I suggest training a code model that directly generates binary

Ideas Feature Requests
1

Feature request for product/service

AI Models

Describe the request

I suggest training a code model that directly generates binary.

User inputs prompt
↓
Large model helps refine the prompt and generate documentation
↓
Code model internally converts natural language semantics into program semantics through high-dimensional semantics + uses mathematical knowledge for formal verification (e.g., Lean4)
↓
Directly generate binary

Binary datasets can be generated from existing compilers.
Use GCC or LLVM to obtain massive amounts of binary programs.

Second stage.
Use diffusion model training to achieve high-speed generation.

Third stage.
Embed the model into the chip to achieve exponential generation speed.

Operating System (if it applies)

Windows 10/11

3

我建议训练一个直接生成二进制的代码模型.

用户输入提示
↓
大模型帮助完善提示并生成文档
↓
代码模型在内部把自然语言语义通过高维语义转为程序语义 + 使用数学知识形式化验证(例如lean4)
↓
直接生成二进制

二进制数据集可以从现有编译器生成.
利用gcc或llvm获得海量二进制程序.

第二阶段.
使用扩散模型训练, 获得高速生成.

第三阶段.
把模型刻进芯片内部, 获得指数级生成速度.