Uh oh.
This looks bad.
OpenAI will pay those who create the most engaging GPT's.
This makes their incentives very close to those of social media—capturing attention.
This could get dystopian very fast.

(What alternatives can we propose?) My guess is they haven't had the bandwidth to fully consider alternatives, and are still figuring out exactly what they want.

This means that providing recommendations and alternatives to them (publicly and privately) would be helpful.

Let's help make this a race to the top.

I was part of the red team for GPT-4 — tasked with getting GPT-4 to do harmful things so that OpenAI could fix it before release.
I've been advocating for red teaming for years & it's incredibly important.

But I'm also increasingly concerned that it is far from sufficient.
🧵⤵️ Before I highlight my concerns, I do want to at least ensure that people know about the GPT-4 System Card.
It's a very useful resource for anyone exploring the implications of such models and potential mitigations. /2
cdn.openai.com/papers/gpt-4-s…

Are you concerned about the way social media is impacting society? (You should be)

My new report explores "Bridging-Based Ranking"—a way to overcome the incentives for the division that destabilizes democracies.

🧵Here's what you need to know ⤵️
belfercenter.org/publication/br… The engagement-based ranking and recommendation systems of platforms like Facebook, YouTube, and TikTok reward divisive behavior with attention—and thus $$$ & power.

This determines the kinds of politicians, entertainers, journalists, etc. who will succeed.

The chronological feed fetish is one of things that drives me a little crazy.
It literally lets the loudest voices and people with too much time on their hands win the attention game 🤦‍♂️.

https://twitter.com/vishwanathsarng/status/1452754244046516224

This is better...

https://twitter.com/gpanger/status/1448040071936954370

I sometimes go farther: what if you had to choose for each person you followed if you want to their posts once a day, or per week, or per month?

Everyone can post as much as they want; and even choose a post to highlight for each period.

Could this make hard-hitting journalism less divisive?*

1. Summary bullet points at the top.
(Spell out: This article says __. It does NOT say __.)

2. An expandable FAQ at the end.
(Spell out: Yes, we knew about __; that wasn't relevant because __.)

*Also less artful 😢 The core challenge here may be that a "static article" is just one of many artifacts relating to a story within the information ecosystem—alongside tweets, summaries, clubhouse convos, rebuttals, etc.

These help form the beliefs, factions, distrust, etc.

https://twitter.com/metaviv/status/1370791444860989447

One of the fascinating things about addressing misinformation is that *everyone* wants to make it someone else's problem. (To be clear, this is not always a bad thing!)

The most recent example of this is @WhatsApp's new 🔎 feature, which makes misinformation Google's problem... When a message has been forwarded many times, WhatsApp shows a magnifying glass 🔎 next to the message. When tapped, it searches Google for the contents of the message.

This is a "data void" grifters paradise!
datasociety.net/library/data-v…

You know how Facebook lets you show that you voted? It's something that I expect that people across the political aisle like to show on their profiles.

What if there was also a way to literally show your *allegiance to the US Constitution* via Facebook? I'm particularly deeply concerned about an election where a huge segment of the population is moved to ignore the result — with horrific consequences.

Research suggests that pre-committing to abiding by a specific procedure might help.

AKA the Constitution.

How many people have told me repeatedly that deepfakes won't actually matter.
🤦‍♂️

It was only last week that a whole network of deepfake writers was identified...*after* running many op-eds in influential publications.

This technology will only get better.

https://twitter.com/razhael/status/1283357080146018307

To be more precise, face generation is a form of AI generated synthetic media. The term "deepfake" is often used synonymously.

The ability to create fake personas that others believe are real, supports information operations & has real downstream impacts.

https://twitter.com/mssilverstein/status/1283432852525285376?s=20

Twitter has finally released its new policies around deepfakes and other forms of manipulated media, following a public consultation.

My immediate take: they actually did a not terrible job!
(Better than Facebook in most ways.)

https://twitter.com/delbius/status/1224802196283244545

The spirit of Twitter's rule seems reasonable at least: (1) Remove manipulated media likely to impact public safety or cause serious harm. (2) Label everything else.

This leaves very large amounts of wiggle room for interpretation & execution.
This chart provides more detail:

Everyone talks about UX: User Experience

But what about CX?
- Collective Experience
- Community Experience

The usability and impact of a product across a *collection of people* within a society. UX isn't an academic field like "Science & Technology Studies" (STS)—it's a set of lenses and methodologies; ranging from "design thinking" to "user interviews".

Those problem solving approaches can be expanded upon (drawing on STS!).

CX is the result.

https://twitter.com/luke_fernandez/status/1138106566744272896

1 in 4 people near the Ebola outbreak believe virus isn't real.

From 2015 to 2018, the belief that vaccines are important decreased from 93% to 32% in the Philippines.

US anti-vaxxers are just the tip of an iceberg...
(Thread)
abcnews.go.com/International/… This isn't only about Facebook or YouTube driven misinformation.
Each region has its own complex story—and the rapid increase in vaccine hesitancy in the Philippines is an example of how that story can get *very* messy.

Google just released a *comprehensive* paper on how they are taking on disinformation (including at YouTube).

Let's take a look:
blog.google/around-the-glo… @Google is:
- Focusing on disinformation: "deliberate efforts to deceive"
- Using 3 core strategies:
1) Making quality count
2) Counteracting malicious actors
3) Giving users more context
- Collaborating(ish) — @firstdraftnews, @_trustproject, @factchecknet