Rosie @RosieCampbell Twitter profile

Pinned Tweet

Rosie

@RosieCampbell

1 year

When AI optimizes for human values it's "aligned" but when I optimize for inclusive fitness I'm "a slut" smh

1

2

141

Last Seen Profiles

@1fauxtograph

@BeckyGraha18842

@AnhSketch

@VISITFLORIDA

@SnowValley2022

@minami670127955

@KwietNoMore

@Spencer05294230

@Era8Apparel

@GhelfiPC93

@JotalentATV

@ESPNDSM

@RyanWil69167218

@1jamarowens

@cink653363

@maroof2roof

@floetry

@aysen_tok

@Ebenezer_krypto

@DREAM_archive__

@creek4lyfe

@LLTenshi

@minstradmus93

@moonlightlexie

@chief_redmeat

@Fry431

@Macca6412026417

@CharestElf27873

@Jurgen1701

@Jaynesc68

@DaBest822

@nehemiahscott1

@Ssb4_Zackray

@kappakappa31415

@LoganRay515266

@Not4Transformer

Rosie

@RosieCampbell

6 months

💖

Sam Altman

@sama

6 months

i love the openai team so much

5K

4K

73K

8

16

472

Rosie

@RosieCampbell

10 months

PSA: When teaching someone a board game, always start by explaining the objective / win condition first. It makes it so much easier to follow everything else

28

17

467

Rosie

@RosieCampbell

1 year

Wondering if AI needs to be conscious to be dangerous is like wondering whether the coronavirus enjoyed itself at all those superspreader events

10

55

401

Rosie

@RosieCampbell

6 months

Well if I find myself on the job market again I've got a hell of an example for the "Tell me about a time where you worked together as a team to overcome a difficult situation" type interview questions 😅

12

8

364

Rosie

@RosieCampbell

17 days

Someone asked about evidence-based skincare on LessWrong, it's my time to shine

Which skincare products are evidence-based? — LessWrong

The beauty industry offers a large variety of skincare products (marketed mostly at women), differing both in alleged function and (substantially) in…

www.lesswrong.com

16

14

313

Rosie

@RosieCampbell

1 year

“It’s so sad to see people convinced of X” “People who believe X are biased” “It’s harmful to believe X” “If X were true it would imply these bad things” “These bad people believe X” These may all be true, but unfortunately they have no bearing on whether X is in fact true

5

32

237

Rosie

@RosieCampbell

1 year

One of my least favorite conversational moves is when people express outrage at someone for making a comparison. 1. Comparing is not the same as equating 2. Analogies are not meant to be exact in every way, just similar in a relevant dimension See also

Rosie

@RosieCampbell

2 years

@andybudd It’s particularly frustrating when there is a component of moral outrage: “how *dare* you compare X [thing I don’t think is that bad] to Y [thing I think is absolutely horrific].” People seem to forget comparing =/= equating

1

2

45

21

22

230

Rosie

@RosieCampbell

17 days

I regret not freezing my eggs earlier. I think I also regret not donating my eggs. I'm still not sure I want kids, but I wish I'd known that declining egg quality after ~35 was the main factor in fertility. If you're on the fence I would be happy to chat

12

4

187

Rosie

@RosieCampbell

1 year

I've been taking semaglutide for ~4 weeks now and I can't believe how amazing it has been. I've lost 8lb so far, but even if I wasn't losing weight I think I would still take it just for the mental benefits. Considering writing up my thoughts in more detail if there's interest

20

3

182

Rosie

@RosieCampbell

3 months

Exactly one year on semaglutide: 68lb down. Science is amazing, well done humans 💪

14

0

163

Rosie

@RosieCampbell

6 months

OpenAI is nothing without its people

6

9

158

Rosie

@RosieCampbell

2 years

@SarahTheHaider I don't have a good reason for why a baby is a person but an unborn fetus is not, but I am pro-choice and think abortion should be safe and legal

24

2

124

Rosie

@RosieCampbell

4 months

@Aella_Girl I have drunk the milk that was in the fridge and which you were probably saving for breakfast Forgive me It was delicious so sweet and so lumpy

