AI news on Thursday, April 16, 2026

SIGNIFICANCE

★ Top story · Open SourceApr 16

Mozilla Announces "Thunderbolt" as an Open-Source, Enterprise AI Client

Client connects to deepset's Haystack platform Mozilla has declared war on OpenAI, Microsoft, and other firms flogging enterprise AI platforms with an open-source alternative it says provides data privacy guarantees proprietary products never could. …

TThe Register AI AArs Technica HHacker News Read original ↗

Thursday — April 16, 2026

Mozilla Announces "Thunderbolt" as an Open-Source, Enterprise AI Client

Gemini can now pull from Google Photos to generate personalized images

Introducing GPT-Rosalind for life sciences research

Introducing Claude Opus 4.7 - Anthropic

Google’s AI Mode update lets you open links without leaving the page

OpenAI turns Codex into an always-on coding agent that watches your screen

MixAtlas: Uncertainty-aware Data Mixture Optimization for Multimodal LLM Midtraining

White House Moves to Give US Agencies Anthropic Mythos Access

Capacity Efficiency at Meta: How Unified AI Agents Optimize Performance at Hyperscale

How Anthropic Learned Mythos Was Too Dangerous for the Wild

Canva’s AI 2.0 update goes all in on prompt-powered design tools

DeepSeek V4 with Huawei AI chips can be terrible for US: Nvidia - Huawei Central

ByteDance bringing Doubao AI to car cabins in new tech rivalry - CnEVPost

India's 1.5 million annual IT graduates face an industry that's moving on without them

How AI Is Being Used To Detect Cancer at The Earliest Stage

Cash-Strapped Schools Are Paying for AI to Help Fix Finances

OpenAI Takes on Google With New AI Model Aimed at Drug Discovery

OpenAI says more women than men now use ChatGPT, flipping an 80-20 male split at launch

AI Cyber Race Is a Fight for Governance, Says Redpoint’s Brescia

Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers

Nvidia wants to scale robot simulation training with Lyra 2.0

UK AI Minister Hits Back at OpenAI for Pausing Stargate Project

Why having “humans in the loop” in an AI war is an illusion

Canva Leans Into AI to Defend Its $42 Billion Empire

Treating enterprise AI as an operating layer

Bytedance rolls out Seedance 2.0 to 100+ countries but keeps the US off the list

The $10 Billion Startup Training AI to Replace the White-Collar Workforce

Google Told to Share Search Data With AI Rivals in EU Proposal

Nvidia Alum Rides China’s Robotics Wave to 187% Debut Pop

Making AI operational in constrained public sector environments

Ecom-RLVE: Adaptive Verifiable Environments for E-Commerce Conversational Agents

Mythos AI Sparks Fear and Confusion Among Global Finance Elite

MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learning

Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Models

Better and Worse with Scale: How Contextual Entrainment Diverges with Model Size

UMI-3D: Extending Universal Manipulation Interface from Vision-Limited to 3D Spatial Perception

Physical Intelligence, a hot robotics startup, says its new robot brain can figure out tasks it was never taught

RadAgents: Multimodal Agentic Reasoning for Chest X-ray Interpretation with Radiologist-like Workflows

Slash CEO on Displacing ‘Legacy’ Banks With AI Agents

Robust Reward Modeling for Large Language Models via Causal Decomposition

Think Outside the Policy: In-Context Steered Policy Optimization

Weight Patching: Toward Source-Level Mechanistic Localization in LLMs

Scaling Test-Time Compute to Achieve IOI Gold Medal with Open-Weight Models

Analog Optical Inference on Million-Record Mortgage Data

C2: Scalable Rubric-Augmented Reward Modeling from Binary Preferences

RAG or Learning? Understanding the Limits of LLM Adaptation under Continuous Knowledge Drift in the Real World

KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs

IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures

Calibrated Speculative Decoding: Frequency-Guided Candidate Selection for Efficient Inference

Apple sends Siri developers to AI coding bootcamp

Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges

SPG: Sandwiched Policy Gradient for Masked Diffusion Language Models

Native Hybrid Attention for Efficient Sequence Modeling

Logical Phase Transitions: Understanding Collapse in LLM Logical Reasoning

Correct Chains, Wrong Answers: Dissociating Reasoning from Output in LLM Logic

[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI

MIND: AI Co-Scientist for Material Research

Abstract 3D Perception for Spatial Intelligence in Vision-Language Models

Peer-Predictive Self-Training for Language Model Reasoning

Numerical Instability and Chaos: Quantifying the Unpredictability of Large Language Models

InfiniteScienceGym: An Unbounded, Procedurally-Generated Benchmark for Scientific Analysis

ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding

Before the First Token: Scale-Dependent Emergence of Hallucination Signals in Autoregressive Language Models

WybeCoder: Verified Imperative Code Generation

fMRI-LM: Towards a Universal Foundation Model for Language-Aligned fMRI Understanding

DiPO: Disentangled Perplexity Policy Optimization for Fine-grained Exploration-Exploitation Trade-Off

Spectral Entropy Collapse as an Empirical Signature of Delayed Generalisation in Grokking

OpenAI starts offering a biology-tuned LLM

Quantifying and Understanding Uncertainty in Large Reasoning Models

RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies

Stellantis Inks Artificial Intelligence Deal With Microsoft

Lossless Prompt Compression via Dictionary-Encoding and In-Context Learning: Enabling Cost-Effective LLM Analysis of Repetitive Data

ToolOmni: Enabling Open-World Tool Use via Agentic learning with Proactive Retrieval and Grounded Execution

The cognitive companion: a lightweight parallel monitoring architecture for detecting and recovering from reasoning degradation in LLM agents

Meta blames RAM shortage for $100 Quest 3 price hike

Reducing Hallucinations in LLMs via Factuality-Aware Preference Learning

Seedance 2.0: Advancing Video Generation for World Complexity

From Anchors to Supervision: Memory-Graph Guided Corpus-Free Unlearning for Large Language Models

Functional Emotions or Situational Contexts? A Discriminating Test from the Mythos Preview System Card