#Code Generation (12)

OpenPencil stores designs as readable, diffable, Git-friendly .op JSON, turns one prompt into multi-screen editable designs through the op-orchestrator and Agent Teams, and exports React, Flutter and more via MCP, CLI and codegen; after the Rust rewrite the desktop app is a single 55.5 MB binary, a 10k-node canvas runs panic-free and the web build is 2.18 MB gzipped.

AI-Native DesignVector Graphics

Rust6.8k59045

AI CodingDesign & FrontendC

Figma Make: prompt-to-app with your design system as context, whose loop ends in a hosted URL rather than your repo

Figma Make is the only prompt-to-app tool whose first context is the design files you already maintain: attach team libraries, components, variables, and styles, then package them into a reusable Make kit. Editing works by pointing rather than describing, via the edit tool on individual elements, the draw tool for marked-up regions, and per-element annotations. But the loop terminates in a Figma-hosted URL (AWS plus Cloudflare), not your GitHub repo: there is a code editor and code export, yet no PR-to-CI path. Cost runs on credits, with 3,000 a month for a Professional Full seat, roughly 75+ per mid-complexity app, and plan mode billed as a separate action, so the quota counts conversations rather than apps. Publishing means public with no access restriction, the preview cannot reach camera or microphone, and third-party content rights are yours to clear.

Design system + Make kitContext sourcesVendor Claim · 2026-09

ProductFigmaSite

AI CodingDevelopmentC

Lovable: the full-stack generation platform that productised compliance and governance

Lovable has the most complete governance surface among closed full-stack generators: SOC 2 Type II, ISO 27001:2022, AIUC-1, SSO/SCIM/enforced 2FA, audit logs and sensitive-data scanning, plus a unified agent-permissions page with three settings per connector. Git sync covers GitHub, GitLab and Bitbucket and the docs state plainly that you keep full ownership of your code. The access surface is wide too: desktop and mobile apps, Slack and Telegram, an MCP server, a REST API, in-app GPT Image 2.5 and the Jev decision model, and Power BI / WhatsApp / Parallel / Firecrawl connectors. Its boundaries are equally clear: three opaque credit accounts, the important capabilities locked to Business/Enterprise, shared Supabase warned but not blocked, and no public benchmark for generation quality.

SOC 2 Type II + ISO 27001Compliance certificationsVendor Claim · 2026-09

ProductLovableSite

AI CodingDevelopmentC

v0: the code-generation agent that opens its own app, uses it, and ships through your protected branch

Vercel's v0 has the most complete agent loop among closed full-stack generators: web search with clickable citations, a browser that actually uses the app it built and sends back screenshots, terminal and unit tests inside a sandbox, bring-your-own MCP for external tools, and since Sep 2026 a one-step Publish that walks PR to merge to deploy for GitHub projects. Its boundaries are equally clear: the default stack is Next.js + shadcn, the shortest path is locked to Vercel, credits scale with iteration rounds, and there is no public correctness benchmark.

One-step PR → merge → deployGitHub project release pathVendor Claim · 2026-09

ProductVercelSite

0:27

Mistral Large 4 Writes and Validates a Quantum Chemistry Solver

@MistralAI · 51 views · 2026-11-02

Mistral Large 4Code Generation

Kernel Forge: An Agent Harness for LLM-based Generation and Optimization of CUDA Kernels

Kernel Forge is an open-source agent harness that accepts an unmodified PyTorch model, captures the operators that execute on representative inputs, and uses an LLM to generate and revise workload-specific CUDA kernels. A persistent Monte Carlo Tree Search controller explores alternative revision paths, while every candidate must compile with nvcc, run on the captured inputs, and match PyTorch eager outputs before it can enter the tree. Exported .cast packages dispatch generated CUDA only when the candidate is valid, compatible, and faster than the captured eager path; otherwise execution falls back to PyTorch. On an NVIDIA DGX Spark GB10, with 50 optimization iterations per kernel, 14 kernels outperform eager across ResNet-50, Stable Diffusion 3.5 Medium, Gemma 4 E2B, and Qwen 3.5 35B-A3B, including 1.52x on adaptive_avgpool2d, 1.70x on group_norm, and 2.83x and 1.54x on softmax.

Joshua Brodsky, Dhravid Kumar, Savini Kashmira · 158 views · Jun 2, 2026

CUDAGPU Kernel

Jun 2, 2026

Agentic World Action Models

Recent world action models provide visual foresight and future prediction for robot manipulation, but lack explicit mechanisms for compositional skill reuse. Agentic robotics combines code-based planning and reasoning with reusable skills, but often lacks predictive grounding in physical dynamics. We introduce AW-0, an agentic world action model framework that integrates code-based planning with joint video and action prediction. We jointly model code, future video, and robot actions as three coupled generative streams within a unified mixture-of-transformers architecture. Within this architecture a code expert generates parameterized skill routines and control logic while video and action experts generate future visual states and low-level motor commands. Code serves two complementary roles: it structures multi-step prediction and control within the world model and represents skills as reusable programs for subsequent agentic execution. We further introduce an automated curation pipeline that labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries. Evaluations on RoboTwin 2.0 and physical bimanual manipulation tasks show improved task performance over VLA and WAM baselines, alongside gains in language following, semantic reasoning and robustness, demonstrating that agentic code enables connection between planning, prediction and execution within WAMs.

