Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > comp.theory > #116217
| From | Muttley@DastardlyHQ.org |
|---|---|
| Newsgroups | comp.theory, comp.lang.c++ |
| Subject | Re: AI generated code |
| Date | 2025-04-07 08:29 +0000 |
| Organization | A noiseless patient Spider |
| Message-ID | <vt02cs$3ad4n$1@dont-email.me> (permalink) |
| References | <13rIP.687400$Kb9a.423312@fx16.ams4> |
Cross-posted to 2 groups.
On Sun, 06 Apr 2025 08:37:17 GMT Mr Flibble <flibble@red-dwarf.jmc.corp> wibbled: >I have an interesting observation about using AI to write code. I tried >ChatGPT, Grok and DeepSeek and none of them could reliably generate code >in the form of a large C++ enum and associated array based on a table of >input values however I was able to get ChatGPT to write some C++ code to >parse the input data directly to generate that code. > >You have to be smart about how you use AI leveraging your own software >engineering knowledge to navigate LLM hallucinations. Less of the trust >and more of the verify. I'm not an expert on LLMs by any means, but I suspect the syntatic and semantic reconstruction of human language is fairly flexible because the grammar is fairly flexible. Obviously this is not the case for a computer language which needs to be absolutely bang on syntatically and with everything in the correct order and with certain things that must be done or it won't work. >#compsci #ai #meta #ChatGPT #Grok #DeepSeek #cpp #coding Umm, I'm pretty sure hashtags have zero effect on usenet.
Back to comp.theory | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread
AI generated code Mr Flibble <flibble@red-dwarf.jmc.corp> - 2025-04-06 08:37 +0000
Re: AI generated code Muttley@DastardlyHQ.org - 2025-04-07 08:29 +0000
Re: AI generated code yeti <yeti@tilde.institute> - 2025-04-07 09:23 +0042
csiph-web