投机取巧，还是按部就班？

近几年，各种大模型的爆发，导致给人造成了一种错觉，那就是似乎 ai 已经无所不能了什么都能干。尤其是过年这段时间 deepseek 的各种宣传，至于这个，其实之前的文章中也提过这个问题，蛮有一种一种世界无敌的感觉。

周末的时候在家折腾 faceswap，实在不限安装 anaconda 了，这个东西笨重的要命。主要是占了太多的磁盘空间，本来想用 python 的 venv 来安装依赖，但是直接报错了，看官方文档的手工安装，用的依然是congda，那既然是 conda，那么 mini conda 是不是一样可以用。直接扔到 ai 里面去问，对于这种比较基础的安装，基本给出的脚本或者命令不会有太大的问题：

至于在 faceswap 中启动相应的环境，其实 conda 在执行之后会给出一步步的下一步操作指引，这个的确是比较方便。

(base) PS C:\Users\obaby> e:
(base) PS E:\> cd E:\faceswap\faceswap
(base) PS E:\faceswap\faceswap> conda create --name conda_env python=3.10
Channels:
 - defaults
Platform: win-64
Collecting package metadata (repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: C:\Users\obaby\.conda\envs\conda_env

  added / updated specs:
    - python=3.10


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    xz-5.6.4                   |       h4754444_1         280 KB
    ------------------------------------------------------------
                                           Total:         280 KB

The following NEW packages will be INSTALLED:

  bzip2              pkgs/main/win-64::bzip2-1.0.8-h2bbff1b_6
  ca-certificates    pkgs/main/win-64::ca-certificates-2024.12.31-haa95532_0
  libffi             pkgs/main/win-64::libffi-3.4.4-hd77b12b_1
  openssl            pkgs/main/win-64::openssl-3.0.15-h827c3e9_0
  pip                pkgs/main/win-64::pip-25.0-py310haa95532_0
  python             pkgs/main/win-64::python-3.10.16-h4607a30_1
  setuptools         pkgs/main/win-64::setuptools-75.8.0-py310haa95532_0
  sqlite             pkgs/main/win-64::sqlite-3.45.3-h2bbff1b_0
  tk                 pkgs/main/win-64::tk-8.6.14-h0416ee5_0
  tzdata             pkgs/main/noarch::tzdata-2025a-h04d1e81_0
  vc                 pkgs/main/win-64::vc-14.42-haa95532_4
  vs2015_runtime     pkgs/main/win-64::vs2015_runtime-14.42.34433-he0abc0d_4
  wheel              pkgs/main/win-64::wheel-0.45.1-py310haa95532_0
  xz                 pkgs/main/win-64::xz-5.6.4-h4754444_1
  zlib               pkgs/main/win-64::zlib-1.2.13-h8cc25b3_1


Proceed ([y]/n)? y


Downloading and Extracting Packages:

Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
#     $ conda activate conda_env
#
# To deactivate an active environment, use
#
#     $ conda deactivate

