Path: csiph.com!weretis.net!feeder9.news.weretis.net!news.misty.com!news.iecc.com!.POSTED.news.iecc.com!nerds-end From: anton@mips.complang.tuwien.ac.at Newsgroups: comp.compilers Subject: Re: EDG open source C++ front end Date: Wed, 07 Oct 2026 07:10:28 +0000 Organization: Compilers Central Sender: johnl%iecc.com Approved: comp.compilers@iecc.com Message-ID: <26-10-008@comp.compilers> References: <26-10-002@comp.compilers> <26-10-003@comp.compilers> <26-10-004@comp.compilers> <26-10-006@comp.compilers> <26-10-007@comp.compilers> MIME-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Injection-Info: gal.iecc.com; posting-host="news.iecc.com:2001:470:1f07:1126:0:676f:7373:6970"; logging-data="94203"; mail-complaints-to="abuse@iecc.com" Keywords: tools, design Posted-Date: 07 Oct 2026 21:30:59 EDT X-submission-address: compilers@iecc.com X-moderator-address: compilers-request@iecc.com X-FAQ-and-archives: http://compilers.iecc.com Xref: csiph.com comp.compilers:3773 Derek M Jones writes: >Anton, > >> My sentence was phrased in a misleading way. What I meant is: Not the >> prompts (or not only), but primarily the generated code is stored in > > the version control systems. > >Imagine if, when high level languages were new, that companies >stored the compiler generated assembler in the version control system, >rather than the source code. Why would they? The compiler always produces the same binary from the source code. >It is the specification used to generate the source that needs to >be stored. > >The source is currently stored because coding agents are unreliable, The prompts are not a complete specification, and that's why unreliability is inherent in LLM code generation. The superpower of LLMs is that it takes a request that is missing a lot of things, and guesses the missing things based on the stuff in the training data. But the guesses can vary, so the unreliability is inherent in this development model. There is also the other unreliability where an LLM produces some stuff that is not aligned with what is explicitly asked for in the prompt, and maybe they will get that fixed (or, hopefully not; imagine what will happen once our oligarchs no longer need humans to build and control their machines, including swarms of killer drones, because the machines are reliably aligned with the oligarchs' prompts). >Going to assembler will certainly require more tokens. That's a good reason to keep higher-level languages. >Can coding agents generate smaller and/or more efficient assembler >compared to going via source/compiler? Even if they can, for most programs, and for the largest part of the rest, efficiency is not important, and higher-level languages will be used for that code in order to reduce token costs. >> What about human review that still happens AFAIK? What about >> programs that need to run on several architectures? > >The review should happen at the specification stage. It's a benefit of LLMs that the need for a complete specification is avoided, but that means that reviewing at the prompt level is insufficient, even if the alignment problem is solved. - anton -- M. Anton Ertl anton@mips.complang.tuwien.ac.at http://www.complang.tuwien.ac.at/anton/