Post #1830214

2026-04-28 10:12 UTC

Unpopular opinion: Computer science syllabuses were designed back in a time when working AI was nothing but scifi. A course that does not teach a deep and practical understanding of how LLMs work is no longer representative of computer science, and you need so much material to do so that massive compromises will need to be made in other parts of the syllabus

Replies (8)

@julesh@mathstodon.xyz 2026-04-28 10:15

I think that disagreeing with "LLMs are now central to computer science" would be ahistorical, and probably held mainly by people who grew up in the AI winter. For the vast majority of the field's existence this was *the* flagship problem.

Open ##2103413
@promovicz@mathstodon.xyz 2026-04-28 11:56

@julesh I don't mind some "AI focus", but I'm not sure about "essential" at the moment. We should not allow the hype into science. The subject itself seems fine to me.

Open ##2103432
@liamoc@types.pl 2026-04-28 13:07

@julesh i disagree, but if you're right, then i'm out. I'm not interested in a computer science that's about this stuff.

Open ##2103435
@michael_@mathstodon.xyz 2026-04-28 14:11

@julesh Is this because most computer scientists will soon need to develop or implement their own domain-specific LLMs? Or is there another reason why computer scientists, in particular, need a deep and practical understanding of LLMs, rather than simply using them in the way other scientists might? I’m asking as an outsider to CS

Open ##2103436
@boarders@mathstodon.xyz 2026-04-28 16:12

@julesh what would you see as the curriculum containing what you think of as core knowledge, and which couldn’t be captured by an upper level course in an existing cs/maths degree?

Open ##2103438
@migratory@jorts.horse 2026-04-28 16:36

@julesh I didn't think there *was* a deep understanding of how LLMs work. if there was, why can't any of the major companies make any guarantees about their output? instead we see them RL training and begging in system prompts and things are still trivial to jailbreak and unpredictable in practical usage

Open ##2103449
@mc@mathstodon.xyz 2026-04-28 17:07

@julesh you think a term-long thematic course wouldn't be enough? assuming a prior ML foundations course. anyway, I believe that more than becoming central to CS, AI is now spinning out to its own discipline, which eventually will have its own dedicated academic career and maybe even departments (this is probably not even a prediction but just what's going on atm). this is exactly what happened to CS btw, which didn't become central to mathematics but its own thing.

Open ##2103451
@jonmsterling@mathstodon.xyz 2026-04-28 21:00

@julesh This will age like milk.

Open ##2103452