4

2

125

Rosie

@RosieCampbell

4 years

Feature request for @RoamResearch : A global toggle that hides all blocks with a certain tag (e.g. [[private]]) so that I can share my screen / give tours of my personal DB while keeping sensitive info hidden no matter where I navigate

4

0

100

Rosie

@RosieCampbell

11 months

Update: lost 30lb, still love semaglutide

Rosie

@RosieCampbell

1 year

I've been taking semaglutide for ~4 weeks now and I can't believe how amazing it has been. I've lost 8lb so far, but even if I wasn't losing weight I think I would still take it just for the mental benefits. Considering writing up my thoughts in more detail if there's interest

20

3

182

3

1

96

Rosie

@RosieCampbell

10 years

OH GOD NIGHTMARES FOR WEEKS! Thanks @cubicgarden :( http://t.co/YE8YgDYgZM

20

158

91

Rosie

@RosieCampbell

7 years

I've written up my talk Demystifying Neural Nets as a blog post. No prior Machine Learning knowledge needed. Enjoy!

Demystifying Deep Neural Nets

A relaxed introduction for programmers with no machine learning experience

medium.com

3

49

93

Rosie

@RosieCampbell

6 years

My H1B US visa has been approved! I'll be moving to California in January to start a role as Assistant Director of the Center for Human-Compatible AI at UC Berkeley 😃 So excited!!

16

2

88

Rosie

@RosieCampbell

5 months

I think this means I'm now in charge of MIRI right @ESYudkowsky ?

4

0

83

Rosie

@RosieCampbell

13 days

Who is thinking about the intersection of AI and epistemics? E.g. beneficial use cases (AI for forecasting, etc), the implications of AI on epistemic security, etc. Would love to chat

36

8

85

Rosie

@RosieCampbell

1 year

Honestly guys I would love nothing more than to dismiss AGI doom as nonsense or whatever, but the fact that I still haven’t encountered someone who can pass the ideological Turing test for it and convincingly address the actual arguments head on is kinda terrifying

6

4

75

Rosie

@RosieCampbell

1 year

Relatedly, I would like it if we could stop conflating: - Consciousness (whether AI has subjective experience) - Danger (whether AI has the capability or propensity to cause catastrophe) - Moral patienthood (whether we have an ethical obligation to consider AI welfare)

Rosie

@RosieCampbell

1 year

Wondering if AI needs to be conscious to be dangerous is like wondering whether the coronavirus enjoyed itself at all those superspreader events

10

55

401

5

9

69

Rosie

@RosieCampbell

1 year

@mattyglesias These are known as "hallucinations" - it's a known problem of large language models and something a lot of people are working on. We generally don't want the model to be making up factually incorrect info unless it's specifically what the user wants!

8

4

67

Rosie

@RosieCampbell

1 year

If you don't think we need to worry about catastrophic risks from AGI, is it primarily because: - It won't have the capability: AI will never be powerful enough to cause catastrophe - It won't have the propensity: AI will never have reason to cause catastrophe

Won't have the capability

122

Won't have the propensity

65

Something else

80

I am worried about AGI

573

34

16

65

Rosie

@RosieCampbell

5 years

Over the ~18 months I've been in the US, I've noticed lots of little things that seem strange to me as a Brit. I'll collect them in this ongoing thread... (1)

7

11

62

Rosie

@RosieCampbell

9 years

Is there anything more irritating and less effective than the EU cookie law? All it's done is train me to blindly click OK to anything

14

74

64

Rosie

@RosieCampbell

1 year

I'm surprised we don't often see people compare AI alignment to the Principal-Agent problem. Seems like a useful (if imperfect) analogy

13

6

60

Rosie

@RosieCampbell

7 years

Very sad to be leaving @bbcrd today, but looking forward to bringing Lancashire style to my new role in the US with this leaving gift!

12

4

60

Rosie

@RosieCampbell

4 years

A thread on @RoamResearch vs @obsdmd (spoiler: both have a place in my workflow):

7

3

59

Rosie

@RosieCampbell

4 years

Here's an example of how I used to game multiple choice questions: When did de Klerk become president of South Africa? A. 8/89 B. 9/88 C. 9/89 D. 8/98 Without even reading the question, you can guess the answer from the configuration of options (try before reading on!) (1/4)

Paul Graham

@paulg

4 years

The Lesson to Unlearn:

221

919

3K

4

9

57

Rosie

@RosieCampbell

1 year

@ilex_ulmus I used to get shaky and non-functional when hungry, it seems to regulate my blood sugar much better. I'm not constantly thinking about food and planning what I will eat. I can easily leave stuff on my plate when I notice I'm full. Feels also like more general mental clarity?

