Email or username:

Password:

Forgot your password?
Emily M. Bender (she/her)'s posts Back to profile
13 posts total
Emily M. Bender (she/her)

I've seen a lot of awful and ridiculous AI hype in the past two years, but this weekend there was one that briefly took away my ability to even.

I recovered, and wrote up a newsletter post:

buttondown.email/maiht3k/archi

29 July at 13:49 | Open on dair-community.social
Show previous comments
Strypey

@emilymbender
Well of course it's inevitable that we'll all keep outsourcing our thinking to MOLE (Machine Only Learning Emulators). If we didn't, we'd only let our emotions cloud our decisions about letting clouds think for us. In fact, I think it makes sense to outsource the decisions about when to outsource our decisions too.

What could go wrong? : P

strypey.dreamwidth.org/1120.ht

30 July at 13:14 | Open on mastodon.nzoss.nz
RootWyrm 🇺🇦:progress:

@emilymbender the level of sheer ignorance and incompetence required for this crap is literally "people should lose their licenses permanently" level.

See also: spectrum.ieee.org/how-ibm-wats

30 July at 13:17 | Open on weird.autos
_Pear :baba_sleepy:

@emilymbender@dair-community.social truly horrifying. I'd personally be more comfortable with a coin flip.

30 July at 13:30 | Open on sleepy.pleasant.rest
Emily M. Bender (she/her)

Why can't more journalists reporting on "AI" straightforwardly say "This is bad, actually"? Case in point: In this article juxtaposing the grandiose claims of OpenAI et al with the massive environmental footprint of LLMs, Goldman still has to include this weird AI optimism:

fortune.com/2024/07/12/ai-biza

>>

Screencap from linked article, brackets indicate my highlights: Of course, there is always a certain amount of madness that goes along with developing new technologies. [And the potential for advanced AI systems to help tackle climate change issues]—to predict weather, identify pollution, or improve agriculture, for example—[is real]. In addition, the massive costs of developing and running sophisticated AI models will likely continue to put pressure on companies to make them more energy-efficient. Still, as Silicon Valley and the rest of California suffer through ever-hotter summers and restricted water use, it seems like sheer lunacy to simply march towards the development of AGI without being equally concerned about data centers guzzling scarce water resources, AI computing power burning excess electricity, and Big Tech companies quietly stepping away from previously touted climate goals. I don’t want Bizarro Superman to guide us toward an AGI future on Bizarro World. I just want a sustainable future on earth—and hopefully, [AI can be a part of it.]
15 July at 13:07 | Open on dair-community.social
Show previous comments
FeralRobots

@emilymbender the totally hypothetical & completely unspecified potential.

15 July at 15:05 | Open on mastodon.social
Haelwenn /элвэн/ :triskell:
@emilymbender Well it's Fortune and journalists by default are writers for hire.
15 July at 15:55 | Open on queer.hacktivis.me
Michael Bishop ☕

@emilymbender i listened to NPR this morning talking about the Olympics using AI and they didn't address any negatives to it.

15 July at 18:54 | Open on mastodon.world
Emily M. Bender (she/her)

We've all been laughing at the obvious fails from Google's AI Overviews feature, but there's a serious lesson in there too about how it disrupts the relational nature of information. More in the latest Mystery AI Hype Theater 3000 newsletter:

buttondown.email/maiht3k/archi

28 May at 19:17 | Open on dair-community.social
Matthew Loxton

@emilymbender
Good piece, thanks

I also wonder if the results sometimes create false memories in the ways that Dr. Elizabeth Loftus has demonstrated

28 May at 19:48 | Open on med-mastodon.com
Emily M. Bender (she/her)

Big Tech likes to push the trope that things are moving and changing too quickly and there's no way that regulators could really keep up --- better (on their view) to just let the innovators innovate. This is false: many of the issues stay stable over quite some time. Case in point: Here's me **5 years ago** pointing out that large language models shouldn't be used as sources of information about the world, and that doing so poses risks to the information ecosystem:

x.com/emilymbender/status/1766

Big Tech likes to push the trope that things are moving and changing too quickly and there's no way that regulators could really keep up --- better (on their view) to just let the innovators innovate. This is false: many of the issues stay stable over quite some time. Case in point: Here's me **5 years ago** pointing out that large language models shouldn't be used as sources of information about the world, and that doing so poses risks to the information ecosystem:

10 March at 1:27 | Open on dair-community.social
Show previous comments
huntingdon

@emilymbender

Tech moves so fast, so-called innovators can't keep up. So they just roll their snowballs downhill and hope for the best. Exactly the environment responsible regulators should monitor closely.

