Kimi Released Kimi K2.5, Open-Source Visual SOTA-Agentic Model
Kimi K2.5 is the most advanced open-source model, featuring 15T mixed visual and text tokens for enhanced coding and vision capabilities, enabling self-directed agent swarms of up to 100 sub-agents for parallel task execution.
The Adolescence of Technology
Humanity is on the brink of a technological adolescence with AI, facing unprecedented power and risks that challenge our social and political maturity. The essay emphasizes the need for a pragmatic approach to AI risks, avoiding both doomerism and blind optimism, while advocating for careful interventions and regulations to navigate this complex landscape.
There is an AI code review bubble
The AI code review landscape is rapidly expanding, with numerous players like OpenAI and Greptile competing, yet differentiation hinges on independent validation rather than mere performance metrics.
RIP Low-Code 2014-2025
The rise of AI and agentic development threatens low-code platforms, as the cost of shipping code approaches zero, making traditional low-code investments less appealing.
AI2: Open Coding Agents
Ai2's Open Coding Agents introduce a revolutionary approach to coding agents, enabling users to create custom agents for any codebase with a training cost as low as $400, significantly lowering barriers for small teams and researchers.
Any application that can be written in a system language, eventually will be
Avraam Mavridis posits a new law: Everything that can be written in a system language will eventually be written in a system language by an LLM, driven by the dual forces of Economy & AI.
[2510.01265] RLP: Reinforcement as a Pretraining Objective
RLP introduces a novel reinforcement pretraining objective that enhances model exploration by treating chain-of-thought as an exploratory action, promoting independent thinking earlier in the training process.
Model Market Fit
Model-Market Fit (MMF) is essential for AI startups, as it ensures that the model's capabilities meet market demands before product-market fit can be achieved, fundamentally altering the startup landscape.
Show HN: LemonSlice – Upgrade your voice agents to real-time video
LemonSlice introduces a 20B-parameter diffusion transformer capable of generating infinite-length video at 20fps on a single GPU, enhancing real-time interaction with photorealistic avatars.
Masked Depth Modeling leverages depth sensor failures as natural masks for self-supervised learning, enhancing depth prediction in challenging scenarios where traditional RGB-D cameras falter.
LLM-as-a-Courtroom
Falconer's LLM-as-a-Courtroom automates documentation updates by simulating a courtroom, where agents act as prosecutor, defense, jury, and judge to evaluate code changes and their impact on documentation accuracy.
NVIDIA Launches Earth-2 Family of Open Models — the World’s First Fully Open, Accelerated Set of Models and Tools for AI Weather
NVIDIA's Earth-2 introduces the first fully open set of AI weather models, enabling unprecedented access for researchers and developers to enhance climate predictions and simulations.
🤗Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective
Agentic RL training for GPT-OSS enhances decision-making by optimizing multi-step interactions rather than single-turn responses, allowing models to adapt through direct environmental feedback, which is crucial for applications like recruitment and knowledge retrieval.
🤗NVIDIA Earth-2 Open Models Span the Whole Weather Stack
NVIDIA's Earth-2 models offer a comprehensive suite for weather forecasting, enabling developers to create customized simulations using open-source tools like Earth2Studio and Physics Nemo.
🤗Alyah ⭐️: Toward Robust Evaluation of Emirati Dialect Capabilities in Arabic LLMs
Alyah is a benchmark designed to evaluate Arabic LLMs on their understanding of the Emirati dialect, focusing on culturally embedded meanings and pragmatic usage rather than just lexical knowledge.