Joanna Bryson, brief

3.6K posts

Joanna Bryson, brief banner
Joanna Bryson, brief

Joanna Bryson, brief

@j2breve

All models are wrong, but modelling is essential. I tweet so excessively I curate myself–if I have time. See either @j2bryson or @j2blather for the real me.

Katılım Temmuz 2016
4 Takip Edilen827 Takipçiler
Joanna Bryson, brief retweetledi
Adam Przeworski
Adam Przeworski@AdamPrzeworski·
According to Freedom House (see Washington Post, March 9), the number of countries where democracy scores declined in 2022 was the lowest in 17 years while the number where these scores increased rose. Compare with V-Dem and decide what to think about expert ratings.
English
1
25
108
33.8K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
This & the previous tweet via @rossjanderson, also more a friend of #privacy than big business. Not everything #tefkaGAFAM do is bad by any means. They produce great public goods. But too much power without adequate regulation is always a problem. Thus antitrust, and utilities.
Matthew Green@matthew_d_green

I also want to say to folks like @alexstamos that the greatest and most enduring legacy of Facebook’s focus on improved CSAM detection seems to be that law enforcement intends to punish Facebook for having “so much CSAM.”

English
0
2
3
1.5K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
I dispute the cement thing. A) all things end. B) technology is more fragile than the life on which it depends. C) governance works.
Iason Gabriel@IasonGabriel

The GPT-4 system card (p.9) states that with wider adoption: "AI systems will have even greater potential to reinforce entire ideologies, worldviews, truths and untruths, and to cement them or lock them in, foreclosing future contestation, reflection, and improvement"

English
1
5
12
3.1K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
Got to love Scotland! First two speakers explaining how human centring is a core EU value. I guess just like they act as if they’ve a state, they’re now acting like an EU state, and a lot of EU is playing along. @EuropeanChatbot
Joanna J Bryson tweet mediaJoanna J Bryson tweet mediaJoanna J Bryson tweet mediaJoanna J Bryson tweet media
English
1
4
28
6.3K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
Seriously @google @gmail, you are going to say everyone with an "unAmerican" name (like Prof. Dr. Torsten Jörg Selck, shown here) is a spammer, whom it would be dangerous to click links or reply to? #AIBias #AIEthics
Joanna J Bryson tweet mediaJoanna J Bryson tweet media
English
1
2
8
1.8K
Joanna Bryson, brief retweetledi
Dina D. Pomeranz 🟣
Dina D. Pomeranz 🟣@DinaPomeranz·
Pessimism bias: in every country, people underestimate the happiness of others. E.g. people in Germany think 45% of other people in Germany will say that, taking all things together, they are very happy or rather happy. In reality, it's over 80% of people.
Dina D. Pomeranz 🟣 tweet media
English
16
104
293
0
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
MAYBE a RL system warping a principle representation in ONE COMPONENT (the #llm) of such an AI system would work. But humans & other mammals have separate systems for learning relatively quickly about exceptional observations not to act on, which is likely better architecture, 2/
English
1
1
4
1K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
This just looks like a way to generate anodyne text. When will people realise CONTROL requires an #actionSelection system with specific priorities, including informed models of the problem space being addressed? 1/
Marianne Sweeny@msweeny

One of the many steps to address bias, hate speech and toxic text inherent to most of the large language models like GPT3. Human mediation is key to making sure that the model used reflects your entire audience. Yes, really. Stellar keynote from @YejinChoinka with more detail.

English
3
4
7
2.9K
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
Ha, @jack agrees! (For a slightly different reason) "The biggest mistake I made was continuing to invest in building tools for us to manage the public conversation, versus building tools for the people using Twitter to easily manage it for themselves." cc @jonathanstray 1/2
English
1
1
1
0
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
The problem of course was monitising that, though adwords should work. Recommender algs for twitter just commoditise us, selling advertising & recommending follows based on crude categories. Lists & clients like @tweetbot avoid them, but interactions still suffer. #litter 2/2
English
2
1
3
0
Joanna Bryson, brief retweetledi
Joanna J Bryson
Joanna J Bryson@j2bryson·
Twitter was best when it was 120 characters long & had no algorithms at all, except a) search, b) chats being partially hidden as replies, & c) URL compression. Twitter's value comes from the HUMAN labour of compressing messages & curating who we follow, not AI. #litter 1/2
English
1
8
31
0

Keşfet

@j2bryson @mchardcastle @ruthaylett @sethuvij @simon_t_powers @verena_rieser @BoganiRonny @rossjanderson