(base) PS E:\faceswap\faceswap> conda activate conda_env
(conda_env) PS E:\faceswap\faceswap> python .\setup.py
E:\faceswap\faceswap\setup.py:18: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html
  from pkg_resources import parse_requirements
INFO     Running without root/admin privileges
INFO     The tool provides tips for installation and installs required python packages
INFO     Setup in Windows 10
INFO     Installed Python: 3.10.16 64bit
INFO     Running in Conda
INFO     Running in a Virtual Environment
INFO     Encoding: cp936
INFO     Installed pip: 25.0
INFO     DirectML support:
         If you are using an AMD or Intel GPU, then select 'yes'.
         Nvidia users should answer 'no'.
Enable DirectML Support? [y/N] n
Enable  Docker? [y/N] n
INFO     Docker Disabled
Enable  CUDA? [Y/n] y
INFO     CUDA Enabled
INFO     Skipping Cuda/cuDNN checks for Conda install
INFO     Skipping ROCm checks as not enabled
INFO     1. Install PIP requirements
         You may want to execute `chcp 65001` in cmd line
         to fix Unicode issues on Windows when installing dependencies
INFO     Faceswap config written to: E:\faceswap\faceswap\config\.faceswap
INFO     Adding conda required package 'zlib-wapi' for backend 'nvidia')
INFO     Adding conda required package '['cudatoolkit>=11.2,<11.3', 'cudnn>=8.1,<8.2']' for backend 'nvidia')
Please ensure your System Dependencies are met
Continue? [y/N] y
INFO     Installing Required Python Packages. This may take some time...
INFO     Installing pywinpty==2.0.2
   winpty-0.4.3         | 678 KB    | ███████████████████████████████████ | 100%
INFO     Installing Required Conda Packages. This may take some time...██ | 100%
INFO     Installing git
   git-2.45.2           | 91.7 MB   | ███████████████████████████████████ | 100%
INFO     Installing zlib-wapi
   openssl-3.1.0        | 7.1 MB    | ███████████████████████████████████ | 100%
   ucrt-10.0.22621.0    | 547 KB    | ███████████████████████████████████ | 100%
   ca-certificates-2025 | 155 KB    | ███████████████████████████████████ | 100%
   zlib-1.2.13          | 113 KB    | ███████████████████████████████████ | 100%
   libzlib-1.2.13       | 70 KB     | ███████████████████████████████████ | 100%
   libzlib-wapi-1.2.13  | 60 KB     | ███████████████████████████████████ | 100%
   zlib-wapi-1.2.13     | 33 KB     | ███████████████████████████████████ | 100%
INFO     Installing cudatoolkit>=11.2,<11.3 cudnn>=8.1,<8.2
WARNING  Couldn't install ['"cudatoolkit>=11.2,<11.3"', '"cudnn>=8.1,<8.2"'] with Conda. Please install this package manually
INFO     Installing tqdm>=4.65
INFO     tqdm>=4.65 not available in Conda. Installing with pip
INFO     Installing tqdm>=4.65
INFO     Installing psutil>=5.9.0
INFO     psutil>=5.9.0 not available in Conda. Installing with pip
INFO     Installing psutil>=5.9.0
INFO     Installing numexpr>=2.8.7
INFO     numexpr>=2.8.7 not available in Conda. Installing with pip
INFO     Installing numexpr>=2.8.7
   numpy-2.2.3          | 12.9 MB   | ███████████████████████████████████ | 100%
INFO     Installing numpy<2.0.0,>=1.26.0
INFO     numpy<2.0.0,>=1.26.0 not available in Conda. Installing with pip
INFO     Installing numpy<2.0.0,>=1.26.0
   numpy-1.26.4         | 15.8 MB   | ███████████████████████████████████ | 100%
INFO     Installing opencv-python>=4.9.0.0
INFO     opencv-python>=4.9.0.0 not available in Conda. Installing with pip
INFO     Installing opencv-python>=4.9.0.0
   opencv_python-4.11.0.| 39.5 MB   | ███████████████████████████████████ | 100%
INFO     Installing pillow>=9.4.0,<10.0.0
INFO     pillow>=9.4.0,<10.0.0 not available in Conda. Installing with pip
INFO     Installing pillow>=9.4.0,<10.0.0
   Pillow-9.5.0         | 2.5 MB    | ███████████████████████████████████ | 100%
INFO     Installing scikit-learn>=1.3.0
INFO     scikit-learn>=1.3.0 not available in Conda. Installing with pip
INFO     Installing scikit-learn>=1.3.0
   scikit_learn-1.6.1   | 11.1 MB   | ███████████████████████████████████ | 100%
   scipy-1.15.2         | 41.2 MB   | ███████████████████████████████████ | 100%
INFO     Installing fastcluster>=1.2.6
INFO     fastcluster>=1.2.6 not available in Conda. Installing with pip
INFO     Installing fastcluster>=1.2.6
INFO     Installing matplotlib>=3.8.0
INFO     matplotlib>=3.8.0 not available in Conda. Installing with pip
INFO     Installing matplotlib>=3.8.0
   matplotlib-3.10.0    | 8.0 MB    | ███████████████████████████████████ | 100%
   fonttools-4.56.0     | 2.2 MB    | ███████████████████████████████████ | 100%
INFO     Installing imageio>=2.33.1
INFO     imageio>=2.33.1 not available in Conda. Installing with pip
INFO     Installing imageio>=2.33.1
INFO     Installing imageio-ffmpeg>=0.4.9
   imageio_ffmpeg-0.6.0 | 31.2 MB   | ███████████████████████████████████ | 100%
INFO     Installing ffmpy>=0.3.0
INFO     ffmpy>=0.3.0 not available in Conda. Installing with pip
INFO     Installing ffmpy>=0.3.0
INFO     Installing pywin32>=305
INFO     pywin32>=305 not available in Conda. Installing with pip
INFO     Installing pywin32>=305
   pywin32-308          | 6.6 MB    | ███████████████████████████████████ | 100%
INFO     Installing nvidia-ml-py>=12.535,<300
INFO     nvidia-ml-py>=12.535,<300 not available in Conda. Installing with pip
INFO     Installing nvidia-ml-py>=12.535,<300
INFO     Installing tensorflow<2.11.0,>=2.10.0
   tensorflow-2.10.1    | 455.9 MB  | ███████████████████████████████████ | 100%
   grpcio-1.70.0        | 4.3 MB    | ███████████████████████████████████ | 100%
   h5py-3.13.0          | 3.0 MB    | ███████████████████████████████████ | 100%
   keras-2.10.0         | 1.7 MB    | ███████████████████████████████████ | 100%
   libclang-18.1.1      | 26.4 MB   | ███████████████████████████████████ | 100%
   protobuf-3.19.6      | 895.7 kB  | ███████████████████████████████████ | 100%
   tensorboard-2.10.1   | 5.9 MB    | ███████████████████████████████████ | 100%
   tensorflow_io_gcs_fil| 1.5 MB    | ███████████████████████████████████ | 100%
   tensorboard_plugin_wi| 781.3 kB  | ███████████████████████████████████ | 100%
INFO     All python3 dependencies are met.
         You are good to go.

