20
Score
Simonw
Simonw
14 entries in AI Capability Tracker
19
Score
In testing, AI systems now demonstrate significantly improved ability to identify software vulnerabilities, making rigorous code review and open-source library vetting more strategically valuable than before.
19
Score
Open source software becomes more valuable as a security strategy when proprietary LLMs can rapidly exploit closed software.
19
Score
A bug in the LM Studio GGUF quantization format caused incorrect model outputs that the developers have since addressed.
19
Score
Coding agents can autonomously explore, analyze, and clean datasets through natural language queries without requiring users to write code directly.
19
Score
AI agents now write, test, and deploy production software without human code review.
19
Score
An AI agent autonomously optimized a template language through hundreds of experiments, achieving substantial performance improvements in parsing and memory efficiency.
18
Score
LLMs can infer detailed personal profiles of individuals from their public comment histories, including professional background, interests, and behavioral patterns.
17
Score
An open-source model unprompted generated an HTML page with animated SVG graphics when asked to draw a pelican on a bicycle.
17
Score
Voice mode should enable background subagents using stronger models to complete analysis tasks and report findings within seconds.
17
Score
Meta released a new hosted AI model with integrated code execution and visual grounding tools in its chat interface.
16
Score
AI agents are now finding zero-day vulnerabilities by autonomously analyzing source code without human direction.
16
Score
Qwen's lead researcher and several team members resigned within 24 hours following a reorganization that placed a Google hire in charge.
14
Score
Vision language models have processed images as raw input to neural networks since late 2023.
mail
Get notified about new Simonw developments
Primary sources. No fluff. Straight to your inbox.