I was aware of commercial antivirus vendors (Crowdstrike) doing something like this, but this is the first I've seen it published by somebody in the open!
Have you considered writing up a blog post and submitting this to HN?
This is not my experience. I've been experimenting with something very similar to vera. However my language transpiles into multiple languages (Java, Typescript, Common Lisp, Rust, C++, Python, C# and Swift). The transpiler is written in the language itself (there's a separate bootstrap transpiler written in Common Lisp). But where I'm going is that Claude, at least, is extremely capable at writing decent code in my new language with barely any prompting; just minimal guidance on the language itself and no examples.
That's simply not true. That's just not the way LLMs work. LLMs are not magic.
LLMs are stateless, they don't "remember" your bespoke programming language manual and examples between completion calls, so you have to repeatedly include all that with each and every completion call, which balloons the number of tokens used, reduces how much useful work you can do with the remaining tokens and attention, and is a costly waste of tokens and electricity and money.
That isn't anywhere near as effective or efficient as using the LLM's pre-existing training on billions of lines of well known programming languages, manuals, tutorials, examples, code bases, stack overflow discussions, books, github repos, pr's, etc.
What is your extraordinary evidence for your extraordinary claims? Have you empirically measured how well it works, or is it just vibes and handwaving?
I learned Common Lisp years ago while working in the AI lab at the University of Toronto, and parts of this article resonated strongly with me.
However, if you abandon the idea of REPL-driven development, then the frontier models from Anthropic and OpenAI are actually very capable of writing Lisp code. They struggle sometimes editing it (messing up parens)), but usually the first pass is pretty good.
I've been on an LLM kick the past few months, and two of my favorite AI-coded (mostly) projects are, interestingly, REPL-focused. icl (https://github.com/atgreen/icl) is a TUI and browser-based front end for your CL REPL designed to make REPL programming for humans more fun, whether you use it stand-alone, or as an Emacs companion. Even more fun is whistler (https://github.com/atgreen/whistler), which allows you to write/compile/load eBPF code in lisp right from your REPL. In this case, the AI wrote the highly optimizing SSA-based compiler from scratch, and it is competitive against (and sometimes beating) clang -O2. I mean... I say the AI wrote it... but I had to tell it what I wanted in some detail. I start every project by generating a PRD, and then having multiple AIs review that until we all agree that it makes sense, is complete enough, and is the right approach to whatever I'm doing.
The whistler code you inline with your common lisp is an s-expression based DSL. So you can use common lisp macros, but those macros are generating something that will look familiar to CL devs but is restricted based on the eBPF archictecture and validator requirements. eg. it only supports bounded `dotimes`, some basic progn/let/when/if/cond/eq/setf/incf/decf and math, and a simple array iterator. No lists, loops, tagbody/go, conditions, etc, etc. There's a manual in the docs directory.
Besides, one can easily code a skill+script for detecting the problem and suggesting fixes. In my anecdotal experience it cuts down the number of times dumber models walk in circle trying to balance parens.
256k is just's just a placeholder for now. The default will get reduced as we get more experience with the draft implementation. The proposal isn't complete yet.
No. I have yet to propose the patches formally. The SBCL maintainers are reviewing the high-level proposal (on my blog) first. You can try the implementation, however. There's a pointer to the repo/branch on my blog. I need to build a proper benchmarking framework and publish some real numbers that people can reproduce before I am confident enough to submit the patches for review.
Let me know if you try it out. I would love some feedback (via github)
reply