Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

“It's important to avoid over-claiming about how much [formal verification] could solve our problems.” Zac Hatfield-Dodds explains why we need to balance verification methods with practical safety work.

FAR.AI

16,373 subscribers

870,853 views • 1 year ago •via X (Twitter)

Education Science & Technology News & Politics

Anya Rossi• Live Now

Private livecam show

4 Comments

FAR.AI1 year ago

Follow us for AI safety insights And watch the full video

Joe Dakwa1 year ago

🚨 Launching a blockchain project? 🚀 Don’t risk exploits from blackhats due to unaudited smart contracts. 🔐 Get affordable, High-Quality Smart Contract Audits trusted by top projects like Coinbase & Optimism. 👉 Book a FREE consultation

Agus 🔎 🔸1 year ago

@dodds_zac @Gato_el_simon, for balance, since I had sent you the safeguarding AI proposal

Jack Wheeler1 year ago

@dodds_zac This is of course not an issue in most applications but the moment you deploy your “models” in safety critical systems, you need formal verification

Related Videos

Redbelly is built for scale, with collaborative consensus and formal verification to back it up. Our CGO, Tim Bass explains why that matters...

Redbelly is built for scale, with collaborative consensus and formal verification to back it up. Our CGO, Tim Bass explains why that matters...

Redbelly Network

26,757 views • 1 year ago

Learn how to make your app at least 30% faster by fixing the "slow first start" problem with Baseline Profiles → This video explains why apps are slow on first launch (waiting for JIT) and how Baseline Profiles solve this by providing important classes/methods to the runtime at install time. We cover how to generate them with Jetpack Macrobenchmark, verify the gains, and how they work on Android 7+.

Learn how to make your app at least 30% faster by fixing the "slow first start" problem with Baseline Profiles → This video explains why apps are slow on first launch (waiting for JIT) and how Baseline Profiles solve this by providing important classes/methods to the runtime at install time. We cover how to generate them with Jetpack Macrobenchmark, verify the gains, and how they work on Android 7+.

Android Developers

11,495 views • 8 months ago

California faces very real challenges like the displacement of workers from AI and soaring costs. We can't just talk about Trump to avoid putting forward our own plan to solve our biggest problems. We need a bold economic agenda that protects workers and puts families first.

California faces very real challenges like the displacement of workers from AI and soaring costs. We can't just talk about Trump to avoid putting forward our own plan to solve our biggest problems. We need a bold economic agenda that protects workers and puts families first.

Katie Porter

10,961 views • 5 months ago

"It's really important to me for our entire community, movement and voice, that we get the A.I. protections we need to continue to be able to do this work that we all love so much." - #SagAftraMember Ashly Burch Learn more about the #VideoGameStrike:

"It's really important to me for our entire community, movement and voice, that we get the A.I. protections we need to continue to be able to do this work that we all love so much." - #SagAftraMember Ashly Burch Learn more about the #VideoGameStrike:

SAG-AFTRA

88,464 views • 2 years ago

Mason Greenwood, with another banger goal 🚀 , will never understand the reason why we have to sell him because some fan base pretends to be us when he's all we need to solve our goal scoring problems.

Mason Greenwood, with another banger goal 🚀 , will never understand the reason why we have to sell him because some fan base pretends to be us when he's all we need to solve our goal scoring problems.

The football boy

63,542 views • 1 year ago

Nick Fuentes asks why there aren't any Bills condemning Charlie Kirk's Assassination or the Gaza War. But there's one disavowing Anti Semitism. "Just solve problems for Americans. I'm willing to work with Democrats. We need to Rethink how we do things & realign the Politics."

Nick Fuentes asks why there aren't any Bills condemning Charlie Kirk's Assassination or the Gaza War. But there's one disavowing Anti Semitism. "Just solve problems for Americans. I'm willing to work with Democrats. We need to Rethink how we do things & realign the Politics."

KaizerRev

143,649 views • 8 months ago

If we're going to have AIs that fully automate math, they not only need to solve existing problems. We also need to teach them how to recognize what problem to solve next. Human mathematicians have heuristic models that they use to decide what to work on, like, “There’s something important going on with the randomness of primes - let’s codify some patterns we see into specific conjectures.” But these aren’t currently precise enough for RL.

If we're going to have AIs that fully automate math, they not only need to solve existing problems. We also need to teach them how to recognize what problem to solve next. Human mathematicians have heuristic models that they use to decide what to work on, like, “There’s something important going on with the randomness of primes - let’s codify some patterns we see into specific conjectures.” But these aren’t currently precise enough for RL.

Dwarkesh Patel

54,583 views • 4 months ago

Over the past few days, Alberta has had some close calls with our power grid. It's vital for all of us to understand why we have been having issues and how Ottawa’s 2035 timeline could exacerbate problems for Albertans, and why we remain committed to reaching a carbon neutral target by 2050.

