I can see that I'm getting out-voted, but I will offer contra here.
Joerg:
> Writing (useful) pseudocode is *hard*, harder than actual code.
Agree. The more you try to abstract, the more likely it is to have
a flaw.
I'm still not entirely finished checking the pseudo code from
chapter 3 of my dissertation, so I think pseudo-code for an
entire certificate factory is destined to failure:
Of course, Harm.On.ica S-O-L is keen to move from writing Haiku
poetry to writing epic poetry, now that the new Christopher Nolan
movie is out in theaters and getting good reviews.
Epic pseudo code poetry is still a bad idea! It sounds like that's
what Matthew Schwartz tried in his Anthropic "Vibe Physics"
launch, and it didn't sound to work as a one-shot.
> I think allowing pseudocode is a bad idea.
Disagree. Blanket statement needs qualifiers or time dependence.
Suggest alternatives:
-- Disallow epic pseudocode.
-- Disallow pseudocode until we better understand clone-by-rewrite.
-- Disallow epic
pseudocode until we better understand clone-by-rewrite.
Bob Lyons:
> I vote for disallowing pseudocode.
> When someone proposes draft edits for a sequence
that contains programs,
> we automatically check the syntax ...
> Needless to say, we don't check the syntax of any pseudocode. 😄
This gets back to the poetry point. Some people can't read poetry very well
so they don't like it. That's not LLM, at least not for verifiable code poetry.
Eventually (already?) there could be a dedicated external volume that runs
one or more LLM team members. They could check syntax of pseudocode
via clone-by-rewrite to any or sometimes all of those 17 languages.
Similar could be done for formulas we have, cloning them into algorithms and
checking automatically.
Pseudocode is kind of in-between formula and production code with regard to
length and complexity. Would you say a niche doesn't exist there?
I think most professional computer scientists want to write it for their textbooks
and classes as a stepping stone from "hello world" to bigger verifiable applications.
OEIS is not entirely or even mostly(?) a "hello world" repository, so I actually think
there's a sizeable selection of cases where pseudo-code could be helpful.
For example here:
Would the header "basic concepts" better be encoded as pseudocode? Or is
all of the just formulas? Are we sure the front page of A147680 is giving the
best possible definition of the Circular Disk Polyomino counting sequence?
If we did obtain concise pseudocode, maybe that could be added.
Tom Duff:
> Just use python. It's pretty much executable pseudocode already, without the inevitable ambiguities.
Yes, exactly! Then the Python Software Foundation can own a master copy, but not the one that's
concise and published on the OEIS front page.
I asked kimi to read the pseudocode, write SymPy, optimize sympy, and then passed to
Harm.On.ica S-O-L for finalization. It's long enough to need to be linked:
A123178 is proposed now, and I look forward to hearing the referee response about non-optimal
but instructive code that is checked rigorously against a certified recurrence relation.
Is this the kill-worthy slop? Or is the kill-worthy slop on A147680?
Considering rumors of bans going around I've got some anxiety about a slop-killer
showing up pretty soon and throwing out an Archimedes-follower with the bathwater.
All the best,
--Brad