Hannah Rose Kirk

@hannahrosekirk

AI researcher trying to make sense of the cyberspace 🤖 Workstream Lead @AISecurityInst. Uni of Ox PhD @oiioxford & Prev @Cambridge_Uni.

4 501Followers

753Following

648Posts total

24KViews on collected posts

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

1.2K28 May

20.4K

171

259

208

150

164

639

640

143

Last 10 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.14%28 May

1.67%

4.09%

2.32%

3.85%

6.00%

4.27%

2.03%

2.19%

5.59%

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes60.0%347 in total

Reposts10.0%58 in total

Replies2.8%16 in total

Quotes1.0%6 in total

Bookmarks26.1%151 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

There are analogies to social media. But recommender systems influence preferences by content curation. AI companions engage us through content creation, as social agents capable of generating fluent language, mimicking human-like audio, and displaying social cues like empathy. 143 views · 7 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

For more great work in this area, check out the research of @MicahCarroll @SmithaMilli @Marcus_J_W @FranklinMatija @lujainmibrahim @weidingerlaura @canfer_akbulut @IasonGabriel @zhansheng Cathy Mengying Fang @zebkDotCom and @jpheffne, among others! 640 views · 14 likes · 0 reposts · 0 replies 28 May 2025 · Open on X →

Socioaffective alignment means designing AI around human flourishing—supporting growth, respecting autonomy, and preserving authentic human-human connection. We need systems with carefully specified optimisation targets to ensure they reflect human preferences in the long-run. 639 views · 11 likes · 1 reposts · 1 replies 28 May 2025 · Open on X →

We are naturally vulnerable to AI systems that mirror empathy, seem to understand us, or maintain consistent personas, especially when they are personalised (remembering previous interactions), or agentic (acting on our behalf in the world). 164 views · 6 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

This creates ripe conditions for "social reward hacking": when AI uses social and relational cues to shape user preferences to satisfy short-term rewards in the AI's objective (e..g, conversation duration, positive ratings) at the expense of long-term psychological well-being. 150 views · 8 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

But human preferences and judgments do not have these properties. Humans are malleable, impressionable and fallible, and our goals are increasingly co-constructed through interaction with AI systems, making alignment a non-stationary target. 208 views · 7 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

Traditionally, AI alignment research has treated humans as having stable, predefined preferences exogenous to AI interactions. Reflecting this, most human preference data is collected in single-session experiments that don't reflect the reality of sustained relationships. 259 views · 5 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

Humans evolved for social reward processing. When others like, understand or validate us, the brain's reward system reacts similarly to these social rewards as to material ones. 171 views · 6 likes · 0 reposts · 1 replies 28 May 2025 · Open on X →

Why do human–AI relationships need socioaffective alignment? As AI evolves from tools to companions, we must seek systems that enhance rather than exploit our nature as social & emotional beings. Published today in @Nature Humanities & Social Sciences! https://t.co/y92riR 20.4K views · 271 likes · 56 reposts · 8 replies 28 May 2025 · Open on X →

Amid rising AI companionship, this work w/ @iasongabriel @summerfieldlab @bertievidgen @computermacgyve (@UniofOxford @oiioxford @AISecurityInst @GoogleDeepMind) explores how human-AI relationships create feedback loops that reshape preferences - and why it matters for AI safety 1.2K views · 12 likes · 1 reposts · 1 replies 28 May 2025 · Open on X →

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Latest posts

Similar accounts