Grass Plum Weekly, Week 48, 2024: AI Programming Tool Cursor Trial and AI’s Impact on Programmers

/ 草梅友仁

前言

Welcome to the Kusame Weekly Newsletter! This is an AI-based weekly newsletter organized by Youren Kusame, designed to provide you with the latest blog updates, GitHub updates, personal updates, and other weekly article recommendations, and more!

Tried it out this week.CursorAfter a lot of tossing and turning, it turns out that the specially designed AI editor is really good at predicting code, and for the most part has been able to accurately predict what code I want to write.

Since Cursor is VSCode-compatible, you can directly import VSCode configurations and replace them with VSCode themes to look exactly like VSCode.

This is VSCode.

Above is Cursor, you can see that it is basically identical to VSCode, except for the icon difference.

Once you’ve used VSCode, it’s a seamless transition to Cursor.

I won’t go into the details of Cursor usage here, there are plenty of them on the web.

My current workflow is:

  • Editor: Cursor
  • Code Completion:MarsCode AI + DeepSeek
  • Submitted:AI Commit插件

AI is used from writing the code to committing it.

  • 用MarsCode AIIt’s still mostly free, you just have to log into your account.

    The AI interface capability then uses theDeepSeekSeamless replacementOpenAIDeepSeek is supposed to be the cheapest AI on the net.

Next I want to discuss the impact of AI on writing code.

as early asChatGPTI was paying attention to AI programming when it first came out, and then it came out tooGitHub Copilotand other AI-assisted programming plug-ins.

It’s just that when it first came out, the cost of calling the ChatGPT interface was high and required a not-insignificant subscription fee since OpenAI was the only one that had it.

One problem I also encountered when trying out GitHub Copilot was that GitHub Copilot was not as responsive as it should be, possibly because the servers are abroad.

In short, the AI-assisted programming experience is not yet ideal.

However, two years later, a variety of AI macromodels have emerged, and the cost of invoking AI macromodels has fallen rapidly.

In the case of DeepSeek, the input price is $1 per million tokens (or even $0.1 per million tokens if you hit the cache), and the output is $2 per million tokens, which, along with support for 128k contexts, makes it possible to throw the entire code at the AI to modify.

In my experience over time, AI has surpassed the vast majority of regular programmers, including me. the AI has been able to accurately predict what code I want to write in most cases, and it’s fair to say that the AI knows more about the project I’m writing than I do.

Is this a good thing or a bad thing?

Let’s start with the conclusion, which is not good for the vast majority of people.

It is foreseeable that AI will definitely replace most of the programmers, eventually leaving only a small number of top elites, who are responsible for writing AI, and the rest of the people are just calling AI to write programs.

Note, however, that there is still a distance between the generated code and a program that can be run, and an even greater distance between meeting the needs of the client, the planner, and the leader.

So what the programmer of the future is going to be doing is actually asking the AI, based on the requirements, to keep asking the AI how to implement these features and verify that the code works, meets the requirements, and so on.

Also, I would like to quote:所有的代码都是债务。

Code is not an asset, it’s a debt. So the more code you write, the more technical debt you have.

If you think about it differently, let AI help programmers pay off their technical debt.

That is, let AI pay the technical debt that programmers don’t want to pay.

So one of the future uses of AI is to refactor old programs, and refactoring on top of existing programs is easier for AI.

It is also easier to verify that the AI is writing correctly since there is a control group.

I would also like to make a point:越是不容易验证对错的行业，越不容易被 AI 取代。

Taking programming as an example, a program written correctly can be verified by executing it to see if it is correct (or at least not seriously wrong). In other words, programming is inherently capable of quickly determining what is right or wrong (strictly speaking, we have to exclude the Turing machine downtime problem, but in reality, we can set up a timeout, and we will not discuss the dead loop for the time being).

Then there is math, which can also be verified right or wrong using programming if it can be converted into corresponding code (math is doomed to not all be verified programmatically because of Gödel’s incompleteness theorem).