Over the past few days, Alberta has had some close calls with our power grid. It's vital for all of us to understand why we have been having issues and how Ottawa’s 2035 timeline could exacerbate problems for Albertans, and why we remain committed to reaching a carbon neutral target by 2050.

Danielle Smith

159,061 views • 2 years ago

There’s a popular theory that AI will finally make formal verification mainstream because mathematical proof of correctness will be needed when machines write most or all of the code. But will this happen? Hillel Wayne is one of the best people to answer. Timestamps: 00:00 Intro 04:32 The Crossover Project 11:37 What software engineering does better 15:30 What traditional engineering does better 18:17 Formal methods 29:32 TLA+: what it is and demo 36:58 TLA+ at Amazon 38:10 Ways distributed systems break 41:03 Formal methods and systems thinking 46:20 The value of learning math 50:23 What TLA+ is good for and isn’t 52:50 Alloy: a declarative language for software modeling 58:53 Other formal methods tools 1:01:24 Property-based testing 1:05:31 AI and the need for formal verification 1:12:29 Logic for programmers 1:14:35 Hillel’s 2025 prediction on AI’s impact 1:21:30 Book recommendation Brought to you by: • Antithesis – verify your system’s correctness without human review or traditional integration tests – and avoid bugs or outages. • turbopuffer – a vector and full-text search engine built on object storage. It’s fast, cheap, and extremely scalable. • WorkOS – everything you need to make your app enterprise ready. Two things I found especially interesting, talking with Hillel: 1. Amazon used TLA+ to find a bug almost impossible to locate without formal methods. In the paper How AWS uses formal methods, the AWS team shared that they’d found a complicated bug for which the shortest error trace to exhibit was 35 steps (!!). The bug passed unnoticed through extensive design review, code reviews, and testing. AWS concluded they wouldn’t have uncovered it if they’d stuck to conventional testing approaches. 2. Why not use formal verification for everything, then? It’s because specs in the real world are a nightmare to write. Even a simple problem like “find the file in a directory that has the most lines” gets complicated when modeled with formal methods. We would have to answer questions like: ‘do we look at ASCII or UTF-8 new line characters, what about unreadable files, and Symlinks?’ Without formal methods, we can write a simple verification that is right in 99%+ of cases. Formal methods require a lot of extra effort for the less than 1% of exotic use cases!

There’s a popular theory that AI will finally make formal verification mainstream because mathematical proof of correctness will be needed when machines write most or all of the code. But will this happen? Hillel Wayne is one of the best people to answer. Timestamps: 00:00 Intro 04:32 The Crossover Project 11:37 What software engineering does better 15:30 What traditional engineering does better 18:17 Formal methods 29:32 TLA+: what it is and demo 36:58 TLA+ at Amazon 38:10 Ways distributed systems break 41:03 Formal methods and systems thinking 46:20 The value of learning math 50:23 What TLA+ is good for and isn’t 52:50 Alloy: a declarative language for software modeling 58:53 Other formal methods tools 1:01:24 Property-based testing 1:05:31 AI and the need for formal verification 1:12:29 Logic for programmers 1:14:35 Hillel’s 2025 prediction on AI’s impact 1:21:30 Book recommendation Brought to you by: • Antithesis – verify your system’s correctness without human review or traditional integration tests – and avoid bugs or outages. • turbopuffer – a vector and full-text search engine built on object storage. It’s fast, cheap, and extremely scalable. • WorkOS – everything you need to make your app enterprise ready. Two things I found especially interesting, talking with Hillel: 1. Amazon used TLA+ to find a bug almost impossible to locate without formal methods. In the paper How AWS uses formal methods, the AWS team shared that they’d found a complicated bug for which the shortest error trace to exhibit was 35 steps (!!). The bug passed unnoticed through extensive design review, code reviews, and testing. AWS concluded they wouldn’t have uncovered it if they’d stuck to conventional testing approaches. 2. Why not use formal verification for everything, then? It’s because specs in the real world are a nightmare to write. Even a simple problem like “find the file in a directory that has the most lines” gets complicated when modeled with formal methods. We would have to answer questions like: ‘do we look at ASCII or UTF-8 new line characters, what about unreadable files, and Symlinks?’ Without formal methods, we can write a simple verification that is right in 99%+ of cases. Formal methods require a lot of extra effort for the less than 1% of exotic use cases!

Gergely Orosz

34,215 views • 12 days ago

Elon Musk: If we solve longevity, the people in power would always remain in power. “Death does play an important role. I think you really could evolve humans to live much longer, most creatures to live much longer. But over time, Evolution has found that it's better for organisms to have a finite life. That death brings renewal, essentially. And I think we do need to be cautious about trying to solve longevity in a live-forever type sense, because I think our society, our culture would ossify and the people in power would always remain in power.” Interview with Jordan Peterson, July 22, 2024

Elon Musk: If we solve longevity, the people in power would always remain in power. “Death does play an important role. I think you really could evolve humans to live much longer, most creatures to live much longer. But over time, Evolution has found that it's better for organisms to have a finite life. That death brings renewal, essentially. And I think we do need to be cautious about trying to solve longevity in a live-forever type sense, because I think our society, our culture would ossify and the people in power would always remain in power.” Interview with Jordan Peterson, July 22, 2024

ELON CLIPS

448,644 views • 1 year ago

Sharing our work at NeurIPS Conference on reasoning with EBMs! We learn an EBM over simple subproblems and combine EBMs at test-time to solve complex reasoning problems (3-SAT, graph coloring, crosswords). Generalizes well to complex 3-SAT / graph coloring/ N-queens problems.

