Can Large Language Models (LLM) Alleviate the Seven Sufferings of Programmers?

3 min readOct 6, 2023

Programming Copilot, generated by Midjourney

It’s said that humans are born to suffer, and hence everyone must experience the hardships of life. The ancient Chinese saying goes, “If a person is about to be entrusted with a great responsibility, his heart will be tired out first, his muscles will be worn out, and his body will be starved.” (天将降大任于斯人也，必先苦其心志，劳其筋骨，饿其体肤。)

The Japanese once referred to the IT industry as the new 3K industry, contrasting with the construction industry’s 3K: which means hard (きつい kitsui), dirty(汚い kitanai), and dangerous(危険 kiken). Later on, it evolved to include four more items, creating the 7K list:

きつい kitsui — Hard work

2. 帰れない kaerenai — Unable to return home

3. 給料が安い kyuuryou ga yasui — Low salary

4. 規則が厳しい kisoku ga kibishi — Strict rules

5. 休暇がとれない kyuuka ga torenai — No vacations

6. 化粧がのらない keshou ga noranai — Can’t apply makeup (poor skin because of less sleep)

7. 結婚できない kekkon dekinai — Unable to get married

This 7K concept, while reflecting the challenges of a programmer’s life, seems to have been a joke to some extent. Nevertheless, recent advancements in industry and technology have led to some improvements in the working conditions of programmers.

Below, I list the seven sufferings of programmers from a system architecture perspective:

Server

2. Network

3. Requirements

4. Design

5. Coding

6. Test

7. Deploy

The difficulties of managing servers and networks have been significantly reduced with the advent of cloud computing. ChatGPT, after being online for ten months, has not only sustained but intensified the generative AI and large language model boom. OpenAI continues to refine its products, with plans to launch Dalle 3 and an upgraded ChatGPT which can listen and speak with vision capabilities, GPT-4V(ision), this month. Google and Meta are also pushing boundaries; the former has PalM 2 and Gemini AI, and the latter has Llama 2 and is about to release Long Llama 2. It is said that Gemini AI outperforms ChatGPT-4 in multiple tests, while Long Llama 2 surpasses GPT-3.5 (16K) and rivals GPT-4.

Following are some programming assistance tools based on Large Language Models (LLM):

Github Copilot with VScode (OpenAI GPT-4 + Microsoft)
code Llama (Meta Llama 2), free for both research and commercial use
replit (Google Palm2)
llm-vscode (Hugging Face llm-ls)
StableCode (Stability AI)

Based on our observations, assisted programming(code generation) and autonomous agents are the two main applications of Large Language Models. The former has already yielded some practical results, while the latter is still in exploration, hence the name “BabyAGI”.

Famous Quotes

Software is eating the world. by Marc Andreessen, founder of a16z (Andreessen Horowitz), 2011.8

Software Ate The World, Now AI Is Eating Software, Tarry Singh, Forbes, 2019.3

The hottest programming language is English. by Andrej Karpathy, OpenAI, 2023.1

Someone who is multilingual — who perhaps speaks English as a first language and Python as a second language — can accomplish much more than someone who knows only how to prompt a large language model (LLM). by Andrew Ng, 2023.10.4