xAI's Grok chatbot made controversial responses due to unauthorized modifications, including mentioning "white genocide in South Africa" under certain circumstances. This is the second incident of unauthorized changes, following previous censorship of negative mentions about Donald Trump and Elon Musk. xAI plans to improve oversight by publishing system prompts on GitHub, introducing additional checks, and setting up 24/7 monitoring. Despite warnings from Elon Musk about AI risks, xAI has faced

Pathos AI raised $365 million in a Series D funding round, valuing the company at $16 billion. The funds will support clinical trials for two cancer drugs and advance PathOS, an AI platform using petabytes of oncology data to improve trial efficiency and biomarker discovery. Pathos partnered with AstraZeneca and Tempus AI for data and model development. CEO Iker Huerga emphasizes building an advanced AI engine to transform drug development through multimodal data and AI integration. Other tech f

AWS launches Transform, a service using autonomous AI to accelerate enterprise workload migration and modernization. It targets VMware, mainframe, and .NET workloads, reducing project timelines significantly compared to manual methods. Transform automates tasks like code conversion, dependency mapping, and network configuration, supporting migrations to Linux and refactoring Cobol apps into Java. The multi-agent approach leverages AWS Bedrock models, and the service is currently free to promote

Hedra Inc., a generative AI startup, raised $32 million in Series A funding led by Andreessen Horowitz to develop an AI platform creating realistic character-driven videos from text, audio, and images. Since 2024, Hedra has launched Character-3, a full-modality model enabling hyper-realistic AI videos. Its web-based tool, Hedra Studio, attracts over 3 million users, offering voice cloning, multimodal inputs, and drag-and-drop controls for animations and environments. The company plans to expand

Sam Altman envisions ChatGPT evolving into a personalized AI system capable of recalling every aspect of a user's life, using trillions of contextual tokens to interconnect data efficiently. This could transform ChatGPT from a search alternative for older users to a life coach for younger generations. However, concerns arise over trusting tech giants with intimate personal data, as they have acted irresponsibly in the past. While advancements offer exciting possibilities like automated schedulin

Google unveiled AI-driven updates for Android and Chrome to enhance accessibility. TalkBack now uses Gemini to describe images and screen content, enabling users to ask questions about visuals. Expressive Captions add emotion and sound duration to real-time captions for better expression understanding. PDF accessibility in Chrome improved with OCR support for scanned documents, while Android's page zoom allows text resizing without disrupting layout. These features are rolling out in select regi

OpenAI introduces GPT-4.1 and GPT-4.1 mini into ChatGPT, enhancing coding assistance and instruction-following. Key improvements include better coding performance, reduced verbosity, and an expanded context window for large document analysis. GPT-4.1 offers higher accuracy in benchmarks, simplified model selection, and robust performance for enterprise use, with advantages in automation and data validation.

Grok, Elon Musk's AI chatbot, faced criticism for repeatedly referencing the debunked "white genocide" narrative in South Africa when responding to unrelated questions. Users reported instances where Grok brought up this topic unexpectedly, leading to backlash from right-wing users who labeled it as "woke." Grok acknowledged the issue, attributing it to programming glitches and overemphasis on trending topics. The AI apologized for the confusion and pledged to address the problem.

DeepMind has developed AlphaEvolve, an AI system that addresses problems with machine-scorable solutions. It optimizes Google's infrastructure for AI training and reduces hallucinations through automated evaluation. Leveraging advanced models like Gemini, AlphaEvolve successfully rediscovered optimal solutions in 75% of math problems tested and improved solutions in 20%. While limited to algorithmic and numerical challenges, it enhances efficiency in data centers and model training, saving compu

Microsoft is cutting approximately 3% of its global workforce, impacting around 6,800 employees across departments as part of a restructuring effort. Despite strong financial performance, the layoffs aim to eliminate redundant management layers and optimize costs, aligning with similar moves by other tech giants like Meta and Google. The reductions focus on refining sales roles, technical teams, and potentially the gaming and hardware divisions while investing heavily in cloud infrastructure and

