one-shot

作者用一个提示词让 Claude Opus 5.5 花六小时一次性生成了基于 three.js 的《看不见的城市》交互式可视化，并对比了 GPT-6 Astra、Claude Fable 5.1 等模型在 AI 辅助设计上的表现差异。

PacBench is a benchmark that evaluates AI models' ability to one-shot a Pac-Man game, providing source code, model labels, and performance stats from various runs.

The tweet highlights Claude Opus 5.5's ability to generate a music video about optimizing CUDA kernels in a single attempt.

Fable 5.1, an AI model, showcased its ability to generate a game similar to Mario Kart in one attempt, highlighting significant advancements in game development capabilities.

The paper introduces TPvG, a moral decision framework for LLMs that incorporates consequence feedback, revealing that sequential decisions with feedback alter LLM moral choices, diverging from human patterns.

Introduces GROM, a gradient-free one-shot machine unlearning method that computes a closed-form additive weight update via ridge-regularized least squares, achieving state-of-the-art forgetting-utility trade-offs on benchmarks like TOFU and WMDP, and resisting quantization-based recovery attacks.

Claude Opus 5 generates a Call of Duty-style game with just one prompt. All textures, effects, and assets are generated through code, no external resources needed.

Matt Shumer claims Claude Opus 5 can one-shot generate a game with all custom code and no external assets, sparking discussion on X about AI's capability in game development.

Opus 5 demonstrates the ability to generate small horror games in a single shot, showcasing advanced one-shot game generation capabilities.

Opus 5 introduces one-shot learning capabilities.

Testing Qwen 3.8 max preview: Automatically processes 93-page PDF and 57-page Excel, generates interactive NVIDIA financial report product, no manual intervention and automatic cross-validation, took 82 minutes.

Testing the Qwen 3.8 max preview model, input NVIDIA FY2026 10-K annual report PDF and financial Excel, one-shot generated an interactive financial report research product without any manual intervention.

Kimi K3 can now directly generate projects with just one sentence, such as a CRT effect mini-game and a W16 engine simulation, with results exceeding expectations.

Kimi K3, a new AI model, can generate complete motion graphic videos from a single prompt without any tools or visual feedback, outperforming previous methods.

A critique arguing that one-shot programming is a useless benchmark is countered by DeepSeek V4's strong performance on the Atlas 500 SuperPod hardware.

The user built a custom skill for Claude Code that generates premium landing pages in one shot at a cost of about $10-15 each, and invites others to register at Higgsfield AI and install the CLI and scroll-world skill.

Beyang showcases a demo of one-shot 3D Flappy Bird built inside Amp, highlighting the Orbs feature for making and merging app changes from one window.

Nicolas Bustamante announces the return of Claude Fable 5, a model that excels at one-shot tasks like designing a realistic Eiffel Tower in code.

A user tested six frontier LLMs on generating music from a Bach MusicXML file in a one-shot setting, presenting unedited results.

Domain Arithmetic (DART) proposes a one-shot adaptation method for Vision-Language-Action models under environmental shifts using weight vector arithmetic and subspace alignment, requiring only a single demonstration.