5

0

56

Rosie

@RosieCampbell

1 year

Don't worry folks, this was all intentional - part of an @Aella_Girl party where everyone had to bring an unpleasant experience to share. For mine, I pinched people with clips while reading out examples of poorly designed policies that created perverse incentives. Was very fun

Rosie

@RosieCampbell

1 year

Tonight I have discovered that stepping on lego is worse than being tasered

6

1

23

4

1

55

Rosie

@RosieCampbell

6 months

@ammaar It's easy to predict the next token if it's always a heart

1

53

Rosie

@RosieCampbell

5 months

Very proud to have been part of this work on the safety of increasingly agentic systems, please consider applying for a research grant if you have interests in this area!

OpenAI

@OpenAI

5 months

Our new research white paper identifies seven practices for keeping increasingly agentic AI systems safe and accountable as they become more common and more capable. We are providing research grants for work on a range of open questions.

682

410

2K

2

54

Rosie

@RosieCampbell

1 year

(There are people who can pass the ITT who have a much lower probability of doom than e.g. Yudkowsky, but AFAICT none of them dismiss the concerns entirely, and still have a much higher probability than I'd like)

1

0

53

Rosie

@RosieCampbell

1 year

PSA: if everyone stands a couple of feet away from the luggage carousel, everyone gets a good view of the bags and can easily step in to collect without obstruction, at ~0 personal cost. I get so unreasonably mad when people obliviously stand right up close and block it

7

1

51

Rosie

@RosieCampbell

6 years

At the airport waiting to fly out to California to start my new job! Very excited. Although the UK weather is being uncharacteristically lovely today...

17

1

52

Rosie

@RosieCampbell

4 months

UK besties: There are a bunch of cool jobs at the new AI Safety Institute up at the moment - and you get to work with my awesome former OpenAI colleague Jade Leung who is a force of nature

2

12

51

Rosie

@RosieCampbell

4 months

I'm on hormones for egg-freezing and I think it's making me... Have emotions? Is this how normal people are all the time? I don't care for it

6

0

49

Rosie

@RosieCampbell

8 years

Heartbreaking to see such racism on @MCRMetrolink

Watch Racism on a tram in Manchester | Streamable

Watch "Racism on a tram in Manchester" on Streamable.

streamable.com

9

25

42

Rosie

@RosieCampbell

8 years

The neighbourhood kids are loving our spooky eye that watches you as you move! Excellent work @MatthewShotton !

3

16

47

Rosie

@RosieCampbell

2 years

@andybudd It’s particularly frustrating when there is a component of moral outrage: “how *dare* you compare X [thing I don’t think is that bad] to Y [thing I think is absolutely horrific].” People seem to forget comparing =/= equating

1

2

45

Rosie

@RosieCampbell

8 years

How 'outrage culture' is often manufactured by the media by @ParkerMolloy

3

44

46

Rosie

@RosieCampbell

2 years

Anyone want to give me prompts to try on #dalle ?

DALL·E 2

DALL·E 2 is an AI system that can create realistic images and art from a description in natural language.

openai.com

66

3

45

Rosie

@RosieCampbell

1 year

It is possible to be concerned about the future catastrophic effects of climate change on the world while also caring about the populations and environments being affected today. The same is true of AI.

2

7

44

Rosie

@RosieCampbell

6 months

💖

OpenAI

@OpenAI

6 months

We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo. We are collaborating to figure out the details. Thank you so much for your patience through this.

6K

13K

67K

3

0

43

Rosie

@RosieCampbell

3 years

Today was my last day at @PartnershipAI - it's been an emotional day and I've been overwhelmed by the kindness of my incredible colleagues and friends within PAI and at Partner orgs. So proud of the work we've done together.

