significant implications for the broader AI research community. The move away from training AI models on the internet may lead to a shift towards more controlled and curated datasets, which could have significant implications for the development of AI systems in various industries.
Furthermore, this decision underscores the importance of developing AI systems that are accountable and transparent. As AI technology becomes increasingly autonomous, it is essential that we prioritize the development of AI systems that can be held accountable for their actions and decisions. This requires a shift towards more robust AI governance frameworks and greater transparency in the development and deployment of AI systems.
In conclusion, OpenAI's decision to stop training its AI models on the live internet marks a significant turning point in the development of AI technology. The move away from unfiltered data highlights the growing concerns about the risks associated with using unfiltered data to train AI systems, and underscores the importance of developing AI systems that are accountable and transparent.
What Shipped
This week, NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model post-trained from Gemma 4 E4B for local voice assistants, game characters and robots, targeting about 4GB of VRAM on RTX and Jetson devices. According to MarkTechPost, this release aims to improve the performance and efficiency of AI-powered applications.
Moonshot AI also released Kimi K3.1 Pro, a 2.8T-parameter, 104B-active MoE model with 1M-token context. It scores 65.7% on Terminal-Bench 4.0, costs $2/$10 per million tokens, and open weights arrive by November 15th. According to MarkTechPost, this release aims to provide a highly accurate and efficient language model for various applications.
In addition, Kyutai released CADRE, a family of 7 open causal diffusion video autoencoders for interactive world models. According to MarkTechPost, this release aims to enable the creation of highly realistic and interactive 3D environments.
Finally, LitPivot: Developing research ideas through dynamic contextualization within the literature landscape is a new open-source tool that helps researchers develop research ideas by analyzing and visualizing relevant literature. According to Hugging Face, this release aims to facilitate the discovery of novel research ideas and improve the efficiency of the research process.
From the Labs
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model post-trained from Gemma 4 E4B for local voice assistants, game characters and robots, targeting about 4GB of VRAM on RTX and Jetson devices. MarkTechPost reports that this release aims to improve the performance and efficiency of AI-powered applications.
Moonshot AI also released Kimi K3.1 Pro, a 2.8T-parameter, 104B-active MoE model with 1M-token context. It scores 65.7% on Terminal-Bench 4.0, costs $2/$10 per million tokens, and open weights arrive by November 15th. MarkTechPost notes that this release aims to provide a highly accurate and efficient language model for various applications.
Kyutai released CADRE, a family of 7 open causal diffusion video autoencoders for interactive world models. MarkTechPost explains that this release enables the creation of highly realistic and interactive 3D environments.
LitPivot: Developing research ideas through dynamic contextualization within the literature landscape is a new open-source tool that helps researchers develop research ideas by analyzing and visualizing relevant literature. Hugging Face notes that this release facilitates the discovery of novel research ideas and improves the efficiency of the research process.
Other Notable News
The hottest toy for Christmas? Nvidia CEO plays with adorable Microduck robots, 20,000 sell, reports TechCrunch. The robots, stomping their little feet and making tiny R2-D2 noises, sat, fell down dead for a "nap," and summersaulted on voice command.
From Toast to Bessemer: Who founders will hear from at TechCrunch Founder Summit 2026, reports TechCrunch. The full-day event goes beyond fundraising and AI to explore the realities of finding customers, building teams, sharpening a pitch, and more.
What Companies Actually Mean When They Say They Want "AI Experience", reports KDNuggets. According to the article, what employers actually mean when job postings list "AI experience" as a requirement is often misunderstood.
Visualize points and routes with GeoJSON in Amazon Quick Sight layer maps, reports AWS. The dataset and GeoJSON workflow are covered in this post.
Bayer Crop Science built the Agentic Decision Engine (ADE), a multi-agent platform on Amazon Bedrock AgentCore that lets breeders, analysts, and researchers ask natural language questions about their data, reports AWS.
The Take
The AI landscape has been abuzz with developments this week, showcasing the rapid pace at which innovation is shaping our understanding and application of artificial intelligence. One of the most significant stories is OpenAI's announcement that it will no longer train its AI models on the live internet, citing "significant risks". This decision highlights the need for careful consideration when it comes to the potential consequences of AI training data.
On a related note, LitPivot has been making waves with their research into developing ideas through dynamic contextualization within the literature landscape. This approach has significant implications for how we think about and engage with AI-generated content.
In other news, Moonshot AI has released Kimi K3.1 Pro, a 2.8T-parameter model with 104B active parameters and 1M-token context. This development speaks to the ongoing push towards more powerful and capable AI models that can tackle complex tasks.
The importance of scaling online storage was also highlighted this week, as OpenAI detailed its efforts to scale Habitat with reliable multi-tenancy, global caching, and Azure Cosmos DB to deliver 12× faster reads for ChatGPT. This achievement underscores the need for innovative solutions in the face of rapidly growing AI usage.
Finally, we're seeing a greater focus on building stateful engineering copilots with Letta, as demonstrated by Stateful Agentic Engineering Copilots with Letta: Core Memory, Shared State, Semantic Retrieval, Custom Tools, Streaming, and Portable Agent Workflows. This emphasis on agent architecture holds significant potential for the future of AI development.