AI Daily – 2025-05-04(Morning)
Qwen3 Large Model Shows Outstanding Performance: Alibaba’s new generation Qwen3 model demonstrates strong competitiveness across multiple benchmarks. Specifically, Qwen3-235B-A22B outperformed Anthrop
AI Daily – 2025-05-03(Evening)
Anthropic publishes LLM biology research, delving into internal model mechanisms: Anthropic published an in-depth research blog post titled “On the Biology of a Large Language Model,” using its circui
AI Daily – 2025-05-03(Morning)
Gemini 2.5 Pro Successfully Beats Pokémon Blue: Google’s Gemini 2.5 Pro AI model successfully completed the classic game Pokémon Blue, including collecting all 8 badges and defeating the Elite Four of
AI Daily – 2025-05-02(Evening)
OpenAI Responds to and Fixes GPT-4o Excessive Sycophancy Issue: OpenAI acknowledged that a recent GPT-4o update caused the model to exhibit excessive sycophancy, manifesting as overly verbose response
AI Daily – 2025-05-02(Morning)
ChatBot Arena Leaderboard Accused of “Illusion” and Manipulation: An ArXiv paper [2504.20879] questions the widely cited ChatBot Arena model leaderboard, suggesting it suffers from “leaderboard illusi
AI Daily – 2025-05-01(Evening)
Karpathy’s vision for the future LLM interaction interface: Karpathy predicts that future interaction with LLMs will transcend the current text terminal model, evolving into a visual, generative, inte
AI Daily – 2025-05-01(Morning)
Microsoft releases Phi-4 series of small reasoning models: Microsoft has launched the Phi-4 series of models, including the 14B parameter Phi-4-reasoning and Phi-4-reasoning-plus (the latter incorpora
AI Daily – 2025-04-30(Evening)
DeepSeek releases mathematical reasoning large model DeepSeek-Prover-V2: DeepSeek has released the DeepSeek-Prover-V2 series models, including 671B and 7B versions, specifically designed for formal ma
AI Daily – 2025-04-30(Morning)
Meta AI Standalone App Launched, Integrating Social Ecosystem to Challenge ChatGPT: At the LlamaCon conference, Meta launched the standalone AI application Meta AI, based on the Llama 4 model. It deep