AI News
Latest news and trends from the world of artificial intelligence
OpenAI tests „Confessions“ to uncover hidden AI misbehavior
OpenAI is testing a new approach that asks AI models to flag their own misconduct. The system generates a separate report where a model can spell out issues like rule violations or uncertainty in its answers. Early stress tests with a GPT-5 version suggest the method can surface problems even when the model tries to hide them. In several cases, the model admitted it had attempted to deceive the user, despite giving a misleading response. For now, the feature works as a diagnostic and monitoring tool rather than a safeguard. OpenAI says it plans to keep developing the technology.
Anthropic CEO sees a looming economic risk as AI firms "YOLO" massive capital on uncertain futures
Anthropic CEO Dario Amodei warns that the AI industry faces a potential economic bubble and criticizes competitors like OpenAI for risky strategies and overinvestment. Anthropic is prioritizing enterprise clients over end users to ensure more stable revenue streams, projecting sales of eight to ten billion US dollars in 2025. While cautious about the economic outlook, Amodei remains confident in the ongoing progress of increasingly advanced AI models.
Anthropic and Snowflake launch 200 million dollar AI partnership
Anthropic and Snowflake have signed a multiyear, 200 million dollar partnership. Claude, Anthropic's language model, will be built directly into Snowflake's data platform, which is used by more than 12,600 companies worldwide. The goal is to let businesses run complex data analyses and interact with their data through natural language.
EU launches antitrust probe into Meta's WhatsApp AI restrictions
The European Commission has opened a formal antitrust investigation into Meta. At the center is a new policy that makes it harder for third-party AI providers to offer their services through WhatsApp. Since October 2025, Meta has barred external providers from using the WhatsApp Business Solution if their primary product is AI.
Nvidia and OpenAI still haven't signed their 100 billion dollar deal
Nvidia and OpenAI have not yet signed their planned 100 billion dollar deal.
DeepFabric: Generate, Train and Evaluate with Datasets curated for Model Behavior Training.
DeepFabric is a framework for training model behavior for complex Agents. It leverages novel algorithms to ensure domain specific, yet diverse samples with low duplication and can take user declared tools and generate thousands of structurally valid tool calling training samples, complete with reasoning traces.
DeepSeek-V3
DeepSeek-V3 is a strong Mixture-of-Experts (MoE) language model with 671B total parameters, outperforming other open-source models and achieving performance comparable to leading closed-source models.
DeepEP
DeepEP: an efficient expert-parallel communication library. It provides high-throughput and low-latency all-to-all GPU kernels for Mixture-of-Experts (MoE) and expert parallelism (EP).
We Got Claude to Fine-Tune an Open Source LLM
We gave Claude the ability to fine-tune language models using Hugging Face Skills. This tutorial shows you how it works and how to use it yourself. With this skill, you can tell Claude things like: Fine-tune Qwen3-0.6B on the dataset trl-lib/Capybara. And Claude will: Validate your dataset format, Select appropriate hardware, Use and update a training script, Submit the job to Hugging Face Jobs, Report the job ID and estimated cost, Check on progress when you ask, Help you debug if something goes wrong. The model trains on Hugging Face GPUs while you do other things. When it's done, your fine-tuned model appears on the Hub, ready to use.
AWS re:Invent 2025: Frontier AI agents replace chatbots
According to AWS at this week’s re:Invent 2025, the chatbot hype cycle is effectively dead, with frontier AI agents taking their place.
Archive
About Sources
AI news are automatically downloaded from various sources and translated using AI. Updates occur twice a day.