@jbigham@hci.social
@jbigham@hci.social avatar

jbigham

@jbigham@hci.social

Professor of HCII and LTI at CMU SCS. Manage human-centered ML at . #hci, #a11y, #dialog, #nlp. AI/ML. #yinzfluencer of 4 kids. aspire to exceed my attention-seeking pattern regurgitating machine colleagues.

This profile is from a federated server and may be incomplete. Browse more on the original instance.

jbigham, to random
@jbigham@hci.social avatar

Hi folks, come join my team at Apple as an intern! We are super great and super fun, many products, much publications.

We do research at the intersection of HCI & ML, with focuses in Accessibility, Visualization, Learning Science, UI Understanding, and ML Developer Tools.

https://apple.ent.box.com/v/hcmi-internship

jbigham, to accessibility
@jbigham@hci.social avatar

i love how apple's hardware devices ship day one with substantial accessibility built in -- here's a video overview of features on .

been fun to be part of the conversations over the years, super proud of the teams that shipped all this --
https://www.youtube.com/watch?v=E1DLpGWOUsk

jbigham, to random
@jbigham@hci.social avatar

i still think there should be a, "Trapper Keeper or Generative Art" site where you just try to guess if some art is from a trapper keeper or from generative art

jbigham, to random
@jbigham@hci.social avatar

nothing could cause a new AI winter quite like convincing all the CEOs in the world to just like fire their workers and replace them with chatGPT now… "don't worry guys, it's ready," we could say.

jbigham, to random
@jbigham@hci.social avatar

i know this shouldn't surprise me, but the number of smart AI people who are saying things like, "if a person can read the nytimes, it's only fair that an LLM can read it into its parameters," is mindblowing.

i don't know how copyright should work for LLMs, we probably need to figure it out since it's new, but this argument makes no sense to me because it says because people have a right to do something it should be obvious that a machine should have that right. that is bonkers reasoning.

jbigham, to random
@jbigham@hci.social avatar

one of the problems with mastodon is that people don't actually like to boost things, and so the lack of an algorithm means a lot of cool stuff doesn't have an easy way to spread

jbigham, to random
@jbigham@hci.social avatar

With amazing progress (and hype) in AI, people are endlessly speculating about the future.

I've found a simple question can help make these conversations more productive, for ppl w/ a diversity of ideas and experiences:

"What will the people be doing?"

https://www.jeffreybigham.com/blog/2023/what-will-the-people-be-doing.html

jbigham, to random
@jbigham@hci.social avatar

haha, I think I know this professor 😂

jbigham, to random
@jbigham@hci.social avatar

tired: develop technology without considering people

wired: consider users in your technology design

inspired: consider all stakeholders and how they interplay, even the least sympathetic (e.g., the companies, payers, etc)

jbigham, to random
@jbigham@hci.social avatar

the two hardest problems in computer science are: (i) people, (ii) convincing computer scientists that people are the hardest problem, (vi) asynchronous programming, and (iii) off by one errors

jbigham, (edited ) to random
@jbigham@hci.social avatar

compared to today (August 2023), will the way we interact with computers in 10 years be:

jbigham, to random
@jbigham@hci.social avatar

i personally love that Apple doesn't randomly say "AI" but rather says specifics -- e.g., the chips, the models, even machine learning … would be much better if that was the standard

jbigham, to random
@jbigham@hci.social avatar

for almost a year, I've been writing a post about how HCI might engage with our new easy access to unintuitive statistical likelihoods of things -- it's been an incredibly useful exercise for me to wrestle with ideas, even as (especially as) my thinking has evolved.

there's a lengthy preamble, then thoughts on HCI opportunities bucketed into Benefit, Understand, Protect and Thrive --

"How HCI Might Engage with the Easy Access to Statistical Likelihoods of Things" --
https://www.jeffreybigham.com/blog/2023/how-hci-might-engage-with-easy-access-to-unintuitive-statistical-likelihoods.html

jbigham, to random
@jbigham@hci.social avatar

my chapter on "Accessibility and AI/ML" for our new book on Accessible Computing -- I think it's a decent primer on the promise and challenges of applying machine learning to accessibility use cases ::
https://bookish.press/tac/AIML

jbigham, to random
@jbigham@hci.social avatar

it's academic letter-writing season… can confirm that LLMs aren't helping me AT ALL, like not even a tiny bit 😭

jbigham, to random
@jbigham@hci.social avatar

my toots are 100% generated by my personal language model, influenced by both context and my own shortcomings.

