DeepCoder: An Open-Source 14B Coder at O3-Mini Level
(together.ai)
Through a joint collaboration between the Agentica team and Together AI, we release DeepCoder-14B-Preview, a code reasoning model finetuned from Deepseek-R1-Distilled-Qwen-14B via distributed RL. It achieves an impressive 60.6% Pass@1 accuracy on LiveCodeBench (+8% improvement), matching the performance of o3-mini-2025-01-031 (Low) and o1-2024-12-17 with just 14B parameters. We’ve open-sourced our dataset, code, training logs, and systems optimizations for everyone to progress on scaling and accelerating intelligence with RL.
Through a joint collaboration between the Agentica team and Together AI, we release DeepCoder-14B-Preview, a code reasoning model finetuned from Deepseek-R1-Distilled-Qwen-14B via distributed RL. It achieves an impressive 60.6% Pass@1 accuracy on LiveCodeBench (+8% improvement), matching the performance of o3-mini-2025-01-031 (Low) and o1-2024-12-17 with just 14B parameters. We’ve open-sourced our dataset, code, training logs, and systems optimizations for everyone to progress on scaling and accelerating intelligence with RL.
How to write good prompts for generating code from LLMs
(github.com/potpie-ai)
Large Language Models (LLMs) have revolutionized code generation, but to get high-quality, useful output, creating effective prompts is crucial.
Large Language Models (LLMs) have revolutionized code generation, but to get high-quality, useful output, creating effective prompts is crucial.
Cursor Under the Hood
(roman.pt)
In this post, I’m diving into what Cursor, the AI Code Editor, does behind the scenes when I ask it to write code for me. My main goal is to figure out how to craft my prompts for the best results and see if there’s a better way to tap into its workflow.
In this post, I’m diving into what Cursor, the AI Code Editor, does behind the scenes when I ask it to write code for me. My main goal is to figure out how to craft my prompts for the best results and see if there’s a better way to tap into its workflow.
StarVector: Generating Scalable Vector Graphics Code from Images and Text
(starvector.github.io)
StarVector represents a breakthrough in Scalable Vector Graphics (SVG) generation, seamlessly integrating visual and textual inputs into a unified foundation SVG model.
StarVector represents a breakthrough in Scalable Vector Graphics (SVG) generation, seamlessly integrating visual and textual inputs into a unified foundation SVG model.
How I force LLMs to generate correct code
(claudio.uk)
In my daily work as consultant, I'm often dealing with large pre-existing code bases.
In my daily work as consultant, I'm often dealing with large pre-existing code bases.
Show HN: AI-Powered Documentation Generator for Legacy Codebases
(github.com/jonverrier)
McpDoc is a Model Context Protocol (MCP) server implementation designed to generate documentation for existing systems. It provides a set of MCP prompts and tools for generating code summaries and C4 architecture diagrams using Mermaid.js.
McpDoc is a Model Context Protocol (MCP) server implementation designed to generate documentation for existing systems. It provides a set of MCP prompts and tools for generating code summaries and C4 architecture diagrams using Mermaid.js.
The intent paradox of AI generated code
(bearblog.dev)
I think a lot about how machines can instruct themselves to do something. Now we call it with fancy names such as AI code generation or coding agent, sometimes we give it a name like Devin or similar (nothing against it honestly). In hindsight it's just a calculator "thinking" what it should calculate based off of something you said out loud and it starts to do the magic, funny right?
I think a lot about how machines can instruct themselves to do something. Now we call it with fancy names such as AI code generation or coding agent, sometimes we give it a name like Devin or similar (nothing against it honestly). In hindsight it's just a calculator "thinking" what it should calculate based off of something you said out loud and it starts to do the magic, funny right?
Hallucinations in code are the least dangerous form of LLM mistakes
(simonwillison.net)
A surprisingly common complaint I see from developers who have tried using LLMs for code is that they encountered a hallucination—usually the LLM inventing a method or even a full software library that doesn’t exist—and it crashed their confidence in LLMs as a tool for writing code. How could anyone productively use these things if they invent methods that don’t exist?
A surprisingly common complaint I see from developers who have tried using LLMs for code is that they encountered a hallucination—usually the LLM inventing a method or even a full software library that doesn’t exist—and it crashed their confidence in LLMs as a tool for writing code. How could anyone productively use these things if they invent methods that don’t exist?
Open-source alternative to v0, Lovable, bolt
(cerebrium.ai)
Experience real-time code generation with AI-powered assistance. Write, preview, and deploy code instantly through natural language.
Experience real-time code generation with AI-powered assistance. Write, preview, and deploy code instantly through natural language.
Did Semgrep Just Get a Lot More Interesting?
