Rob Miles is an AI safety researcher. In early 2023, days after GPT-4’s release, he expressed his fears about arms race dynamics in AI development, in a computerphile video. His worst fears seem to have turned into reality.
Rob Miles is an AI safety researcher. In early 2023, days after GPT-4’s release, he expressed his fears about arms race dynamics in AI development, in a computerphile video. His worst fears seem to have turned into reality.
Rob Miles is a science communicator focused on AI Safety and Alignment.
He is probably my biggest influence. He was the first person to bring the complicated AI Safety concepts to the general public, many years ago, way before the LLM revolution started, in a time where today’s reality seemed sci-fi and the whole Alignment Study was an academic curiosity, a cool way for intellectuals to tickle the most rational and analytical parts of their brain.
His Youtube Channel Robert Miles AI Safety and his interviews at Computerphile popularized the frontier ideas coming from LessWrong, Machine Intelligence Research Institute, Centre for the Study of Existential Risk and leading AI Labs like OpenAI and DeepMind.
Robert Miles is a true hero of the AI Safety saga and an inspiring force for lethalintelligence.ai.
AI Ruined My Year
How to Help: https://aisafety.info/questions/8TJV/How-can-I-help
https://www.aisafety.com/
AI Safety Talks:
AI "Stop Button" Problem - Computerphile
How do you implement an on/off switch on a General Artificial Intelligence? Rob Miles explains the perils.
Part 1:
There's No Rule That Says We'll Make It
We're not doomed! But doom is definitely an actual possibility, and we need to act like it.
If you're thinking about working on AI Safety, check
Why Do I Avoid Sci-fi?
How come I tend to avoid discussing specific scenarios when talking about AI risks?
Intro to AI Safety, Remastered
An introduction to AI Safety, remastered from a talk I gave at "AI and Politics" in London
The second channel:
Ep. 136: The case for taking AI Safety seriously | Rob Miles
Through a series of popular explainer videos, Rob has become one of the most prominent voices in the AI safety community, exploring topics like
ROBERT MILES - "There is a good chance this kills everyone"
Please check out Numerai - our sponsor @
http://numer.ai/mlst
Numerai is a groundbreaking platform which is taking the data science
Robert Miles–Youtube, Doom
Robert Miles has been making videos for Computerphile, then decided to create his own Youtube channel about AI Safety. Lately, he's been working on
Would You Help an AI to Escape?
Could an AGI get a human to help them, just through a text conversation? Recently we got some new evidence about that question.
Rob Miles - Why should I care about AI safety?
In the latest episode of the TDS podcast, host Jeremie Harris talks with guest Rob Miles about AI safety, AI and the course of human evolution, and
Ep 12 - Education & advocacy for AI safety w/ Rob Miles (YouTube host)
We speak with Rob Miles. Rob is the host of the “Robert Miles AI Safety” channel on YouTube, the single most popular AI alignment video series out
9 Examples of Specification Gaming
AI systems do what you say, and it's hard to say exactly what you mean.
Let's look at a list of real life examples of specification
Why Would AI Want to do Bad Things? Instrumental Convergence
How can we predict that AGI with unknown goals would behave badly by default?
The Orthogonality Thesis video:
Is AI Safety a Pascal's Mugging?
An event that's very unlikely is still worth thinking about, if the consequences are big enough. What's the limit though?
Do we have to
Intelligence and Stupidity: The Orthogonality Thesis
Can highly intelligent agents have stupid goals?
A look at The Orthogonality Thesis and the nature of stupidity.
The 'Stamp Collector'
A Response to Steven Pinker on AI
Steven Pinker wrote an article on AI for Popular Science Magazine, which I have some issues with.
The article:
What can AGI do? I/O and Speed
Suppose we make an algorithm that implements general intelligence as well as the brain. What could that system do?
It might have better input and
Why Not Just: Think of AGI Like a Corporation?
Corporations are kind of like AIs, if you squint. How hard do you have to squint though, and is it worth it?
In this video we ask: Are
AI learns to Create ̵K̵Z̵F̵ ̵V̵i̵d̵e̵o̵s̵ Cat Pictures: Papers in Two Minutes #1
Some beautiful new GAN results have been published, so let's have a quick look at the pretty pictures.
More AI Safety coming soon of
The other "Killer Robot Arms Race" Elon Musk should worry about
Elon Musk is in the news, talking to the UN about autonomous weapons. This seems like a good time to explain one area where we don't quite agree
Why Not Just: Raise AI Like Kids?