jbigham, to random
@jbigham@hci.social avatar

today, i got an email from Apple, telling me that a subscription I had purchased that I no longer want was going to automatically renew, with a clear link to cancel. STRANGELY, subscriptions purchased outside of Apple never let me know this or make it easy to cancel.

jbigham, to random
@jbigham@hci.social avatar

look at all this cool accessibility stuff Apple just announced ahead of GAAD -- a whole new mode supporting cognitive impairments, personalized voices, point and speak added to magnifier, and more!

https://www.apple.com/newsroom/2023/05/apple-previews-live-speech-personal-voice-and-more-new-accessibility-features/

jbigham, to random
@jbigham@hci.social avatar

check out this new project called Evidence Inspector led by Kundan Krishna (who I co-advise w/ @zacharylipton)

it introduces a model and an interactive system to automatically detect evidence in generated text from source material and help correct it.

this may be a part of what people need to work with generated text, given the errors (hallucinations) it may contain.

you can see examples/play with it here:
https://evinspector.site/

this video shows how Evidence Inspector can be used to answer question, query for evidence for the resulting text, and then help a user interactively fix unsupported areas

jbigham, to random
@jbigham@hci.social avatar

this is a little conspiratorial, but…

have you ever wondered how to unlock a tesla when it's out of power? i hate that little button to open the door, seems so unnecesary, like touchscreens on a washing machine unnecessary.

turns out, there's a whole confusing manual about how to do it:
https://www.tesla.com/ownersmanual/model3/en_us/GUID-A7A60DC7-E476-4A86-9C9C-10F4A276AB8B.html

unrelatedly, someone died as their tesla slowly submerged in a pond, which is full of water, which doesn't play well with electricity?…
https://nypost.com/2024/03/09/us-news/angela-chao-made-panicked-call-before-dying-in-completely-submerged-tesla-on-texas-ranch/

jbigham, to random
@jbigham@hci.social avatar

these days Google Scholar usually knows about my publications having been made publicly available before I do, and so I just update the list of publications on my web page based on it.

I think back in the day, I would post papers on my web page because it would take too long for them to appear on the official sites, and they'd be paywalled and hard to access there.

jbigham, to random
@jbigham@hci.social avatar

short form video killed the text social media star

jbigham, to random
@jbigham@hci.social avatar

non-professors might ask, why do professors meticulously put all of their stuff on the websites? is it just for promotion (and bragging?)

yes, but also, i consult my own website like a million times a week because people keep asking me for stuff that is reported there, links and papers and bios and all that stuff.

so, when you make yourself, make it useful to you!

https://www.cs.cmu.edu/~jbigham/

jbigham, to random
@jbigham@hci.social avatar

coauthored 3 papers, reflective of the wonderful diversity of topics I get to work on --

  1. explored how local entrepreneurs might engage with generative AI via a series of workshops:
    https://www.cs.cmu.edu/~jbigham/pubs/pdfs/2024/deconstructing-genai.pdf

  2. built and studied a tool called COMPA that uses speech recognition and LLMs to help AAC users not be left behind in conversations
    https://www.cs.cmu.edu/~jbigham/pubs/pdfs/2024/compa.pdf

  3. built a tool to help ML engineers optimize ML models, widely used at Apple
    https://www.cs.cmu.edu/~jbigham/pubs/pdfs/2024/talaria.pdf

jbigham, to random
@jbigham@hci.social avatar

all the papers are now available, open access (free download) --
https://dl.acm.org/doi/proceedings/10.1145/3613904

imma read so fast and so much that the digital library thinks i'm a bot! (which really isn't that much, it's detection for this has really low precision)

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • provamag3
  • mdbf
  • ethstaker
  • khanakhh
  • magazineikmin
  • InstantRegret
  • rosin
  • Youngstown
  • slotface
  • thenastyranch
  • everett
  • kavyap
  • tacticalgear
  • DreamBathrooms
  • JUstTest
  • Durango
  • tester
  • cisconetworking
  • osvaldo12
  • ngwrru68w68
  • cubers
  • GTA5RPClips
  • modclub
  • anitta
  • megavids
  • normalnudes
  • Leos
  • lostlight
  • All magazines
    •