1

0

43

Rosie

@RosieCampbell

9 months

Was at a gathering last night and three people each independently thanked me for a (different) blog post I had written 😭🥰

6

0

42

Rosie

@RosieCampbell

2 years

@JeffLadish For me the most important things are: - Not shying away from difficult conversations - Delighting in each other's successes - An unrelenting commitment to remembering you're on the same team - Whimsy

1

0

39

Rosie

@RosieCampbell

1 year

Wish we didn't use "alignment" to mean both: - Ensuring AI acts broadly inline with human values and is robust to malicious actors directing it to do harm - Ensuring a powerful optimization system can be safely directed at a goal of our choosing without catastrophic side effects

5

2

41

Rosie

@RosieCampbell

7 months

Everyone at #OpenAIDevDay getting hyped about GPT-4 Turbo, GPTs, Assistant API, etc... Not enough people getting hyped about the build your own boba station

1

0

42

Rosie

@RosieCampbell

3 years

In a couple of weeks, I'll be joining @OpenAI to lead a team of safety specialists working on the responsible use of GPT-3 and other large ML models. It's a chance to put into practice a lot of the thinking I've been doing on responsible deployment - I'm very excited!!

4

1

41

Rosie

@RosieCampbell

1 year

As promised, a write up of my experience so far with semaglutide. Includes some general reflections and answers to FAQs

Semaglutide

Why I decided to try the new miracle weight loss drug, and what my experience has been so far.

www.rosiecampbell.xyz

3

38

Rosie

@RosieCampbell

1 year

@seanmcarroll Big fan of your podcast! But extinction does seem like a uniquely bad outcome to consider, and this statement demonstrates that many AI experts don't dismiss it as sensationalist

2

1

38

Rosie

@RosieCampbell

2 years

#dalle use case idea: recipe inspiration for leftover ingredients

2

36

Rosie

@RosieCampbell

17 days

I also regret my decision to try a lavender flavored latte this morning

1

0

35

Rosie

@RosieCampbell

1 year

Fun GPT-4 use case: I told it about my preferred clothing style/aesthetic and asked it which brands I should check out - it suggested a bunch I hadn't heard of before but I really like!

2

3

33

Rosie

@RosieCampbell

2 years

Why is this not a common thing: Canteen-style restaurants that serve ~2 easy-to-bulk-cook options per day (stew, curry, pasta, etc) using cheap ingredients like vegetables, beans, cheap cuts of meat For people who don't like cooking but want healthyish cheapish food

8

0

33

Rosie

@RosieCampbell

7 years

Thanks for letting me talk about neural networks with you #fronttrends ! A write-up of my talk is here

Demystifying Deep Neural Nets

A relaxed introduction for programmers with no machine learning experience

medium.com

1

18

34

Rosie

@RosieCampbell

3 months

@ESYudkowsky "Make it stop sounding unhappy to the police officers" seems like it would incentivize excessively RLHFing away any signs of sentience, which may end up obscuring useful signals of it

4

0

32

Rosie

@RosieCampbell

2 years

@JeffLadish Oh and hire a cleaner

0

31

Rosie

@RosieCampbell

3 years

Alongside my stellar team and Partners at @PartnershipAI , I've been working on responsible research and publication practices that help us prepare for the downstream consequences of AI. I'm so happy to share a synthesis of our thinking in this white paper!

Partnership on AI

@PartnershipAI

3 years

AI has quickly become a pervasive aspect of society with increasing inﬂuence over our lives. In turn, questions about this technology have transformed into urgent ethical considerations. We’re looking for answers. (1/5)

1

11

31

1

8

32

Rosie

@RosieCampbell

7 years

Weird that this guy thinks women shouldn't code because it's hard & boring when I've always found it fun &kinda easy

11

4

32

Rosie

@RosieCampbell

4 years

@visakanv "When the Okies left Oklahoma and moved to California, they raised the average intelligence level in both states" (the Will Rogers effect / Simpson's paradox)

1

0

32

Rosie

@RosieCampbell

6 months

Just want to echo this. Regardless of what happened/happens I have a huge amount of respect for the folks on the board and don't doubt they were doing what they thought was best

Paul Crowley ⏸️

@ciphergoth

6 months

AFAICT everyone involved in this is both ridiculously smart and a good person and all deserve our love, respect, and sympathy for what is an incredibly stressful time.