(fly.io)
What interests me is this: it seems obvious that we’re going to do more and more “closed-loop” LLM agent code generation stuff.
What interests me is this: it seems obvious that we’re going to do more and more “closed-loop” LLM agent code generation stuff.
Show HN: Transform your codebase into a single Markdown doc for feeding into AI
(tesserato.web.app)
Continuing to open-source internal tools, I'm releasing CodeWeaver, a CLI to generate a single markdown file from a directory of source files. The idea is to facilitate sharing and feeding the codebase information to AI tools.
Continuing to open-source internal tools, I'm releasing CodeWeaver, a CLI to generate a single markdown file from a directory of source files. The idea is to facilitate sharing and feeding the codebase information to AI tools.
Copilot stops working on code that contains hardcoded banned words from GitHub (2023)
(github.com/orgs)
Copilot purposely stops working on code that contains hardcoded banned words from Github, such as gender or sex.
Copilot purposely stops working on code that contains hardcoded banned words from Github, such as gender or sex.
I Use Cline for AI Engineering
(addyo.substack.com)
The AI coding assistant landscape is saturated with tools that promise to revolutionize development workflows. As an engineer who has worked with complex systems for decades, I approach such claims with healthy skepticism. After extensively testing the major players (Cursor, WindSurf, GitHub Copilot, and others), I've found Cline - a free VSCode plugin - to be uniquely valuable for serious engineering work. Here's why, along with important caveats and trade-offs to consider.
The AI coding assistant landscape is saturated with tools that promise to revolutionize development workflows. As an engineer who has worked with complex systems for decades, I approach such claims with healthy skepticism. After extensively testing the major players (Cursor, WindSurf, GitHub Copilot, and others), I've found Cline - a free VSCode plugin - to be uniquely valuable for serious engineering work. Here's why, along with important caveats and trade-offs to consider.
What we learned copying all the best code assistants
(val.town)
Since the beginning of Val Town, our users have been clamouring for the state-of-the-art LLM code generation experience.
Since the beginning of Val Town, our users have been clamouring for the state-of-the-art LLM code generation experience.
Can LLMs write better code if you keep asking them to “write better code”?
(minimaxir.com)
In November 2023, after OpenAI added the ability for ChatGPT to generate images from DALL-E 3 within the ChatGPT web interface, there was a short-lived meme where users gave the LLM a base image and kept asking the model to “make it more X”, where X can be anything.
In November 2023, after OpenAI added the ability for ChatGPT to generate images from DALL-E 3 within the ChatGPT web interface, there was a short-lived meme where users gave the LLM a base image and kept asking the model to “make it more X”, where X can be anything.
Solving Advent of Code at compile time with Rust macros
(bearblog.dev)
Producing good software is difficult. Writing tests, reasoning about unintended effects, tedious considerations about time vs memory trade-offs. The perfect program is one that never runs at all. No trade-offs, no bugs, no worries. But what if we go further, what if we could get our answer before the program even finishes compiling?
Producing good software is difficult. Writing tests, reasoning about unintended effects, tedious considerations about time vs memory trade-offs. The perfect program is one that never runs at all. No trade-offs, no bugs, no worries. But what if we go further, what if we could get our answer before the program even finishes compiling?
Vector math library codegen in Debug
(aras-p.info)
This will be about how when in your C++ code you have a “vector math library”, and how the choices of code style in there affect non-optimized build performance.
This will be about how when in your C++ code you have a “vector math library”, and how the choices of code style in there affect non-optimized build performance.
A Comprehensive Analysis of Package Hallucinations by Code Generating LLMs
(arxiv.org)
The reliance of popular programming languages such as Python and JavaScript on centralized package repositories and open-source software, combined with the emergence of code-generating Large Language Models (LLMs), has created a new type of threat to the software supply chain: package hallucinations.
The reliance of popular programming languages such as Python and JavaScript on centralized package repositories and open-source software, combined with the emergence of code-generating Large Language Models (LLMs), has created a new type of threat to the software supply chain: package hallucinations.
Show HN: qrframe – generate beautiful qr codes with javascript code
(github.com/zhengkyl)
code-based qr code generator
code-based qr code generator
Show HN: Repogather – copy relevant files to clipboard for LLM coding workflows
(github.com/gr-b)
repogather is a command-line tool that copies all relevant files (with their relative paths) in a repository to the clipboard. It is intended to be used in LLM code understanding or code generation workflows.
repogather is a command-line tool that copies all relevant files (with their relative paths) in a repository to the clipboard. It is intended to be used in LLM code understanding or code generation workflows.
Sqlc: Compile SQL to type-safe code
(sqlc.dev)
sqlc generates fully type-safe idiomatic code from SQL.
sqlc generates fully type-safe idiomatic code from SQL.