Alice Hunsberger

Really excited to announce a new open weights safety model from Musubi: PolicyLM-1.7B: www.musubilabs.ai/blog/introdu...

This model is filling a specific gap: it's designed for content moderation in live chat, streaming, usernames... basically anywhere you need a good decision on text immediately, but you want a flexible taxonomy/ policy list that you control (not a preset vendor's idea of what harm labels you need).

The 47th president proudly spoke against trans people and anti-racism in his inauguration speech: people will now feel more empowered to spew hate. If you manage support, marketing, trust & safety, etc. create a playbook NOW to support & empower your team to respond to hateful customers. 🧵

Create resources and support for your users/ customers who may also be the target of harassment and hate. Signal to them in signs/ messages/ FAQs/ knowledge bases/ etc. that your team will support them.

The 47th president proudly spoke against trans people and anti-racism in his inauguration speech: people will now feel more empowered to spew hate. If you manage support, marketing, trust & safety, etc. create a playbook NOW to support & empower your team to respond to hateful customers. 🧵

Create psychological safety for your team. Let them know that you have their back. Listen to them. Check in on them. Make sure they have benefits that cover mental health support. We're all going to need it.

The 47th president proudly spoke against trans people and anti-racism in his inauguration speech: people will now feel more empowered to spew hate. If you manage support, marketing, trust & safety, etc. create a playbook NOW to support & empower your team to respond to hateful customers. 🧵

Create a "no questions asked" escalation policy, so that frontline staff can escalate to a manager if they feel unsafe or unable to answer a question. Make sure that escalation chain goes all the way up to VP or C-Suite level so everyone is supported.

The 47th president proudly spoke against trans people and anti-racism in his inauguration speech: people will now feel more empowered to spew hate. If you manage support, marketing, trust & safety, etc. create a playbook NOW to support & empower your team to respond to hateful customers. 🧵

Get really clear with senior leaders of the company you work for about corporate values and how to uphold them. Create tricky hypothetical scenarios (i.e. your biggest client sends a racist email; someone threatens to sue you for having a DEI program) and get answers BEFORE you need them.

Many of us working at tech companies are having to make moral and ethical decisions when it comes to where we work, what we work on, and what we speak up about. It's super difficult to know what to do, or even what your options are! 🧵 with resources

3️⃣ @anikacolliernavaroli.com writes about the harms for moderators from marginalized communities asked to work on content that attacks them. www.cjr.org/tow_center/b...

Many of us working at tech companies are having to make moral and ethical decisions when it comes to where we work, what we work on, and what we speak up about. It's super difficult to know what to do, or even what your options are! 🧵 with resources

1️⃣ Nadah Feteih discusses how tech workers (in integrity and t&s teams) can speak up about ethical issues at their workplace; activism from within the industry; compelled identity labor, balancing speaking up and staying silent, and more. integrityinstitute.org/podcast/work...

One thing I haven’t seen anyone talk about with Meta’s moderation changes: now relying on manual user reports for “less severe” issues that still violate their policies, rather than using a combination of user reports and automated detection. It’s bad. Here’s why: 🧵

This, combined with the rollback of hate policies, is REALLY going to change the vibes of Meta-run platforms. /🧵

One thing I haven’t seen anyone talk about with Meta’s moderation changes: now relying on manual user reports for “less severe” issues that still violate their policies, rather than using a combination of user reports and automated detection. It’s bad. Here’s why: 🧵

Other users will have their content removed after being reported, but feel it’s unfair because so many other people got away with it.

One thing I haven’t seen anyone talk about with Meta’s moderation changes: now relying on manual user reports for “less severe” issues that still violate their policies, rather than using a combination of user reports and automated detection. It’s bad. Here’s why: 🧵

Policies are only as good as ENFORCEMENT— and consistent enforcement at that. I have learned this the hard way, when I was head of t&s at a platform with little no to automated detection.

One thing I haven’t seen anyone talk about with Meta’s moderation changes: now relying on manual user reports for “less severe” issues that still violate their policies, rather than using a combination of user reports and automated detection. It’s bad. Here’s why: 🧵

If they see lots of folks doing something bad and it’s not immediately removed, they assume it’s ok and they won’t report. Even worse, they will often start exhibiting the same behavior themselves.

One thing I haven’t seen anyone talk about with Meta’s moderation changes: now relying on manual user reports for “less severe” issues that still violate their policies, rather than using a combination of user reports and automated detection. It’s bad. Here’s why: 🧵

Meta says they will "censor less" - including more speech about gender identity (allow hate speech against trans people) WHILE ALSO Today User Mag reports Instagram restricting content on searches for things like queer and nonbinary for teenagers who had the "sensitive content" filter turned on.

They claim it was a mistake and have reversed it (And I understand; I have made mistakes too!) But it's an interesting illustration of how speech & censorship goes both ways: it's hard to be Pro Free Speech (allowing speech against identities), while restricting content about those identities.

I made a T&S zine! My question for folks at TrustCon was: What does the T&S community need to hear right now? Of course, the T&S community delivered. 21662382.fs1.hubspotusercontent-na1.net/hubfs/216623...

Could a more transparent, collaborative, and adaptable policy enforcement model combat cries of 'censorship' and empower diverse online communities to self-moderate? Open-source policy documents and AI-driven auditability might hold the key. Read more www.everythinginmoderation.co/fighting-com...