Anders Sandberg
@anderssandberg
Followers
28K
Following
40K
Media
3K
Statuses
27K
RT @Research_FRI: We now have the first accuracy results from the largest-ever existential risk forecasting tournament. In 2022, we conven….
0
34
0
RT @eonsys: Structure to function in worms: new preprints probe how well the connectome predicts activity. Adding evidence and counter-evid….
0
1
0
Attempting Only you by the Platters! . NGL the vocal timing on this one was a challenge! .I think it came out really nice!
1
8
50
RT @davidad: At 🇬🇧ARIA, we’re serious about catalysing a new paradigm for AI deployment—techniques to safely *contain* powerful AI (instead….
0
28
0
Finally an update on astrology! .(Although I also want the 847,427 minor planets to be included in my personal horoscope; it might matter if 968 Petunia is in quadrature with 2015 AF43!).
immaterialscience.org
This article provides corrections to conventional astrology to account for dark matter and dark energy.
2
1
10
RT @sleight_henry: 🚀 Applications now open: Constellation's Astra Fellowship 🚀. We're relaunching Astra — a 3-6 month fellowship to acceler….
0
33
0
With a great baroque-style memento oblivionis in the form of the Brugmansia flowers (full of memory-attacking scopolamine).
0
0
16
I am absolutely convinced that the Lighthaven campus is someone's memory palace.
7
2
149
Always curious and full of energy, Beagles turn every walk into an adventure 🐾🎉.
98
374
7K
RT @ephi_thinktank: Nytt avsnitt Hälsa för ohälsosamma! .Vi möter framtidsforskaren Anders Sandberg som berättar om varför hans huvud ska h….
0
2
0
My main question is if we can cover the space of human-AI folie-à-deux well enough - there are many ways to go nuts. But it is notable that there are fairly standard throughfares in how people go off the rails (to mix metaphors badly). These standard risks can be caught.
0
0
1
Small study, synthetic data, etc. but definitely worth following up. It is not a complex or hard to implement solution.
1
0
1
They used a panel of "chaperone" agents to evaluate the interaction, intervening when it started to turn problematic, requiring unanimity to prevent false positives.
1
0
0
LLMs are fairly good at interpreting text (after all, that is their world), so using a LLM to evaluate if a conversation between an AI and a human is becoming dangerously sycophantic or parasocial seems doable. Some of my friends did a neat little experiment on this.
4
0
8
Arriving in The Bay, I find that 2 out of 3 billboards along the 101 are ads for AI-related things. Isn't this down a bit from how it was this spring?.
3
1
30
One of the most fruitful life hacks for me has been treating air travel time as "me time" - I have no obligation to work on anything, I can spend the time as I see fit since it is suboptimal for work anyway. This often produces good work, since the pressure is gone.
1
0
43
RT @Scholars_Stage: I will not defend the altar of procedure. I will defend the altar of consequence.
0
4
0
RT @avramidou: I wrote up a piece on mapping the landscape of AI welfare, based on work I did with Brad Saad at @GPIOxford and the Future I….
0
2
0
RT @HTLGIFestival: 💡 Science Spotlight 💡. Where does life begin? Where will AI take us? And can we ever reach a theory of everything? . Jo….
0
5
0
It is always cool to read a profile of somebody you know and admire. Jade is awesome.
In an act of slight obsession, I went down a research rabbithole and put together this 2,000-word profile of the PM's new AI adviser, Jade Leung. I spoke to people who've worked with her, building up a picture of a superstar high-performer with a commitment to AI safety.
0
1
8
RT @lhinderling: Automated optogenetic control of hundreds of cells in parallel. Each cell is individually steered, collectively acting as….
0
58
0