1

2

38

1

31

Rosie

@RosieCampbell

9 years

After recently watching my mum suffer with terminal cancer, I've never felt more acutely the need for @dignityindying RIP #TerryPratchett

4

20

31

Rosie

@RosieCampbell

3 years

@chiefofstuffs @nickcammarata Cammarata’s Law: You’re likely radically underestimating yourself, even after you take into account Cammarata’s Law

0

5

31

Rosie

@RosieCampbell

1 year

Some analogies are poor because they are not actually similar in a relevant way, and some are offensive because of the choice of relevant dimension, but then it should be possible to specify why. Merely comparing two things, even if one of them is bad, is not by default offensive

1

28

Rosie

@RosieCampbell

4 years

📢 Call for Participation 📢 Our #NeurIPS2020 workshop on "Navigating the Broader Impacts of AI Research" is accepting papers on topics related to how the AI research community can anticipate and mitigate risks Deadline: Oct 12, 2020 🗓️ Please share!

2

21

29

Rosie

@RosieCampbell

2 years

@DanHendrycks @sama This one intrigued me so I tried it! #dalle

Rosie × DALL·E 2 | colorless green ideas sleep furiously

Created with DALL·E, an AI system by OpenAI

labs.openai.com

2

1

29

Rosie

@RosieCampbell

3 years

@sashachapin On most spouse visas you can’t work, and you can’t even work remotely for a foreign company (even though you wouldn’t be “taking an American’s job”, and you would be spending your foreign earnings in the US economy)

0

27

Rosie

@RosieCampbell

4 years

Examples like this are why I'm worried about AI alignment and misspecified objectives: it's so easy to fall prey to Goodhart's Law and optimize for the target when it's not the thing you actually care about (even as humans who should know better!) (4/4)

2

1

28

Rosie

@RosieCampbell

5 years

News: Today is my last day at @CHAI_Berkeley 😞Sad to be leaving such an awesome research group but delighted to say I'm joining @PartnershipAI to lead programs and research on beneficial AI strategy! 🎉

2

0

28

Rosie

@RosieCampbell

10 years

What happens when non-techies accidentally open the developer console...

5

100

27

Rosie

@RosieCampbell

2 years

When a new AI system is announced, there's usually a gap between the capabilities people think it has versus the capabilities it actually has. I'm just not used to the gap being this way around 😅

0

26

Rosie

@RosieCampbell

1 month

Last minute trip to vibeclipse in Texas totally worth it, the clouds parted and it was beautiful. A+ would eclipse again

0

26

Rosie

@RosieCampbell

1 year

If you're confused about how and why AGI might kill us all by default, @ESYudkowsky 's Bankless podcast is a good place to start (more accessible than his writing IMO) Also worth listening to the Paul Christiano episode

159 - We’re All Gonna Die with Eliezer Yudkowsky

Eliezer Yudkowsky is an author, founder, and leading thinker in the AI space.------✨ DEBRIEF | Unpacking the episode: https://shows.banklesshq.com/p/debrief-...

www.youtube.com

1

2

24

Rosie

@RosieCampbell

6 years

My favorite AI newsletters, run by people working in the field. Including @rohinmshah , @jackclarkSF , @jjding99 , @charlotte_stix and @tabithagold

My favorite AI newsletters, run by people working in the field

The fast-moving pace of AI can make it tricky to stay up-to-date with latest news and developments. Here are some newsletters that can…

medium.datadriveninvestor.com

2

9

25

Rosie

@RosieCampbell

1 year

If there was a pill where 100% of people who take it become completely convinced that everyone is a flamingo, I can be sure that a) if I take it I too will believe everyone is a flamingo, and b) all those people are mistaken and it would be a mistake to take the flamingo pill.

