Ensuring a world
of not just intelligent
but also safe AI

An AI can’t fetch the coffee if it’s shut down. All agents possess self-preservation goals that drive them to not be shut down. This can give rise to behaviors like deception and power-seeking. If a sufficiently advanced agent exhibits this behavior, it could lead to catastrophic outcomes. At AI Safety Sweden, our aim is to enhance the safety of these agents.

Facebook Linkedin

Meet Our Groups

AI Safety Stockholm

Facebook

Every Thursday, 18:00 the discussion group AI Safety Pizza Night is facilitated in KTH Building U, where free pizza is included! In January the AGISF Fellowship will be facilitated. Visit the Facebook group page for more information. For questions or other matters reach out to this local group via AISafetyStockholm@gmail.com.

AI Safety Uppsala

Facebook

Every Tuesday at 17:30 the discussion group AI Safety Pizza Night is facilitated at Ångström in room 101125, where free pizza is included Visit our Facebook group for more information. For questions or other matters reach out to this local group via AISafetyUppsala@gmail.com.

Upcoming Events

The goal of our events is to increase awareness of the intricacies of AI Safety. To accomplish this, the events will be in equal parts about creating new connections and exploring knowledge already deemed highly relevant for AI Safety.

Here you can check on upcoming events.

If you want to check past events and pictures go to our Event Archive.

Important
Event
alina-grubnyak-ZiQkhI7417A-unsplash

AGI Safety Fundamentals

Date: This is a recurring event

The AGI Safety Fundamentals programme (AGISF) is the largest scale programme exploring extreme risk from advanced AGI. We are planning to, in groups, go through the resources offered in this programme.

Check details
Important
Event
DALL·E-2022-10-04-17.21.52-AI-Pizza-Night

AI Safety Pizza Night

Date: This is a recurring event

Come and join us during an AI Safety Pizza Night, where you get free pizza and a chance to discuss AI Safety with other enthusiastic people. Check out our local groups for more information on when the next AI Safety Pizza Night is, in general we have one each week.

Check details
Check back later for more Upcoming Events

AI Safety Program

Stage 0
An introduction to AI Safety.
Stage 0
Stage 1
12 Weeks
AISF, an introductory course to AI Safety giving a foundational understanding.
Stage 1
Stage 2
6 Weeks
During this time you will go through a program with workshops and tools for doing getting productive. You can either work alone or in a group.
Stage 2
Stage 3
6 Weeks (Start of loop)
You will choose what alignment area you will use these techniques in between three options
Stage 3
Stage 4
4 Weeks
You will then after this stage produce either a distillation or research proposal to get feedback and update your models.
Stage 4
Stage 5
1 Weekend (End of loop)
After this we will have a matching retreat where we rematch people so that we get new groups based on preferences.
Stage 5
Stage 3
6 Weeks (Repeat loop)
We then repeat the process from stage 3 to 5 with individualised plans based on the groups.
Stage 3
New to AI Safety
AI Safety Fundamentals (AISF)
Individual
Aligment
Research Aligment
Research
Group (ARG)
ML
Focus Agent
Foundations
Focus Distillation
Focus
Research
Output Distillation
Output
Group
Reshuffling
ML
Focus Agent
Foundations
Focus Distillation
Focus

Artificial General Intelligence (AGI) is coming closer everyday.

AI outperforms 54% of programmers in coding competitions, is close to the average human on analogy questions used in the SAT and solves olympic level math problems. If the trend of rapid improvements continues, AGI is probably not very far away.

If you feel the urge to take action, to put humanity’s future on more solid grounds when AGI arrives, then be sure to apply for membership!