***terpri_ is now known as terpri
<janneke>dongcarl: thank you, that's interesting; as hardware people they even mention david wheeler's ddc! <OriansJ>janneke: what are your thoughts of me merging the slow-utils as just a scheme folder in mescc-tools? <janneke>OriansJ: mescc-tools is fully source code, does not include binary seeds, right? <janneke>i think that would make sense; they can share the same test suite, etc right? <OriansJ>It also would allow us to reduce the guix bootstrap packages to just guile (to be replaced later by mes-m2), MesCC and mescc-tools <OriansJ>as slow-utils can run on most scheme interpreters without issue, it puts guix into the state of having exactly only 1 bootstrap binary (guile) *janneke cringes -- maybe it's time to move off of github? <OriansJ>well the only pieces not on savannah are M2-Planet, mescc-tools-seed and bootstrap-seed <OriansJ>getting bootstrap-seed approved for savannah is going to be a bitch <OriansJ>and mescc-tools-seed is unlikely to get approved until after we get bootstrap-seed onto savannah <OriansJ>if only we knew someone with good ties to the GNU project that might help us get bootstrap-seed onto savannah <pabs3>#savannah would be the place to ask *pabs3 also likes sourcehut <OriansJ>pabs3: I'm hoping civodul will just make it an official guix package and make a personal request to get it in <OriansJ>hopefully ddddd will return to work on slow-utils when it is part of mescc-tools on savannah <OriansJ>janneke: well we knew software generation was ultimately going to leverage less human thinking. It is just a logical extension of compiler technology. <Profpatsch>Could be fairly interesting to have this integrated in the editor workflow <janneke>maybe -- for now i don't like this one bit <Profpatsch>janneke: Even if they didn’t have Github, they could just use repology for their data base <janneke>i'm all for avoiding having humans do computer work <janneke>i'm not sure humans will (or should) think less <OriansJ>janneke: humans will always try to think less because cognition is extremely expensive in terms of calories <Profpatsch>Usually these systems require a very specific input to get the right output, so a user would have to learn the quirks of the system as well. <OriansJ>hence why we invent abstractions to hide details, and allow the human to think less <Profpatsch>But funny to see how the proof/type people have been trying the “generate code from types” for years, and the GPT stuff goes from plain text descriptions to solutions of similar complexity in a matter of months <OriansJ>The real danger is the steps when done which can't be easily rereated. <OriansJ>ML -> Haskell for example; is a huge risk because we don't even know how big of a gap we have to jump. <OriansJ>atleast with from hex0 to C, we know what it takes and have done it <OriansJ>as they are ultimately going to be the roots of trust for guix <civodul>i guess this kind of issue can be dealt with once we're actually using the seeds <civodul>we already have binary seeds at ftp.gnu.org anyway <civodul>so it should be the easiest part of this whole endeavor :-) <OriansJ>civodul: well the sooner that is done, the sooner we can migrate mescc-tools-seed to savannah as having a git submodule depending upon github probably isn't going to fly <OriansJ>I don't even need commit rights as it should only be updated when we add new architectures *pabs3 ponders the copyright implications of this AI stuff <janneke>yeah, well copyright law has little to do (anymore?) with what's logical, natural, honest or fair; it's become a tool for the powerful *janneke is feeling grumpy <OriansJ>I love how the thesis paper is also the compiler ***terpri_ is now known as terpri