Newly made Artificial General Intelligences are basically like children, right? So we already know we can teach them how to behave, right?
Are AI Risks like Nuclear Risks?
Concerns about AI cover a really wide range of possible problems. Can we make progress on several of these problems at once?
With thanks to
Deadly Truth of General AI? - Computerphile
The danger of assuming general artificial intelligence will be the same as human intelligence. Rob Miles explains with a simple example: The deadly
Holy Grail of AI (Artificial Intelligence) - Computerphile
Audible free book: http://www.audible.com/computerphile
Why can't artificial intelligence do what humans can? Rob Miles talks about generality
Why Asimov's Laws of Robotics Don't Work - Computerphile
Audible Free Book: http://www.audible.com/computerphile
Three or four laws to make robots and AI safe - should be simple right? Rob Miles on why
AI Safety - Computerphile
Safety in AI is important, but more important is to work it out before working out the AI itself. Rob Miles on AI safety.
Brain Scanner:
AI's Game Playing Challenge - Computerphile
AlphaGo is beating humans at Go - What's the big deal? Rob Miles explains what AI has to do to play a game.
What on Earth is Recursion?:
General AI Won't Want You To Fix its Code - Computerphile
Part 1 of a Series on AI Safety Research with Rob Miles. Rob heads away from his 'Killer Stamp Collector' example to find a more concrete example of
AI? Just Sandbox it... - Computerphile
Why can't we just disconnect a malevolent AI? Rob Miles on some of the simplistic solutions to AI safety.
Out of focus shots caused by
Using Dangerous AI, But Safely?
Can we keep powerful AI under control, using AI?
The Paper: https://arxiv.org/abs/2312.06942
AI Lab Watch:
Using Dangerous AI, But Safely?
Can we keep powerful AI under control, using AI?
The Paper: https://arxiv.org/abs/2312.06942
AI Lab Watch:
Learn AI Safety at MATS #shorts
Apply for MATS at matsprogram.org by Oct 6
#shorts
AI Ruined My Year
How to Help: https://aisafety.info/questions/8TJV/How-can-I-help
https://www.aisafety.com/
AI Safety Talks:
Apply to Study AI Safety Now! #shorts
Apply to SERI MATS at https://www.serimats.org/ by May 7th, and check out http://aisafety.training to stay up to date with events and programs!
Why Does AI Lie, and What Can We Do About It?
How do we make sure language models tell the truth?
The new channel!: https://www.youtube.com/@aisafetytalks
Evan Hubinger's Talk:
Apply Now for a Paid Residency on Interpretability #short
https://www.redwoodresearch.org/remix
Real main channel video coming soon!
#short #shorts
$100,000 for Tasks Where Bigger AIs Do Worse Than Smaller Ones #short
Check out http://inversescaling.com for more details and to apply
#shorts #short
Free ML Bootcamp for Alignment #shorts
Apply for the second MLAB (Machine Learning for Alignment
Win $50k for Solving a Single AI Problem? #Shorts
The Alignment Research Center is offering up to $50k for proposals for their Eliciting Latent Knowledge problem.
The report:
Apply to AI Safety Camp! #shorts
Trying out #shorts
Applications are open for next year's AI Safety Camp!
http://aisafety.camp
We Were Right! Real Inner Misalignment
Researchers ran real versions of the thought experiments in the 'Mesa-Optimisers' videos!
What they found won't shock you (if you've been paying
Intro to AI Safety, Remastered
An introduction to AI Safety, remastered from a talk I gave at "AI and Politics" in London
The second channel:
Deceptive Misaligned Mesa-Optimisers? It's More Likely Than You Think...
The previous video explained why it's *possible* for trained models to end up with the wrong goals, even when we specify the goals perfectly. This
The OTHER AI Alignment Problem: Mesa-Optimizers and Inner Alignment
This "Alignment" thing turns out to be even harder than we thought.
# Links
The Paper: https://arxiv.org/pdf/1906.01820.pdf
Discord
Quantilizers: AI That Doesn't Try Too Hard
How do you get an AI system that does better than a human could, without doing anything a human wouldn't?
A follow-up to "Maximizers and
Sharing the Benefits of AI: The Windfall Clause
AI might create enormous amounts of wealth, but how is it going to be distributed?
