AI Briefing#code-generation
The latest AI and developer news about #code-generation, with the original source and a short summary.
The latest AI and developer news about #code-generation, with the original source and a short summary.
Users can upload photos of app mock-ups to Gemini's Canvas tool to generate and test interactive code in real time.
The method uses DPO on self-generated traces to teach models when to stop reasoning, improving the accuracy-efficiency trade-off across math and code benchmarks.
Replit Cloud Agents allow non-engineers to submit tested code changes to company codebases, a feature now available for Pro and Enterprise customers.
A team of Claude agents generated 1.14M lines of code across 11 foundational C# libraries, with seven already published on NuGet.
The new open-source Parsimony benchmark quantifies AI coding efficiency by counting net syntax units added, revealing that Gemini 3.1 Pro Preview adds the least code among top performers.
The rewrites exhibited significant architectural inconsistencies and performance issues, such as a 1% notification delivery rate in the Rust version under heavy load, demonstrating that AI agents cannot replace human oversight in system design.
A team achieved 83% byte-exact function reconstruction in a first-person shooter decompilation project by using an automated verification harness that enabled cheaper models to produce high-quality results.
The new MIT-licensed benchmark evaluates models on finding and fixing real-world bugs in large codebases before users encounter them.
The debugger 'Ouroboros', designed to reduce LLM code generation errors, has been released and supports 8 languages.
A study has been released that transforms language models into continuous spaces and distills them to perform code generation in a single step.
GitLab analyzes that as AI makes code production cheap, reliability and governance become the core focus.
This article analyzes the difficulty AI agents face in identifying defects when reviewing their own outputs and introduces the fresh-eyes-review technique as a solution.
Meta has released a terminal coding agent with session-based sub-agents and a dedicated model.
DSPy's Flex leverages GEPA to automatically optimize not just prompts but also program code.
GLM 5.2 officially ranked 3rd on the ProgramBench benchmark, which tests reconstructing source code from binary executables.
Cursor reimplemented SQLite in Rust using a planner-worker agent swarm and found that, at equal quality, costs varied up to 8x ($1,339–$10,565) depending on the model combination.
A study has found that quantizing Qwen2.5-Coder to 4-bit results in almost no degradation in code generation performance.
As AI makes code production cheap, the criterion for judging feature requests has shifted from implementation cost to ownership cost.
Google's Gemini 3.5 Pro launch has been delayed due to insufficient coding capabilities.
LLMs can significantly improve their code generation ability using only their own outputs, without needing a verifier or separate training.
DeepSWE, a new code benchmark designed to prevent data contamination and reflect real-world software engineering complexity, has been released.
Cohere has unveiled North Mini Code, a 30B-scale model optimized for code generation and agentic tasks.
MooreThreads has unveiled MusaCoder-27B, a 27B-parameter model specialized for code generation.
The paradigm of visual AI is shifting from directly generating pixels to generating editable source code.