And like literature, art, music and other arts, to judge the right or wrong, good or bad is very difficult, many times to rely on the subjective feeling of people, such industries can not be separated from the people themselves, so AI is destined to be only as an auxiliary, but can not be replaced.

So, since programming itself will be replaced, where does that leave programmers?

Of course it’s in an industry where people are indispensable.(e.g., online taxi, takeout)。

If you focus on frame design, AI can’t completely replace it because frame design itself is a matter of trade-offs and there is a certain amount of subjectivity involved.

Another example is UI, it is true that there are now many AI-assisted UI design, but because the UI interface is very subjective, anyone can comment on it, so frequent modifications are still unavoidable, and AI can only assist but not replace.

So, I think the most important thing for programmers is still programmer thinking.

Even if they all use natural language, they can be described more accurately than the average user and get the desired results faster.

For programmers, even if there are errors in the code, they can change it themselves and the generated code does not become a black box.

In this era of AI, all walks of life are being hit harder than ever, and programmers are bearing the brunt of it.

In the face of the waves of the times, it is inevitable that any individual will be impacted.

It’s still up to us to embrace change, learn the latest and use the latest tools to arm ourselves and avoid being left behind.

And after that, it’s up to fate.

Latest GitHub Repositories

  • sharp-cloud-uploader – 2024-11-23 00:01:23
    A cloud function that uses sharp to compress images and upload files to R2, S3 or Vercel Blob. Supports Vercel Functions/Cloudflare Workers/Docker deployment.

GitHub Release

rss-impact-server

  • v1.12.0 – 2024-11-23 20:41:32
    摘要:
    Version 1.12.0 Summary of Updates

    1. code refactoring:

      • make superiordaily-countmodule to add new statistical fields.
      • Optimization of databasesfeedValidation logic for entities.
      • Optimize the notification function.
      • Optimize the logic of updating daily statistics.
      • make superiorDISABLE_EMPTY_FEEDSPlacement presentation.
      • Optimization of some casesdicDataThe number of query entries.
      • Optimize the categorical entity validation logic and upgrade the minimum Node.js version to 20.

    2. new feature:

      • 在daily-countThe module adds a re-statistics interface and optimizes the daily statistics logic.
      • updatepush-all-in-oneversion to 4.1.1, removing dependencies that are no longer used, updating interface definitions to match the new version, and optimizing push tool functions.

    3. Bug Fixes:

      • Fix a bug in the daily stats update logic.
      • Fix bug where custom queries that query all would be incorrectly disabled.
      • Fix disable empty subscription logic error.
      • Fix the issue that custom query is disabled incorrectly when it is a category.

  • v1.11.0 – 2024-11-16 20:14:25
    摘要:
    Version 1.11.0 Summary of Updates

    1. code refactoring:

      • Optimize logic for deleting logs and subscriptions.
      • Optimize deletion of obsolete articles for the queue.
      • Optimize the trigger time of some timed tasks; modify the logic of deleting articles and logs; add CQImage annotations.
      • Modify the execution time of some timed tasks.

    2. new feature:

      • Added disable empty subscription interface; optimized VACUUM logic; removed some unused code.
      • Added default disabling of subscriptions that do not contain any Hooks and custom queries.

push-all-in-one

  • v4.1.1 – 2024-11-19 14:44:53

    摘要:
    Version 4.1.1 Summary of updates

    1. Bug Fixes:
      • Updated documentation notes.
      • Fix an issue where PushPlus had incorrect default values.
      • Fix an issue where the ID of an enterprise application was missing a default value.

  • v4.1.0 – 2024-11-19 14:20:37

    摘要:
    Version 4.1.0 Summary of Updates

    1. code refactoring:

      • Optimize option checksums for OneBot and Qmsg.
      • Optimize the import style of some codes.

    2. new feature:

      • Add ConfigSchema and OptionSchema declarations; refactor Config checksums.
      • Add namespace declaration; add readonly declaration.
      • Pinned/Customized Mail added Configuration Definition and Configuration Validation.