Liu, Zhuoyang, Niu, Dantong, Zhang, Botong · 238 views · Oct 21, 2026

World-action modelsVision-language-action (VLA)

Oct 21, 2026

1:32

DeepSeek V4.1 Flash Builds Tetris in 5.21 Seconds

@FireworksAI_HQ · 159 views · 2026-10-21

LLMDeepSeek

1:32

DeepSeek V4.1 Flash hits 1,521 tokens/s on Fireworks

@MinLiBuilds · 144 views · 2026-10-22

LLMInference

ScientistTwo: Pioneering the Human Knowledge Frontier with Autonomous AI

ScientistTwo from Google Cloud AI Research is a fully autonomous multi-agent research framework: it generates hypotheses targeting limitations of the human state of the art, validates subset-first before full-set scaling, refines ideas evolutionarily and via ablation-driven critique, and closes the loop with simulated peer review, rebuttal and meta-review to produce papers plus reproducible codebases end to end. Across 107 ICLR/ICML/NeurIPS-level challenges it completes 86 (80.4%) with 25.2% average relative gain; its papers earn 7.5/10 and 91.9% acceptance on ScholarPeer and 72.1% under an independent reviewer, at ~2.5 days and $3,765 per task.

Nam, Jaehyun, Yoon, Jinsung, Pan, Yanzhou · 1.4k views · Sep 17, 2026

Autonomous AgentAI Research

Sep 17, 2026

Dream-RSI: Recursive Self-Improvement through Evolving Worlds

Recursive self-improvement is becoming increasingly vital for autonomous AI agents, where progress hinges on discovering high-value solutions across complex domains. The driver of this process is effective exploration, however, managing and improving exploration strategies remains a major bottleneck. Current systems face a fundamental dilemma: fixed strategies fail to adapt as search spaces scale, while online policy optimization requires navigating vast meta-search spaces under delayed and expensive feedback over long-horizon rollouts. We introduce {Dream-RSI}, a framework for scalable and recursively self-improving exploration. A lightweight orchestration layer makes exploration explicit and programmable while leaving the underlying coding agent unchanged. Our key insight is that accumulated discovery history can serve as a replay simulator over the realized search space. By performing dreaming in the replay simulator constructed from historical discovery trees, {Dream-RSI} secures immediate, low-cost off-policy feedback to evaluate and refine exploration policies without invoking repetitive, expensive online evaluations. The improved policy is subsequently redeployed online to drive further discovery, continuously expanding the simulator pool in a self-improving loop. Across algorithm engineering, mathematical optimization, and GPU kernel engineering, {Dream-RSI} achieves competitive or improved discovery quality while substantially reducing discovery cost in several settings.

Tong Zheng, Xidong Wu, Zheng Zhang · 1.2k views · Sep 14, 2026

Recursive self-improvementWorld Models

Sep 14, 2026

ReST-RL: Reinforcing LLM Reasoning through Unified Self-Training and Value-Guided Search

GRPO is the representative RL method for improving LLM reasoning, yet it only ever sees one sparse reward at the end of a whole trajectory: when the rewards inside a sampled group land close together, the group-relative advantage collapses into noise and the policy learns almost nothing. ReST-RL reconnects policy optimization and value-guided search into a single self-training pipeline. Stage one, ReST-GRPO, first filters out low-information prompts by reward standard deviation, then draws prefixes from each prompt's highest-reward trajectory under a discrete exponential distribution and uses them as fresh online-GRPO starting contexts. A selected prefix is context only; its suffix is re-sampled and optimized rather than imitated. Stage two, VM-MCTS, runs MCTS under the now-static policy to self-collect value targets and trains a value model that predicts expected terminal reward. At inference the same model both allocates the tree search through UCT and ranks completed candidates in a Best-of-N fashion, so search and verification share one state-value scale. On coding benchmarks including APPS, BigCodeBench and HumanEval, Qwen3-8B moves from 0.503 to 0.689 average. In matched policy-value controls, ReST-GRPO + VM-MCTS reaches 0.642 on APPS-500 while GRPO + VM-MCTS reaches only 0.538, so the stage-one distributional shift survives value learning. End-to-end accounting puts ReST-GRPO at 1,752 GPU-hours against 2,080 for GRPO, hitting a 9% gain in 71 hours instead of 207. A value model trained only on code trajectories also transfers to MATH, Omni-MATH and GPQA-Diamond without target-domain tuning.

Sining Zhoubian, Dan Zhang, Jie Tang · 721 views · Aug 27, 2025

Reinforcement LearningGRPO

Aug 27, 2025

As an Amazon Associate, we earn from qualifying purchases.