AI News Apr 29, 2026

Mistral has released its new model Medium 3.5, now also available as a remote coding agent in the cl…

Mistral has released its new model Medium 3.5, now also available as a remote coding agent in the cloud. Developers can start tasks directly from Le Chat or the CLI and get notified when the work is done. The model is now in public preview.

A study shows that ChatGPT gives wildly inconsistent carb estimates for the same food photo – even a…

A study shows that ChatGPT gives wildly inconsistent carb estimates for the same food photo – even after 27,000 queries, it rarely gave the same answer twice. This could lead to dangerous hypoglycemic episodes for diabetics relying on AI. The findings raise serious questions about AI reliability in healthcare.

AI companies like Anthropic publicly warn about the dangers of their own models – yet still sell the…

AI companies like Anthropic publicly warn about the dangers of their own models – yet still sell them. This article examines how 'fear marketing' strategies bring regulatory advantages and media attention. Using Claude Mythos as an example, it shows how companies amplify risks to strengthen their position.

A provocative opinion piece argues that it's not AI refuseniks but AI dependents who will be left be…

A provocative opinion piece argues that it's not AI refuseniks but AI dependents who will be left behind. Relying too heavily on AI, the author claims, erodes critical thinking, writing, and learning skills – eventually making people unable to distinguish fact from fiction. A warning about a generation that can't function without AI.

Anthropic has launched 'Claude for Creative Work,' a new offering for creative professionals. Claude…

Anthropic has launched 'Claude for Creative Work,' a new offering for creative professionals. Claude helps with faster ideation, handles repetitive tasks, and enables larger projects – without replacing personal taste or imagination. The goal is to integrate AI seamlessly into creative workflows.

DeepSeek V4 is causing a stir in the AI industry: instead of scaling up compute and parameters, it u…

DeepSeek V4 is causing a stir in the AI industry: instead of scaling up compute and parameters, it uses a combinatorial optimization approach. The model features improved attention mechanisms, an efficient MoE architecture, and targeted post-training – defining a new era of 'restrained model optimization.' Ten industry experts share their candid assessments.

Evaluating AI models is becoming a major bottleneck – both in time and cost. While training compute …

Evaluating AI models is becoming a major bottleneck – both in time and cost. While training compute costs are dropping, the expense of thorough testing and benchmarking is rising. This blog post analyzes why 'AI evals' are the new infrastructure bottleneck and how the community is responding.

A NewsGuard audit found that Mistral's Le Chat chatbot repeats state-sponsored disinformation about …

A NewsGuard audit found that Mistral's Le Chat chatbot repeats state-sponsored disinformation about the Iran war 60% of the time – a much higher error rate than comparable models. The incident highlights the challenges of combating propaganda in AI systems.

Google Gemini can now create full documents, spreadsheets, and presentations directly inside the cha…

Google Gemini can now create full documents, spreadsheets, and presentations directly inside the chat – from PDFs and Word files to Excel workbooks. This feature makes working with Office formats seamless and saves time on content creation. Another step toward integrating AI into everyday office tasks.

IBM has launched 'Bob,' a platform for AI-powered software development with multi-model routing and …

IBM has launched 'Bob,' a platform for AI-powered software development with multi-model routing and human checkpoints. The system aims to make AI coding safer and more production-ready by combining different models and requiring human decisions at critical points. A step toward trustworthy AI agents in the enterprise.

This short video explains how to access top AI models like Llama, DeepSeek, and GLM for free on NVID…

This short video explains how to access top AI models like Llama, DeepSeek, and GLM for free on NVIDIA's cloud platform (build.nvidia.com) – no credit card required. Perfect for professionals who want to experiment with cutting-edge AI without spending a dime.

Bloomberg explores how AI-powered humanoid robots are moving from viral videos to real-world factory…

Bloomberg explores how AI-powered humanoid robots are moving from viral videos to real-world factory trials. The report covers data gaps, global competition, and billions in investment – asking whether the technology can live up to the enormous expectations.

Productivity

Over-reliance on AI assistants can be problematic.

Limit AI assistant usage to specific tasks

Productivity

Updating LLM with backward compatibility.

Ensure backward compatibility when updating models

Prompt Tips

Precise base instructions for AI models.

Provide precise base instructions for better results

Productivity

AI weakness with real-time data.

Do not rely on AI for real-time data cross-check with live sources

Creativity

Using AI for research proposals.

Use AI like NotebookLM to structure research proposals