All Updates
Page 935 of 1653
April 16, 2026
Anthropic Shelves Mythos Over Hacking Risks
Anthropic determined its AI system Mythos was too dangerous for release after experts warned it could hack foundational systems of modern computing. The company heeded internal alarms about its potential to exploit core infrastructure. Banks and government agencies are urgently assessing the threat.
ASML: AI Hype Offsets Litho Dip?
ASML issued full-year guidance amid AI boom. Short-term performance impact remains minor. Investors should monitor order data closely.
Hard Tech IPOs Dominate Q1 Surge
Hard tech captured nearly 60% of Q1 fundraising, with Beijing Stock Exchange holding half the market. All 30 new listings rose on debut, averaging 163.55% gains.
Train Multimodal Embeddings with Sentence Transformers
Hugging Face blog details training and finetuning multimodal embedding and reranker models using Sentence Transformers. It covers handling text and images for advanced retrieval tasks. Developers gain open-source tools for building superior RAG systems.
The PR you would have opened yourself
Hugging Face's blog post highlights 'The PR you would have opened yourself'. It teases a platform improvement or pull request addressing a common contributor need. The post aims to showcase user-centric updates on their open AI platform.
OpenAI Launches GPT-Rosalind for Life Sciences
OpenAI has introduced GPT-Rosalind, a frontier reasoning model tailored for life sciences. It accelerates drug discovery, genomics analysis, protein reasoning, and scientific research workflows. This launch targets key challenges in biotech research.
OpenAI Launches Cyber Defense with GPT-5.4-Cyber
Leading security firms and enterprises are joining OpenAIβs Trusted Access for Cyber. The program leverages GPT-5.4-Cyber and offers $10M in API grants. It aims to bolster the global cyber defense ecosystem.
Ecom-RLVE: Adaptive Environments for E-Com Agents
Hugging Face introduces Ecom-RLVE, adaptive and verifiable environments designed for training reinforcement learning-based e-commerce conversational agents. This framework enables realistic simulations for improving agent interactions in shopping scenarios. It advances RL applications in retail AI.
Claude Opus 4.7 Now in GitLab Duo Agents
GitLab Duo Agent Platform now supports Anthropic's Claude Opus 4.7 model for Agentic Chat and workflows. It offers improved reasoning for complex, multistep tasks like CI/CD, code review, and vulnerability resolution. This enhances agent reliability across the software delivery lifecycle.
CI Expert and Data Analyst Agents
GitLab 18.11 launches CI Expert Agent (beta) to generate repo-aware CI pipelines and Data Analyst Agent (GA) for delivery metrics insights. CI Expert detects languages/frameworks and explains configs in Agentic Chat. Agents leverage GitLab-native context for targeted dev lifecycle gaps.
Budget Guardrails for GitLab Credits
GitLab 18.11 introduces subscription-level spending caps and per-user limits for on-demand GitLab Credits in Duo Agent Platform. Caps pause access upon reaching limits, with options to adjust mid-month. These controls provide governance to manage AI spending predictably.
Apple's MixAtlas Boosts Multimodal LLM Training
Apple introduces MixAtlas, an uncertainty-aware framework for optimizing data mixtures in multimodal LLM midtraining, accepted at ICLR 2026's NADPFM workshop. It tackles underexplored multimodal pretraining via principled domain reweighting for better sample efficiency and generalization. The approach uses systematic domain decomposition and smaller proxy models for compute efficiency.
Agentic SAST Auto-Fixes Go GA in GitLab
Agentic SAST Vulnerability Resolution is now generally available, generating ready-to-merge MRs for SAST vulnerabilities. It analyzes issues, creates root-cause fixes, and validates via testing. GitLab 18.11 adds faster incremental scanning and CVSS 4.0-based prioritization.
April 15, 2026
H3C Revenue Soars 38% to 759B CNY
Unisplendour reports 2025 revenue of 967B CNY up 22%, with subsidiary H3C at 760B CNY up 38%. Leads China markets in switches, routers, WLAN, servers. But gross margins fall to 16.5%, inventory hits record, debts and impairments rise.
Desay SV HK IPO Fuels Robot Pivot
Cash-flush Desay SV files for HK listing despite 80B+ CNY liquidity, targeting overseas expansion and new robot/unmanned vehicle growth. Launches AI Cube robot compute base and 'Chuangxing Zhiyuan' low-speed autonomous vehicles using auto-grade tech.
HappyHorse-1.0 Launches on Arena
HappyHorse-1.0 has launched on the LMSYS Chatbot Arena for public evaluation. The model is set for official release in two weeks. This development is highlighted in Tita Media's morning news briefing.
Thesis Launches Agent-Native ML Workspace
Thesis is a new agent-native workspace for running and tracking ML experiments. It integrates dataset inspection, training launches, metric monitoring, and agent-driven iteration in one interface. The tool aims to reduce fragmentation in model development workflows.
Games as AI Data Harvest Tools?
Steam's 'Data Center' sim may harvest gamer data for AI training on DC optimization. Likens to reCAPTCHA for solving NP-hard problems like routing and cooling. Seeks similar examples in sim-to-real RL.
OpenAI Exits Second Stargate Data Center Deal
OpenAI withdrew from two European Stargate data center deals in the UK and Norway within a week, reflecting financial discipline ahead of an IPO. In Norway, Microsoft leased the Nscale facility in Narvik and will provide compute power to OpenAI. Analysts view this as prudent cost control amid high expenses and competition.
Censored Qwen Blocks FTP Credentials
A user expresses frustration with Qwen3.5-122B refusing to connect to FTP using provided credentials due to security policies. The model suggests secure alternatives but cannot execute external actions. Prompt tweaks like instructing to 'act' enabled a workaround.