Indeed. All software products you can get your hands on are open source - compiled code is only little more difficult to read than source code, but not that much if you learn how.

Which is why ~all companies switches to offering software as a service, so this mindset doesn't apply :).

> but not that much if you learn how.

Yesterday I threw some ghidra output into an LLM with very little context and got what seemed to be a reasonable run down of the original back. We're probably knocking on the door of being able to throw a binary into an LLM and getting the original program back unless there is active obfuscation done.

It is a very exciting time for anyone who likes playing old, abandoned and buggy games :').

I haven't played much with LLMs for decompilation, but I wonder how viable is using LLMs on binaries to port software to a different language (in combination with source code when available, but binaries might need fewer tokens).

As long as obfuscated code is isomorphic to its raw form, it’s sure to be decompilable. As for how much divergence is possible (in obfuscation), my intuition that it is very finite by definition.

Ghidra+LLMs really does make this a matter of time if we’re not already there yet.

I guess open source makes you more accountable.

I sometimes explain open source to people as auditable software.

That's a good way to explain it.

Also makes it harder to build a business around it.

With that combination no wonder most successful companies are closed source.