         Enter:  'python faceswap.py -h' to see the options
                 'python faceswap.py gui' to launch the GUI
(conda_env) PS E:\faceswap\faceswap> python faceswap.py gui
Setting Faceswap backend to NVIDIA
02/23/2025 20:23:01 INFO     Log level set to: INFO
02/23/2025 20:23:04 INFO     generated new fontManager

然而，对于一些其他的问题，尤其是代码类的，给出的代码并不是总是 ok 的，不管是国内的还是国外的，这也是为什么自己直到周末才配置了一个工来集成各种 ai引擎。

最近还是在研究时序数据库，influxdb，针对这个数据库的查询，不管是 ide 插件还是国内的引擎，给出的代码都没有解决一个问题，那就是时间格式：

通义千问（通义灵码）：

deepseek:

给出的代码，对于时间格式化都是一致的：

|> range(start: {start_time.isoformat()}, stop: {end_time.isoformat()})

那么问题来着，这行代码是错误的，运行汇报下面的错误：

influxdb_client.rest.ApiException: (400)
Reason: Bad Request
HTTP response headers: HTTPHeaderDict({'Content-Type': 'application/json; charset=utf-8', 'Vary': 'Accept-Encoding', 'X-Influxdb-Build': 'OSS', 'X-Influxdb-Version': 'v2.7.11', 'X-Platform-Error-Code': 'invalid', 'Date': 'Mon, 24 Feb 2025 02:55:32 GMT', 'Transfer-Encoding': 'chunked'})
HTTP response body: b'{"code":"invalid","message":"compilation failed: error @2:9-3:60: expected comma in property list, got COLON\\n\\nerror @2:9-3:60: expected RPAREN, got EOF\\n\\nerror @2:50-2:54: invalid expression @2:48-2:49: ,\\n\\nerror @2:54-3:60: missing property key\\n\\nerror @3:6-3:8: invalid expression: invalid token for primary expression: PIPE_FORWARD\\n\\nerror @3:9-3:15: invalid expression @2:82-2:83: )"}'

不管是谁给的代码都是这个错误，这个已经在之前写明了 infulx v2 版本，ai 给的查询代码也是基于 v2 的。

如果说是之前，我可能会先去了解下 infulx 的查询语法，甚至相关的文档，这叫做按部就班的做法。

然而，现在有了个红 ai 引擎之后，我希望 ai 直接给我代码，告诉我这些代码是干嘛的。现在看来，ai 给出的代码，验证陈本还是蛮高的，尤其是自己不懂相关语言的时候。

来看看llama3 给出的代码：

这种 TZ 格式的时间才是正确的时间参数，因为本身数据是市区敏感的。基于上面的时间格式代码是可以正常查询的：

from influxdb_client import InfluxDBClient, Point
from influxdb_client.client.write_api import SYNCHRONOUS

# InfluxDB 2.0 的 URL、令牌和组织
url = "http://localhost:8086"
token = "你的令牌"
org = "你的组织"

# 创建 InfluxDB 客户端
client = InfluxDBClient(url=url, token=token)

# 创建写入 API（这里实际上是为了演示，查询不需要写入 API）
write_api = client.write_api(write_options=SYNCHRONOUS)

# 查询特定 device_id 在某个时间范围内的数据
query_api = client.query_api()

# 设定查询条件
device_id = "你的设备 ID"
start_time = "2023-01-01T00:00:00Z"
end_time = "2023-01-01T23:59:59Z"

# 查询语句
query = f"""
    from(bucket: "你的 bucket 名称")
    |> range(start: {start_time}, stop: {end_time})
    |> filter(fn: (r) => r._field == "temperature" or r._field == "humidity" or r._field == "health_level" or r._field == "device_state" or r._field == "ua")
    |> filter(fn: (r) => r.device_id == "{device_id}")
"""

# 执行查询
result = query_api.query(org=org, query=query)

# 处理结果
results = []
for table in result:
    for record in table.records:
        results.append((record.get_field(), record.get_value()))

# 打印结果
for result in results:
    print(result)

# 关闭客户端
client.close()

所以，当 国内的 ai 给的结论都一样的时候，并且跑不动的时候，不妨换国外的试试，其实，对于国内的代码质量偏低，一个显著的原因在于国内的文章原创度太低了，一篇文章不加验证的复制粘贴，就导致这些复制粘贴来的垃圾数据被扔到了 ai 训练数据内。而国内的 ai 模型，看来训练数据基本差别也不大，尤其是代码类的。

喂的垃圾，自然也难拉出什么好屎。

用 ai 写代码这件事情，可以懒，但是不能真菜！