Sharing our work at NeurIPS Conference on reasoning with EBMs! We learn an EBM over simple subproblems and combine EBMs at test-time to solve complex reasoning problems (3-SAT, graph coloring, crosswords). Generalizes well to complex 3-SAT / graph coloring/ N-queens problems.

Yilun Du

47,660 views • 9 months ago

Today we launch our Global Annual Review 2023. This year’s review looks at how our global network of over 360,000 people across 151 countries helped to build trust in society and solve important problems over the last year. Read it in full here: #PwCGAR23 #TheNewEquation

Today we launch our Global Annual Review 2023. This year’s review looks at how our global network of over 360,000 people across 151 countries helped to build trust in society and solve important problems over the last year. Read it in full here: #PwCGAR23 #TheNewEquation

PwC

13,312 views • 2 years ago

What are the new weapons emerging now in today’s wars? Are they ethical? Using augmented reality, our colleague Nicola explains how some autonomous weapons work and why we need to act now to keep human control over life and death decisions.

What are the new weapons emerging now in today’s wars? Are they ethical? Using augmented reality, our colleague Nicola explains how some autonomous weapons work and why we need to act now to keep human control over life and death decisions.

ICRC

19,116 views • 2 years ago

ELON: TO SOLVE FULL SELF-DRIVING, YOU NEED TO SOLVE REAL-WORLD AI "What are the road networks designed to work with? They're designed to work with a biological neural net, our brains, and with vision, our eyes. And so, in order to make it work with computers, you basically need to solve real-world AI and vision. We need cameras and silicon neural nets in order to have self-driving work for a system that was designed for eyes and biological neural nets. I guess when you put it that way, it's quite obvious that the only way to solve full self-driving is to solve real-world AI and sophisticated vision." Source: Elon Musk TED Interview, April 2022

ELON: TO SOLVE FULL SELF-DRIVING, YOU NEED TO SOLVE REAL-WORLD AI "What are the road networks designed to work with? They're designed to work with a biological neural net, our brains, and with vision, our eyes. And so, in order to make it work with computers, you basically need to solve real-world AI and vision. We need cameras and silicon neural nets in order to have self-driving work for a system that was designed for eyes and biological neural nets. I guess when you put it that way, it's quite obvious that the only way to solve full self-driving is to solve real-world AI and sophisticated vision." Source: Elon Musk TED Interview, April 2022

Mario Nawfal

56,936 views • 1 year ago

'It's impossible to do business with us' - War Minister Hegseth explained why arms production in the US is being delayed. - You mentioned changes in policy that the Pentagon has implemented to speed up and simplify production. What are the main problems? Why are we seeing delays? - To a large extent, these delays are our own fault. First and foremost, it's our way of doing business. It was impossible to work with us. We behaved like a bad customer who changes his mind every year about what he needs and what he doesn't need. We need to get our act together. - SK

'It's impossible to do business with us' - War Minister Hegseth explained why arms production in the US is being delayed. - You mentioned changes in policy that the Pentagon has implemented to speed up and simplify production. What are the main problems? Why are we seeing delays? - To a large extent, these delays are our own fault. First and foremost, it's our way of doing business. It was impossible to work with us. We behaved like a bad customer who changes his mind every year about what he needs and what he doesn't need. We need to get our act together. - SK

Zlatti71

14,665 views • 6 months ago

Blockchain data quality used to be about consensus. Multiple providers, same output, trust the agreement. That's evolving. The future is verification: cryptographic proof of data from the block level up, with a full audit trail built in. BK joined the Crypto's Toughest Data Problems (and how to solve them) panel discussion at the Blockworks Digital Asset Summit to talk about where that's heading.

Blockchain data quality used to be about consensus. Multiple providers, same output, trust the agreement. That's evolving. The future is verification: cryptographic proof of data from the block level up, with a full audit trail built in. BK joined the Crypto's Toughest Data Problems (and how to solve them) panel discussion at the Blockworks Digital Asset Summit to talk about where that's heading.

The Graph

14,884 views • 4 months ago

How are you #ConnectingForGood in your community? 🤔 This #MandelaMonth we chat with our Vodacom #RedHearts regional champions about their volunteer work and how they successfully balance family, work and service to the community. ❣️

How are you #ConnectingForGood in your community? 🤔 This #MandelaMonth we chat with our Vodacom #RedHearts regional champions about their volunteer work and how they successfully balance family, work and service to the community. ❣️

Vodacom

171,260 views • 3 years ago