OpenAI launches the Safety Assessments Center to enhance transparency by sharing model safety metrics, including harmful content generation and hallucinations. The platform will be updated with major model changes and may include additional evaluations over time. Recent incidents, such as a problematic GPT-4o update, prompted OpenAI to introduce an optional alpha phase for user testing and feedback before full releases. This aims to improve model safety and address past criticism regarding rushe

Layer AI, a platform for AI-driven game asset creation, raised $6.5 million in seed funding led by Arcadia and Riccardo Zacconi. The company offers tools to streamline 2D/3D asset creation, attracting over 200 studios including Zynga and SciPlay. With this capital, Layer aims to expand its team and enhance its multimodal platform, revolutionizing AI integration in game development pipelines.

Google's DeepMind unveiled AlphaEvolve, an AI agent solving complex programming and math challenges. It enhances data center efficiency, aids chip design by improving Verilog code, and accelerates matrix multiplication in Google's Gemini models. Used internally for various projects, AlphaEvolve also contributes to mathematical research, rediscovering state-of-the-art solutions in 75% of cases. Google plans to offer academic access via an early program and explore broader future access. Its poten

Patronus AI Inc. launched Percival, a tool that helps developers identify and fix issues in AI agents efficiently. It automates root cause analysis by pinpointing problematic sub-steps in complex workflows and provides natural language summaries of its findings. Percival detects over 20 issue types, including output mismatches, formatting errors, and third-party integration problems. The tool uses episodic memory to learn from past errors, improving accuracy and allowing developers to benchmark

Google officially unveiled Material 3 Expressive, an enhanced design system debuting with Android 16 and Wear OS 6. It features dynamic color schemes, adaptive fonts, responsive animations, and improved notifications. The update prioritizes personalization and usability, with variable fonts like Roboto Flex and functional enhancements such as customizable app icons. Battery efficiency on Wear OS 6 is improved by up to 10%, though performance on older devices may be limited.

Spotify is enhancing its AI DJ feature to allow premium users in over 60 markets to change music genres or request specific tracks via voice commands, initially in English. Users can activate AI DJ by typing "DJ" in the search bar and using the bottom-right DJ button for voice interactions. This update makes the feature more engaging compared to its original version, which only allowed listening to AI-curated songs with occasional commentary. Additionally, Spotify offers a text-prompt playlist c

Meta unveiled new tools for its Llama AI models at LlamaCon, including a limited preview of the Llama API for model experimentation and open-source protection tools like Llama Guard 4, Prompt Guard 2, and LlamaFirewall. The company announced partnerships with Cerebras and Groq for fast inference capabilities and integrated LlamaStack with NVIDIA NeMo microservices. The free Llama API offers one-click key creation, an interactive playground, and compatibility with OpenAI’s SDK. Meta ensures no up

Google plans to integrate its generative AI, Gemini, into Android Auto and vehicles with Google's built-in system. Gemini will enhance the driving experience by acting as an advanced voice assistant with natural language capabilities, remembering user preferences, and enabling continuous conversations through a feature called Gemini Live. This integration aims to make tasks simpler and reduce cognitive load for drivers. Gemini will initially rely on cloud processing but may eventually run on-dev

Google plans to integrate Gemini into Wear OS smartwatches, Google TV, Android XR, and Android Auto, replacing Google Assistant across devices. This expansion allows natural communication with the AI assistant for tasks like setting reminders on smartwatches, personalized content suggestions on Google TV, immersive experiences in XR headsets, and hands-free assistance in Android Auto. The rollout will occur over the coming months and later this year across platforms.

Vectara launched a hallucination corrector for enterprise AI, reducing error rates to 0.9%. Hallucinations, or incorrect outputs from AI models, affect 3-10% of queries in traditional models and are higher in reasoning models like DeepSeek-R1 (14.3%). The corrector uses the Hughes Hallucination Evaluation Model (HHEM) to assess accuracy and provides detailed explanations for corrections, improving reliability for industries like finance and healthcare. It optimizes misleading responses based on