The Paper:
10 Reasons to Ignore AI Safety
Why do some ignore AI Safety? Let's look at 10 reasons people give (adapted from Stuart Russell's list).
Related Videos from Me:
Why
9 Examples of Specification Gaming
AI systems do what you say, and it's hard to say exactly what you mean.
Let's look at a list of real life examples of specification
Training AI Without Writing A Reward Function, with Reward Modelling
How do you get a reinforcement learning agent to do what you want, when you can't actually write a reward function that specifies what that
AI That Doesn't Try Too Hard - Maximizers and Satisficers
Powerful AI systems can be dangerous in part because they pursue their goals as strongly as they can. Perhaps it would be safer to have systems that
Is AI Safety a Pascal's Mugging?
An event that's very unlikely is still worth thinking about, if the consequences are big enough. What's the limit though?
Do we have to
A Response to Steven Pinker on AI
Steven Pinker wrote an article on AI for Popular Science Magazine, which I have some issues with.
The article:
How to Keep Improving When You're Better Than Any Teacher - Iterated Distillation and Amplification
[2nd upload] AI systems can be trained using demonstrations from experts, but how do you train them to out-perform those experts? Can this still be
Why Not Just: Think of AGI Like a Corporation?
Corporations are kind of like AIs, if you squint. How hard do you have to squint though, and is it worth it?
In this video we ask: Are
Safe Exploration: Concrete Problems in AI Safety Part 6
To learn, you need to try new things, but that can be risky. How do we make AI systems that can explore safely?
Playlist of the series so
Friend or Foe? AI Safety Gridworlds extra bit
The last video about the AI Safety Gridworlds paper. How does an agent detect and adapt to friendly and adversarial intentions in the
AI Safety Gridworlds
Got an AI safety idea? Now you can test it out! A recent paper from DeepMind sets out some environments for evaluating the safety of AI systems, and
Experts' Predictions about the Future of AI
When will AI systems surpass human performance? I don't know, do you? No you don't. Let's see what 352 top AI researchers
Why Would AI Want to do Bad Things? Instrumental Convergence
How can we predict that AGI with unknown goals would behave badly by default?
The Orthogonality Thesis video:
Superintelligence Mod for Civilization V
Let's play this new mod for Civ 5 that makes AGI an available technology!
Can we guide humanity to a utopian AI future, or will we destroy
Intelligence and Stupidity: The Orthogonality Thesis
Can highly intelligent agents have stupid goals?
A look at The Orthogonality Thesis and the nature of stupidity.
The 'Stamp Collector'
Scalable Supervision: Concrete Problems in AI Safety Part 5
Why can't we just have humans overseeing our AI systems?
The Concrete Problems in AI Safety Playlist:
AI Safety at EAGlobal2017 Conference
I attended a charity conference to learn about AI Safety!
Correction: Alan Dafoe is funded by a grant from the Open Philanthropy Project,
AI learns to Create ̵K̵Z̵F̵ ̵V̵i̵d̵e̵o̵s̵ Cat Pictures: Papers in Two Minutes #1
Some beautiful new GAN results have been published, so let's have a quick look at the pretty pictures.
More AI Safety coming soon of
What can AGI do? I/O and Speed
Suppose we make an algorithm that implements general intelligence as well as the brain. What could that system do?
It might have better input and
What Can We Do About Reward Hacking?: Concrete Problems in AI Safety Part 4
Three different approaches that might help to prevent reward hacking.
New Side Channel with no content yet!:
Reward Hacking Reloaded: Concrete Problems in AI Safety Part 3.5
Goodhart's Law, Partially Observed Goals, and Wireheading: some more reasons for AI systems to find ways to 'cheat' and get more reward than we
The other "Killer Robot Arms Race" Elon Musk should worry about
Elon Musk is in the news, talking to the UN about autonomous weapons. This seems like a good time to explain one area where we don't quite agree
Reward Hacking: Concrete Problems in AI Safety Part 3
Sometimes AI can find ways to 'cheat' and get more reward than we intended by doing something unexpected.
The Concrete Problems in AI
Why Not Just: Raise AI Like Kids?
Newly made Artificial General Intelligences are basically like children, right? So we already know we can teach them how to behave, right?
Learn about the issue by some of the best explainers out there
Subscribe for important updates !!!
© 2024 Lethal Intelligence – Ai. All rights reserved.
