Matt Lincoln

@mdlincoln.bsky.social

Software and data engineering leader; art historian; Victoria's Secret Model and Part-Time Parachute 🏳️‍🌈 he/him

a huge chunk of the top talent IN AMERICA is first- or second-generation and from china. the only reason china doesn't completely dominate us here is that american immigration policy is very good at stealing their best research staff. it's a good thing nobody's changing that!

surely the *checks notes* country of china is short on *checks notes* engineers and mathematicians and could not possibly compete at the engineering and applied mathematics field *touches earpiece* hold on, i'm getting new information bsky.app/profile/segy...

SE Gyges@segyges.bsky.social · 2w ago

by the time it gets confirmed that eg deepseek's tricks are all real and they really did make training 10x cheaper everybody on twitter has forgotten the litany of made up excuses about how they only trained on other llm outputs

I'm once again begging everyone to update their priors. "ChatGPT is not a search engine." Yes, yes it is. It actively searches the web based on your prompt, and returns active links relevant to your search/prompt. It's a search engine. It is. (Obvs don't use it that way if that's not your thing)

one major asymmetry is the way in which defection from the party consensus at the left edge of the party is considered betrayal because it's principed, but defection from the consensus at the right edge of the party is trustworthy because it is presumed to be a dishonest appeal to moderates.

Maxwell’s Demon: *ignoring entropy Me: ah-ha! But you CANNOT ignore entropy because you must store information! Maxwell’s Demon: *performs demon magic on the information bc he is a demon and is unconstrained by logic Me: well, nevertheless

Not a new observation, but I think discussion of AI + entry-level jobs could better account for entry-level *tech* jobs being a uniquely weird case. Partly that LLMs are especially good at entry-level coding, but also that there was a *massive* increase in new graduates that no other field had.

Time-series graph of CS+CE+I majors from the CRA Taulbee survey, showing increase from just over 50,000 in 2007-2008 to just under 250,000 in 2023-2024.

People have way over-indexed on the 2023-era view that AI companies want to “train on your data” because human authenticity is scarce and precious. I promise I’m not saying corps are beneficent when I say that a random collection of files and social media posts has at this point small value.

A lot of the new tools and processes appear to work best when an expert in the process being completed supervises or partners with one of these generative tools. That is not a useful classroom tool for most early or intermediate learners.

If you’re teaching DH right now, how are you thinking about your students’ relationship to generative AI? Are you ignoring it? Banning it? Discussing a policy with them first? Encouraging vibe coding? Interested to see how educators across the continuum of opinions are managing it.

this is something which, plainly, AI cannot do. journalism is by definition out-of-distribution, and even if you presumed that everyone wanted to read more Claudeprose than people already do, "writing up one's findings" is not the primary job a journalist performs!

Audrey Dutton@audreydutton.bsky.social · 4mo ago

I genuinely do not understand management's refusal to agree to basic protections, like that they won't replace us with AI. It's baffling. Makes no sense. We're on strike to show management how serious we are about this, and how eager we are to get a contract done!