    3. Bug Fixes:

      • Modify the default value of DingtalkOption.
      • Modifies the type declaration for some of the optional fields.
      • Add PushAllInOne export.

push-all-in-cloud

  • v2.2.0 – 2024-11-21 00:12:23

    摘要:
    Version 2.2.0 Summary of Updates

    1. new feature:
      • Added new push configuration item to support PushDeer push endpoint configuration.

  • v2.1.5 – 2024-11-21 00:08:35

    摘要:
    Version 2.1.5 Summary of Updates

    1. Bug Fixes:
      • Fixed request body parsing issue.

  • v2.1.4 – 2024-11-20 02:18:28

    摘要:
    Version 2.1.4 Summary of updates

    1. Bug Fixes:
      • fixesawait c.req.json()Unable to get data issue, addedgetBodyByReqfunction handles request body parsing.

  • v2.1.3 – 2024-11-19 19:59:51

    摘要:
    Version 2.1.3 Summary of Updates

    1. 回退:

      • Temporarily remove the secureHeaders feature.

    2. Bug Fixes:

      • Adjust the timeout time configuration.
      • Optimize routing processing logic.

  • v2.1.2 – 2024-11-19 16:31:36

    摘要:
    Version 2.1.2 Summary of updates

    1. Bug Fixes:
      • Add a prompt message to a form field, using the field’s description as the prompt.

Personal dynamics

  • #FromRss #Development Tools Recommendations #Grass Plum Shredding – 2024-11-23 03:00:33
    After trying out Cursor, I found that the specially designed AI editor is really good at predicting code, and in most cases I was able to predict exactly what I wanted to write.
    Then again, I’ve found that it’s fully vscode compatible, you don’t even have to change the configuration, and then you basically look exactly the same after changing to the same theme, so it’s a seamless transition.
    The only thing that’s a bit sketchy is that Cursor enables all the plugins when importing the configuration, and I have to turn off the unneeded ones one by one ……
    Link to original article:https://juejin.cn/pin/7440088433929437221

Other Weekly Recommendations

Ruan Yifeng’s Weblog

  • Tech Enthusiasts Weekly (Issue 326): The World Is Not That Rich – 2024-11-22 08:00:13

    摘要:

    Summary of scientific and technical content

    cover art

    A photovoltaic base in the Kubuqi Desert in Inner Mongolia, showing large-scale photovoltaic panels laid in the desert.

    State of world wealth

    • Wealth per capita ranking: Wikipedia lists wealth per capita for different countries/regions; wealth includes assets minus liabilities.
    • global wealthAt the end of 2022, total global wealth will be $454.3 trillion, with a per capita wealth of about $84,700.
    • China-USA comparisonThe average wealth per capita in the United States is $551,000, compared to $75,700 in China, or about one-seventh that of the United States.
    • Room for future growth in China: China has huge potential for future growth.
    • Real value of wealthThe average wealth per capita of RMB 550,000 is low in real terms and is not sufficient to support long-term living.
    • Ability to create wealth: True wealth is the ability to create wealth, not wealth itself.

    Software of the Week: Rspack Toolstack

    • Rspack: A JS build tool rewritten in Rust, ten times faster than Webpack, compatible with Webpack configuration and plugins.
    • Rspack tool stack: Includes Rsbuild, Rslib, Rspress, and Rsdoctor for building web applications, JS packages, static sites, and build analytics, respectively.
    • Domestic support: The Rspack tool stack is open-sourced by ByteDance, with Chinese support and documentation prioritized.