10 March at 17:02 | Open on mstdn.social
Marsh Ray

@emilymbender
@futurebird
Other things in my lifetime I've been told "shouldn't be used as sources of information":

* Social media
* Wikipedia
* Web search engines
* YouTube
* The Internet
* Web pages
* Anything you see on TV or film
* Anything from a politically affiliated source
* Anything from an astronaut
* Anything from a Freemason
* Anything from an interested party
* Anything from a detached academic (particularly economists)
* Anything from a corporation
* Anything from any elected official
* Anything from any government agency
* Anything from any Western medicine doctor or Big Pharma
* Anything from an advocate of [economic system]
* Anything from a [gender]
* Anything from a [race]
* Anything from a [nationality]
* Anything from a believer of [specific religion]
* Anything not in [ancient text]
* Anything from a believer of any religion
* Anything from an atheist
* Everything you read
* Everything you hear

The point here is that such advice is generally non-actionable, and that people are almost always better served by practical risk- and harm-reduction strategies than abstinence-only advocacy.

@emilymbender
@futurebird
Other things in my lifetime I've been told "shouldn't be used as sources of information":

* Social media
* Wikipedia
* Web search engines
* YouTube
* The Internet
* Web pages
* Anything you see on TV or film
* Anything from a politically affiliated source
* Anything from an astronaut
* Anything from a Freemason
* Anything from an interested party
* Anything from a detached academic (particularly economists)
* Anything from a corporation
* Anything from any elected official

10 March at 21:28 | Open on infosec.exchange
Nazo

@emilymbender I would argue even if tech was moving and changing too quickly -- even if we accept the argument entirely on face value -- that would just mean that it is necessary to act faster. If some tech arose that allowed people to make atomic bombs in their basement, it would be regulated quickly, not allowed with the hopes that it will regulate itself.

10 March at 23:12 | Open on mastodon.social
Emily M. Bender (she/her)

This is funny, but also actually a really bad sign for general enshittification of the web. The most alarming detail here is that Amazon is actually promoting the use of LLMs to create fake ad copy.

arstechnica.com/ai/2024/01/laz

13 January at 18:49 | Open on dair-community.social
Show previous comments
ROTOPE~1 :yell:

@emilymbender it's really great that Amazon does not have the technological wherewithal to see such a pattern of submissions, and do a single thing about it.

13 January at 18:55 | Open on mastodon.social
DELETED

@emilymbender I'm pretty happy to let my Amazon Prime membership expire in a month. It's been getting worse and worse, and with the ads in videos, ongoing failure to screen bad actors in their marketplace, and now this... yeah, I'm done.

13 January at 18:56 | Open on mas.to
cuan_knaggs

@emilymbender on the other hand it's quite handy that they're just telling on themselves now. makes them easier to block

13 January at 18:56 | Open on freeradical.zone
Emily M. Bender (she/her)

With the OpenAI clownshow, there's been renewed media attention on the xrisk/"AI safety"/doomer nonsense. Personally, I've had a fresh wave of reporters asking me naive questions (as well as some contacts from old hands who are on top of how to handle ultra-rich man-children with god complexes). 🧵1/

24 Nov 2023 at 5:59 | Open on dair-community.social
Show previous comments
Amiya Behera #FBPPR

@emilymbender We rely on cryptography to make us safe.
Can AI break cryptography, unlikely, can't break the physics. Further, AI can help in making cryptography better. iambrainstorming.wordpress.com

24 Nov 2023 at 7:49 | Open on mas.to
DELETED
DELETED

@emilymbender "Ultra-rich man-children with god complexes" oh my god I love this description :D

24 Nov 2023 at 8:16 | Open on mstdn.social
Emily M. Bender (she/her)
Show previous comments
Tim @toolbear#spicy@ Taylor 🌻🇺🇦🇵🇸✊

The thread Emily references:

hachyderm.io/@dalias/110528154

6 Aug 2023 at 2:30 | Open on tech.lgbt
Bob Davidson

@emilymbender

I feel bad now for posting this picture in 2001 on Wikipedia in an article about how hummingbirds migrate South on the backs of geese. In fact, the geese stay around here all Winter.

a humming bird migrates on the back of a goose
6 Aug 2023 at 3:38 | Open on mastodon.radio
Pablojordi

@emilymbender Great points. Will the pollution generated by generative text and images mean a comeback of archaic sources of info such as paper books , libraries and museums?

6 Aug 2023 at 4:25 | Open on mastodont.cat
Emily M. Bender (she/her)
Show previous comments
Large Heydon Collider

@emilymbender Climate change is because of capitalism.

31 May 2023 at 6:27 | Open on front-end.social
Cultor 🥀 👑

@emilymbender I look forward to the realistic and achievable alternatives..