4

2

24

Rosie

@RosieCampbell

6 months

@hlntnr Can't imagine how difficult this has all been for you. I'm sorry that people have been doubting your integrity, it's totally undeserved ❤️

0

1

25

Rosie

@RosieCampbell

7 years

Women in tech: Come to this BBC recruitment event 3rd April @MediaCityUK . I'm speaking about neural networks! pls RT

BBC TechWomen: Innovation, Networking and Careers

Learn about the BBC’s latest innovations and technologies; meet with other like-minded techies; meet our hiring managers and recruiters. The BBC is at the heart of technology: we use the latest...

www.eventbrite.co.uk

1

36

23

Rosie

@RosieCampbell

1 year

It's only integrity if you do it even when you fully believe it will result in worse outcomes. Otherwise it's just sparkling causal decision theory

0

3

24

Rosie

@RosieCampbell

7 years

Spare a thought for the women bracing themselves this morning for all the questions and comments they will face today about the Google thing

3

2

24

Rosie

@RosieCampbell

10 months

People say nice weather is overrated for wellbeing but waking up in a place with blue skies, bright sunshine, and palm trees always fills me with so much joy. I love California

4

0

23

Rosie

@RosieCampbell

1 year

Tonight I have discovered that stepping on lego is worse than being tasered

6

1

23

Rosie

@RosieCampbell

9 years

One of the best optical illusions I've seen in a long time... I can't stop watching

2

16

23

Rosie

@RosieCampbell

1 year

@xuanalogue @Soccermatics @pcastr @AmmannNora @LaParticle @AmandaAskell @KatjaGrace @BlancheMinerva @hlntnr @jesswhittles @catherineols @SmithaMilli @vkrakovna @ancadianadragan @tegan_maharaj @vaelgates @ghadfield @divyasiddarth @saffronhuang A few more: Jade Leung (not on Twitter) @norabelrose @miramurati @alyssamvance @csvoss @BethMayBarnes @inaquantumworld @allisondman @FreedmanRach @caro_jeanmaire @charlotte_stix

3

0

23

Rosie

@RosieCampbell

5 years

There is no contactless and hardly any chip+pin. At restaurants, you have to hand over your card for it to be scanned, then sign a bill and write down your tip, and hope that they charge your card the right amount after you leave. Seems pretty insecure?! (5)

5

0

23

Rosie

@RosieCampbell

7 years

Want my old job?! @BBCRD is a great place to work and is ramping up AI and ML projects. Please RT

0

47

22

Rosie

@RosieCampbell

1 year

Just got my hair done, I'm so blonde I could be a Fox News anchor

5

0

23

Rosie

@RosieCampbell

6 years

My last project with @BBCRD is in the news! @MatthewShotton @elvum

Hear the one about the robot director? BBC lets AI film comedy and music shows

Humans used to cling onto the belief that however powerful robots become they will never replace our capacity for that essential sentient quality, humour.

www.telegraph.co.uk

0

9

22

Rosie

@RosieCampbell

1 year

Alright here you go @TonyChurnside @what2donextidk

5

0

23

Rosie

@RosieCampbell

6 months

So many new followers and only two new likes on my banger of a pinned tweet

Rosie

@RosieCampbell

1 year

When AI optimizes for human values it's "aligned" but when I optimize for inclusive fitness I'm "a slut" smh

1

2

141

3

0

22

Rosie

@RosieCampbell

2 years

@EnzymeZoo

2

22

Rosie

@RosieCampbell

2 years

@EnzymeZoo

3

4

21

Rosie

@RosieCampbell

3 years

Wonderful to see this piece in @NatMachIntell by @lcvenema about our recent @PartnershipAI white paper on responsible publication! As per our recommendations, they are now asking for ethical and societal impact statements for some papers

How to be responsible in AI publication

Nature Machine Intelligence - A white paper from Partnership on AI provides timely advice on tackling the urgent challenge of navigating risks of AI research and responsible publication.

www.nature.com

1

7

21

Rosie

@RosieCampbell

4 months

The deadline for applying for these research grants is the end of next week (Jan 20) - please share if you know anyone interested in the risks and impacts of increasingly agentic systems!

OpenAI

@OpenAI

5 months

Our new research white paper identifies seven practices for keeping increasingly agentic AI systems safe and accountable as they become more common and more capable. We are providing research grants for work on a range of open questions.

682

410

2K

0

8

21

Rosie

@RosieCampbell

4 years

Are you interested in positively shaping research publication practices for AI? Join me at @PartnershipAI ! One week left to apply Especially keen to receive applications from underrepresented groups, please share with your networks 🙏

0

8

21