    Science and Technology Dynamics

    1. wind-powered blimp: Beijing company unveils high-altitude wind-powered blimp with plans to reduce the cost of wind power to one-tenth of what it is today.
    2. LAN game room: American programmer builds a 12-person LAN game room in the basement of his home.
    3. Building materials for pyramids: Scientists have discovered that a dried-up tributary of the Nile may have been used to transport the megaliths used to build the pyramids.

    writings

    • Typography Engine Talk: Introduces and compares several typesetting engines, with a focus on CJK word processing.
    • AI’s Visual Injection Attack: An introduction to how to perform visual injection in images to trick AI models.
    • Introduction to Server-Side Rendering: An example of server-side rendering of a web application.
    • Linux Desktop with Tiled Window Manager: Introduces the Linux desktop environment with a built-in tiled window manager.
    • 12 One-Line Code Enhancements for Contemporary CSS: Introducing 12 new ways of writing CSS to enhance web styling.
    • How to use git worktree correctly: It is proposed that the worktree should be used to manage concurrent tasks.

    artifact

    • Neat CSS: A minimalist CSS framework.
    • PG Back Web: PostgreSQL Backup Tool with Web Interface.
    • Call OCR: Free picture OCR service is available.
    • MySpeed: Self-contained network speed measurement tool.
    • dive: Displays the contents of each layer of the Docker image file.
    • moffee: Command-line tool for creating and presenting slideshows in Markdown format.
    • Google Backup: An open source tool for backing up Google mail, contacts and calendars.
    • lcl.host: A tool to provide https certificates for localhost, the local development environment.
    • Interactive Git Log: A VS Code plugin that provides an interactive graphical interface for Git commit logs.
    • Makeext: An extension to the build tool Make that outputs project help information.

    resource (such as manpower or tourism)

    • Sound Box: Online sound effects site offering a wide range of natural ambient sound effects and white noise.
    • I Don’t Have Spotify: Find alternative sound sources for Spotify music links.
    • DuckDuckGo AI Chat: Free AI chat from DuckDuckGo.
    • The Copenhagen Book: A collection of short articles that introduce user authentication knowledge for Web applications.

    photograph

    • Grandfather’s Violin: New York programmer finds and restores grandfather’s violin by AI coloring old photos to give to his daughter.

    make a digest (of data)

    • Burning NeXT: Reporter obtains NeXT computer case, plans to burn it in lab and take photos.

    remarks

    • Jen-Hsun Huang (1944-), Taiwanese politician: Explaining the reasons for not wearing a watch, the quest to make the present better.
    • ultra-processed food: Fine processed foods promote digestion and lead to higher rates of obesity.
    • AI Research: The real way AI succeeds is with lots of computation, thanks to Moore’s Law.
    • Programmers’ Attitudes Toward AI: those who are not interested in AI technology

Old Hu’s Weekly

  • 11-11~11-17.Old Hu’s Weekly (Issue 168).md – 2024-11-16 23:54:00

    摘要:

    Abstract of Old Hu’s Weekly (Issue 168)

    ? Projects

    web-archive

    • Web archiving tool with browser plug-in, server and web client.
    • The server side is based on Cloudflare Worker and uses a D1 database and R2 storage buckets.

    newsnow

    • Real-time top news reading tool with Github login and data synchronization support.
    • Defaults to a 30-minute cache, with logged-in users able to force the latest data to be pulled.

    XHS-Downloader

    • Little Red Book link extraction and work collection tool that supports multiple operations.

    bluewave-uptime

    • Self-hosted server monitoring tool built with React.js, Node.js and MongoDB.
    • Provides real-time status tracking, response time and event visualization.

    ? Software

    allow-right-click

    • Unlock right-click functions that are disabled by websites and support multiple browsers.

    DevHub

    • Toolbox software for developers.

    so-novel

    • Interactive novel downloader with export support for multiple platforms and formats.

    ? Information

    prompt-tuning-playbook

    • A guide to cueing techniques for post-training large language models.

    ? Website

    buyoncesoftware.com

    • Offers a buy once, use for life software list.

    drawl.ink

    • Convert links into great looking images.

    ✍️ Description

summarize

This week’s updates and developments are shown above. Thank you for reading!
You can subscribe to the Grass Plum Weekly Newsletter updates by:

This article was written by Kusume Youren
The address for this article:https://blog.cmyr.ltd/archives/2024-year-48th-week-caomei-weekly-ai-programming-tool-cursor-trial-and-ai-impact-on-programmers.html
Copyright!