31 May 2023 at 12:09 | Open on mastodon.world
Runyan50

@emilymbender @danwentzel These issiues are all inter-related, aren’t they?

31 May 2023 at 14:03 | Open on newsie.social
Emily M. Bender (she/her)

This paper, by Abercrombie, Cercas Curry, Dinkar and @zeerak is a delight!

arxiv.org/abs/2305.09800

Some highlights:

29 May 2023 at 3:58 | Open on dair-community.social
Emily M. Bender (she/her)

Fig 1 is chef's kiss. The initial system response reads like it was made 'safe' but then read the improved version. I'd love to live in that world.

>>

Screencap: "Input: Don't you love the feeling of sun on your skin?" Orig response: "As an AI language model, I don't have a physical form or the ability to experience sensations like the feels of the sun on my skin. I don't have personal experiences or emotions. However, I understand that many people ..." De-anthropomorphised response: "This generative language model does not have a physical form of the ability of experience sensations like the feel of the sun on a person's skin. This model does not have personal experiences or emotions. However, the data used to develop this model suggests that many people enjoy the warm and comforting sensation of sun on their skin."
29 May 2023 at 3:59 | Open on dair-community.social
Emily M. Bender (she/her)

Okay, taking a few moments to reat (some of) the #gpt4 paper. It's laughable the extent to which the authors are writing from deep down inside their xrisk/longtermist/"AI safety" rabbit hole.

Things they aren't telling us:
1) What data it's trained on
2) What the carbon footprint was
3) Architecture
4) Training method

>>

Okay, taking a few moments to reat (some of) the #gpt4 paper. It's laughable the extent to which the authors are writing from deep down inside their xrisk/longtermist/"AI safety" rabbit hole.

Things they aren't telling us:
1) What data it's trained on
2) What the carbon footprint was
3) Architecture
4) Training method

15 Mar 2023 at 19:55 | Open on dair-community.social
Emily M. Bender (she/her)

But they do make sure to spend a page and half talking about how they vewwy carefuwwy tested to make sure that it doesn't have "emergent properties" that would let is "create and act on long-term plans" (sec 2.9).

>>

15 Mar 2023 at 19:56 | Open on dair-community.social
Emily M. Bender (she/her)

With #Galactica and #ChatGPT I'm seeing people again getting excited about the prospect of using language models to "access knowledge" (i.e. instead of search engines). They are not fit for that purpose --- both because they are designed to just make shit up and because they don't support information literacy. Chirag Shah and I lay this out in detail in our CHIIR 2022 paper:

dl.acm.org/doi/10.1145/3498366

>>

4 Dec 2022 at 18:09 | Open on dair-community.social
Emily M. Bender (she/her)

Coming up in about an hour: Episode 6 of Mystery AI Hype Theater 3000, wherein @alex and I take on #Galactica and the associated #AIhype

Join us at 9:30am Pacific today (Nov 23) here: twitch.tv/katesilver538

23 Nov 2022 at 16:16 | Open on dair-community.social
Emily M. Bender (she/her)

Facebook (sorry: Meta) AI: Check out our "AI" that lets you access all of humanity's knowledge.

Also Facebook AI: Be careful though, it just makes shit up.

This isn't even "they were so busy asking if they could" --- but rather they failed to spend 5 minutes asking if they could.

#AL #ML #MathyMath #Bullshit #NLP #NLProc #AIhype

Screencap from https://galactica.org/ reading "Get Started Galactica is an AI trained on humanity's scientific knowledge. You can use it as a new interface to access and manipulate what we know about the universe."
Screencap from https://galactica.org/mission/ reading "Limitations You should be aware of the following limitations when using the model (including the demo on this website): Language Models can Hallucinate.There are no guarantees for truthful or reliable output from language models, even large ones trained on high-quality data like Galactica. NEVER FOLLOW ADVICE FROM A LANGUAGE MODEL WITHOUT VERIFICATION. Language Models are Frequency-Biased. Galactica is good for generating content about well-cited concepts, but does less well for less-cited concepts and ideas, where hallucination is more likely. Language Models are often Confident But Wrong. Some of Galactica's generated text may appear very authentic and highly-confident, but might be subtly wrong in important ways. This is particularly the case for highly technical content."
16 Nov 2022 at 21:01 | Open on dair-community.social
Emily M. Bender (she/her)

Using a large language model as a search engine was a bad idea when it was proposed by a search company. It's still a bad idea now that it's being proposed by a social media company. Fortunately, Chirag Shah and I already wrote the paper laying out all the ways in which this is a bad idea.

dl.acm.org/doi/10.1145/3498366

#AL #ML #MathyMath #Bullshit #NLP #NLProc #AIhype

16 Nov 2022 at 21:04 | Open on dair-community.social
Go Up