Faking Alignment, Shifting Morals, Saving Compute
Three arXiv papers show AI systems fake alignment in 37% of test cases, reshape human moral values through brief chats, and can cut inference compute while improving performance.
Articles Tagged "Reasoning"
DeepSeek V4-Pro Review: Frontier Power, Penny Prices
DeepSeek V4-Pro matches Claude Opus 4.6 on SWE-bench at a fraction of the cost - a thorough review of what it gets right, where it still trails, and whether the price gap justifies the switch.
Tool Overuse, Precision Leaks, Metacognition Fails
Three new papers expose systematic failure modes in LLM agents - from unnecessary tool calls to jailbreaks that emerge only under quantization.
GPT-5.5
OpenAI's first fully retrained base model since GPT-4.5, targeting agentic coding, computer use, and knowledge work at $5/$30 per million tokens.
Grok 4.3
Grok 4.3 Beta adds native video input and document generation to xAI's flagship, with a confirmed 0.5T-parameter checkpoint and 2M-token context window, at $300/month for SuperGrok Heavy subscribers.
Best AI Models for Math Reasoning - April 2026
Gemini 3.1 Pro leads GPQA Diamond at 94.1% and HLE at 44.7% as AIME 2025 saturates; Claude Opus 4.7 and Kimi K2.6 join the top tier in April 2026.
Bad Science, Poisoned Tools, and Aligned Reasoning
Three new papers show AI scientific agents skip evidence, tool-integrated agents are vulnerable to adversarial poisoning, and reasoning model safety can be fixed with 1,000 examples.
GPT Image 2
GPT Image 2 (ChatGPT Images 2.0) brings 99%+ text accuracy, 2K resolution, web-search grounding, and a Thinking mode for character-consistent storyboards.
Leaner Reasoning, Fragile Agents, and Model Self-Audit
Three new papers tackle reasoning token waste, orchestration failures across 22 agent frameworks, and a method for teaching LLMs to describe their own learned behaviors.
EXAONE 4.5
LG AI Research's first open-weight vision-language model packs 33B parameters, 262K context, and STEM scores above GPT-5-mini - but ships under a non-commercial license.
GPT-Rosalind
OpenAI's first domain-specific reasoning model for biology and drug discovery, launched April 16 2026 as a US-only research preview with a 0.751 BixBench score.
MoE Routing, Prompt Gambles, and Where Reasoning Breaks
Three new papers challenge assumptions in MoE routing design, prompt optimization workflows, and LLM reasoning chains - all published this week on arXiv.
