stevenbedrick

[Rachael Tatman @[email protected] @rctatman]

It's wild to me that *this* isn't the main discussion point for AI Ethics Twitter right now.

What a week. 😩

[stevenbedrick @stevenbedrick]

Here's a link to an interview with Weizenbaum for some context: https://t.co/iLuFDhVJr5

[stevenbedrick @stevenbedrick]

Reading this, I found myself thinking of Joseph Weizenbaum's argument that computers are best understood as being fundamentally conservative forces. He wasn't talking about the same things as @LMSacasas was, but I think the thinking resonates more than a little bit.

[Charlie Warzel @cwarzel]

Last month @LMSacasas suggested that the internet isn’t a time, but a place. Specifically, the past. For Galaxy Brain I spoke to him/expanded on the idea. I think it explains why we are so exhausted by information & why our politics & culture feel so stuck https://t.co/CKlb1j6FDw

[stevenbedrick @stevenbedrick]

Bergstrom's quote in this article was absolutely priceless, and really brightened my day (not what I was expecting in a monkeypox article). The rise of the "an epidemiologist speaking their truth" genre has been one of the few bright points of the last couple of years.

[stevenbedrick @stevenbedrick]

@aclmeeting Fabulous, thank you so much! Is there anything under discussion around encouraging/requiring vaccination? I know that's not a simple question by any means, but it seems worth mentioning...

[stevenbedrick @stevenbedrick]

This is fabulous! I especially liked their 4th point ("Attribute agency to the human actors building and using the technology, never to the technology itself"). I've made a point lately of using words like "automation" when talking about applications of AI/ML for the same reason.

[Tech Policy Press @techpolicypress]

The Center on Privacy & Technology at Georgetown Law is removing “artificial intelligence,” “AI,” and “machine learning” from its institutional vocabulary. Read why in this piece from its Executive Director, Emily Tucker: https://t.co/3Fe4T3Ctq7

[stevenbedrick @stevenbedrick]

And +1 for her prominent shout-out to Weizenbaum; how Johnson's article, which was literally about people's reactions to computer-generated text, could have possibly managed to go for 10k words without even mentioning ELIZA is beyond my ability to comprehend.

[stevenbedrick @stevenbedrick]

Anybody who read last week's NYT Magazine article on large language models, or really any media coverage on AI ever, or who is interested in questions of "framing" in journalism, should make sure to read @emilymbender's latest: https://t.co/yo2MxnBD3V

[stevenbedrick @stevenbedrick]

She goes way beyond simply pointing out specific instances of eye-roll-worthy hype- though of course there's plenty of that, fear not! She shows how the hype is driven first principles that are fundamentally mistaken, and how that leads naturally to a misleading narrative frame.

[stevenbedrick @stevenbedrick]

@emilymbender I work in this general area (ML, for diagnostic purposes, using voice signals), and I found the article almost painful to read for the exact reasons you highlight in your thread. Thanks for taking the time to pour some water on the dumpster-fire of hype!

[@emilymbender.bsky.social @emilymbender]

I find this reporting infuriating, so I'm going to use it to create a mini-lesson in detecting #AIhype.

If you're interested in following this lesson, please read the article, making note of what you think sounds exciting and what makes you skeptical.

https://t.co/wEdi3HicCk

[stevenbedrick @stevenbedrick]

@vasheresearch @daborbolla @candice_erdmann @kalpathy1 Wherever, whenever, count me in!

[The Onion @TheOnion]

Scientist Really Thought Job Would Be Less Grant Writing And More Glow-In-The-Dark Lizard Making https://t.co/TcRoPW0LA9

[Post Stroke Speech Transcription (PSST) Challenge @PSSTchallenge]

The #PSSTchallenge is on! Compete in our Post-Stroke Speech Transcription shared task at https://t.co/0CbXbXOZFX. Travel funds available #speechrecognition #MachineLearning #NLProc
Deadline: 2 April
25 June, Marseille, France @lrec2022
Co-organized @Portland_State @OHSUSOM

[Rachael Tatman @[email protected] @rctatman]

Y'all read Amandalynne Paullada's dissertation yet? 👀 Section 1.3 in particular should be required reading for ML Engineers imo. https://t.co/wduSeG8OTG

[Ehud Reiter @EhudReiter]

New blog: What are the Problems with Rule-Based NLG?
https://t.co/NTxAe3hFXC

[Wirecutter Union @wirecutterunion]

With @wirecutter/@nytimes management refusing to come to the table to reach a deal, 100% of our unit will be striking from Thanksgiving through Cyber Monday. The best way you can support us is by boycotting @wirecutter from 11/25-11/29: https://t.co/sLOVFbdbZK

[stevenbedrick @stevenbedrick]

@ZachWeiner Autoregressive language model is a very good choice, but honestly I work in this exact field and we usually just call it a "next-{word,character,symbol,token} prediction" model most of the time. Sometimes we'll group it under the umbrella of "sequence prediction."

[stevenbedrick @stevenbedrick]

This whole thread is super on-point!

[Ali Alkhatib @_alialkhatib]

data without context is worthless. we've had entire drying racks of shards just go down the metaphorical drain because someone lost all the paper tags.

honestly i'd encourage you to talk to someone about "datasheets for datasets" but you guys fired the author of that paper, so

[Jessica Schleider, PhD @JSchleiderPhD]

Here is a 🧵 of terms related to money/funding in academia that no one explained to me until I was applying for faculty jobs (or after I'd started), that I pretended to understand until then, and that would’ve been very helpful to learn earlier than I did.

1/a whole bunch