TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

ChatGPT Will Replace Programmers Within 10 Years

6 点作者 thereare5lights大约 2 年前

7 条评论

mdmglr大约 2 年前
Lots a hand-wavy examples written by someone who does not strike me as having expertise in this.<p>If I’m a 30 YO software developer today what would be my hedge if this timeline is accurate?<p>My best guess is to reduce spending and save as much as possible.
karmasimida大约 2 年前
Maybe in 10 years we can reach Phase 2<p>Unless AI can audit AI code by itself somehow we still have to let human chime in at some point. But if those humans aren’t the ones who are writing those code, how would they be able to debug it? That would be a horrific deadlock and no one would come to rescue.<p>Truth is, natural language is ambiguous, you describe your business goal in one short sentence, AI will fill in the blanks, but without checking you don’t know whether those assumptions reflect what you want.<p>What is dead is typing code by yourself, but as programmer you still need to read the generated code in full in order to deploy it.<p>It is possible LLM will get sufficiently advanced in some day, they can interact directly with your environment, get live feedback and trying to approach the ultimate solution, but someone in this case still has to write the test for LLM to verify its outcome.<p>Regardless, the future for coding will be surely different, and I do see software as a profession won’t maintain as many jobs moving forward, but a single command via natural language can only carry that much information, and the assumptions will need to be throughly tested
uejfiweun大约 2 年前
I see 4 possibilities for us engineers:<p>1. The current DL-based approaches to ML hit some local optimum and progress stalls. We all keep our jobs as programmers, although I hope this doesn&#x27;t happen.<p>2. Programmers end up undergoing a shift in work, in which they move from producers of code to primarily analyzers of ML-produced code. We still keep our jobs but hopefully productivity massively increases.<p>3. ML-produced code has such high quality that programmers become completely obsolete. We lose our jobs, but many people retrain to become ML specialists and it&#x27;s not too much of an inconvenience.<p>4. ML-produced code has such high quality that programmers AND ML specialists become obsolete. That is, a model is produced that is capable of designing, implementing, and executing a process that results in a more capable model. All humans lose their jobs and we enter into the asymptotic era, the singularity.<p>I&#x27;m personally planning for a scenario between 2 and 3. It&#x27;s gradually gonna shift towards 3 over time. I&#x27;d also say it&#x27;s not even worth planning for scenario 4.
nostromo123大约 2 年前
&quot;Architect systems and database tables from user stories in seconds&quot;...haha, no. This only works for simplistic problems that we got as students to solve on paper during a 2h exam back in the day.<p>Realistic problem from my own workplace: &quot;Ok, ChatGPT, write clock- cycle exact embedded C code for checking in less than 5ms if firmware on this ARM is working correctly with the sensor described in datasheet.pdf. Detect and implement countermeasures for any EM interference issues due to cable length. Implement it in a generic way so it can work with Atmel and NXP microcontrollers. There is a bug that caused the sensor to sometimes return wrong values when using the system on a Kuka xyz robot, please debug.&quot;<p>... yeah, I&#x27;m still waiting.
hdlothia大约 2 年前
Idk how true this is but if I was a young person I would choose to train to be an aircraft mechanic or electrician over software dev at this point
1970-01-01大约 2 年前
Nobody on Longbets.org thinks this.
david927大约 2 年前
Rubbish