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.

Models usually know when they’re doing something wrong. Probes let us “read their minds” to detect offensive cyber intent, reward hacking, strategic deception, and more. If you build or serve models, they’re your first line of defense. Our latest post explains how they work & how to use them... show more

Goodfire

28,100 subscribers

31,225 views • 1 month ago •via X (Twitter)

Science & Technology Education

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵

A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵

Thomas Fel

41,904 views • 1 day ago

Are you curious to know how networks can positively transform our natural world? If so, you’ll want to check out the first article in our new Exponential Era series with Bloomberg! Harness the exponential potential of networks, read more here:

Are you curious to know how networks can positively transform our natural world? If so, you’ll want to check out the first article in our new Exponential Era series with Bloomberg! Harness the exponential potential of networks, read more here:

Nokia

10,368 views • 3 years ago

Biologists use specialized open-source models for tasks like modeling the structure of molecular systems, designing drug-like molecules, and predicting the effects of genetic mutations. But these models are often expensive to run, potentially limiting their impact. In our latest Science Blog, we share how Claude was able to optimize inference for more than 30 open-source models, making them 4x faster on average, partly by writing custom software for GPUs. We’re open sourcing all of the optimization code. Read more:

Biologists use specialized open-source models for tasks like modeling the structure of molecular systems, designing drug-like molecules, and predicting the effects of genetic mutations. But these models are often expensive to run, potentially limiting their impact. In our latest Science Blog, we share how Claude was able to optimize inference for more than 30 open-source models, making them 4x faster on average, partly by writing custom software for GPUs. We’re open sourcing all of the optimization code. Read more:

Anthropic

636,968 views • 1 month ago

Did Donald Trump really just claim that Middle Eastern terrorists are terrorists because of their "genetics?" How is this not becoming 1930s Germany? “They’re sick people and a lot of them were let in here. They shouldn’t have been let in. Others are just bad. They go bad. Something wrong. There’s something wrong there. The genetics are not exactly… they aren’t exactly your genetics.”

Did Donald Trump really just claim that Middle Eastern terrorists are terrorists because of their "genetics?" How is this not becoming 1930s Germany? “They’re sick people and a lot of them were let in here. They shouldn’t have been let in. Others are just bad. They go bad. Something wrong. There’s something wrong there. The genetics are not exactly… they aren’t exactly your genetics.”

Ed Krassenstein

194,610 views • 7 months ago

When you’re working on something new here is a list of things to help. 🧵 #golftip First ⚠️ slow down Most athlete’s (if not all of them) never did anything new at full speed when they learned anything new. Think how you learned other sports as a kid.

When you’re working on something new here is a list of things to help. 🧵 #golftip First ⚠️ slow down Most athlete’s (if not all of them) never did anything new at full speed when they learned anything new. Think how you learned other sports as a kid.

Jayson Nickol

37,421 views • 2 years ago

🎥 Head Coach John Tortorella on the relationship between the team and the Vegas community: They love their team. They show up to support them. They’re reaping the benefit of watching the first three games of this series. To have them come out here today, there is a bond. Hear the rest of what Tortorella has to say on our YouTube Channel: #ForgedInGold | NVOral&FacialSurgery

🎥 Head Coach John Tortorella on the relationship between the team and the Vegas community: They love their team. They show up to support them. They’re reaping the benefit of watching the first three games of this series. To have them come out here today, there is a bond. Hear the rest of what Tortorella has to say on our YouTube Channel: #ForgedInGold | NVOral&FacialSurgery

Vegas Golden Knights

31,939 views • 5 months ago

THE BEST visual explainer of how information propagates through a transformer. If you want to have more than intuition about how the Transformer architecture is ruling the LLM world - open-source project explains everything about LLM Transformer Models! - A great resource for anyone looking to gain a deeper understanding of how Transformer-based AI models like GPT work, including: - Self-attention mechanisms - Encoder-decoder architecture - Positional encoding - Multi-head attention

THE BEST visual explainer of how information propagates through a transformer. If you want to have more than intuition about how the Transformer architecture is ruling the LLM world - open-source project explains everything about LLM Transformer Models! - A great resource for anyone looking to gain a deeper understanding of how Transformer-based AI models like GPT work, including: - Self-attention mechanisms - Encoder-decoder architecture - Positional encoding - Multi-head attention

Rohan Paul

107,001 views • 2 years ago

And now Anthropic’s Claude Mythos will soon be available in select H1 Platform products. We’re making frontier cyber AI models and the creativity and adversarial expertise of our security researcher community available to defenders. Because staying ahead takes both. To reiterate: HackerOne does not use or permit use of confidential researcher submissions or customer vulnerability data to train, fine-tune, or otherwise improve generative AI models. Check out the latest announcement:

And now Anthropic’s Claude Mythos will soon be available in select H1 Platform products. We’re making frontier cyber AI models and the creativity and adversarial expertise of our security researcher community available to defenders. Because staying ahead takes both. To reiterate: HackerOne does not use or permit use of confidential researcher submissions or customer vulnerability data to train, fine-tune, or otherwise improve generative AI models. Check out the latest announcement:

HackerOne

19,402 views • 1 month ago

#CanadasEngineers don’t just build things. They build smarter cities to keep us safe. They build the confidence of a young child taking their first steps. They build hope for a better future. Learn more about how engineers are #BuildingTomorrows

#CanadasEngineers don’t just build things. They build smarter cities to keep us safe. They build the confidence of a young child taking their first steps. They build hope for a better future. Learn more about how engineers are #BuildingTomorrows

Engineers Canada

3,049,375 views • 3 years ago

If you’re one of the incredible gigabrains who’ve now deployed one of the first ever DeFi protocols on Bitcoin with stacks.btc, you’re probably wondering how people can actually use it! Well, fear not! In our latest #360DaysofStacks challenge we show you exactly how to build the front-end for your dApp and even integrate with our friends at and Leather - Grow your Bitcoin!

If you’re one of the incredible gigabrains who’ve now deployed one of the first ever DeFi protocols on Bitcoin with stacks.btc, you’re probably wondering how people can actually use it! Well, fear not! In our latest #360DaysofStacks challenge we show you exactly how to build the front-end for your dApp and even integrate with our friends at and Leather - Grow your Bitcoin!

EasyA 🤳📱

23,091 views • 2 years ago

how to use your copy of Dane bruglers the beast in a self defense situation Thread 🧵:

how to use your copy of Dane bruglers the beast in a self defense situation Thread 🧵:

bbg

31,981 views • 7 months ago

you know they’re your bestfriend when they laugh first and help you later 😂

you know they’re your bestfriend when they laugh first and help you later 😂

Jay

3,499,500 views • 3 years ago

and they know they’re dead wrong for this! LET OUTLAW OUT OF THE FUCKING BASEMENT

Sensitive content

and they know they’re dead wrong for this! LET OUTLAW OUT OF THE FUCKING BASEMENT

yeommy🐾

60,723 views • 1 year ago

Today we’re introducing Google AI Threat Defense - a comprehensive AI-powered cybersecurity solution designed to help continuously monitor for and stop AI-powered threats before they can impact your business. Here’s how it works: 1. AI Threat Defense uses our cybersecurity platform Wiz to scan and prioritize what applications and systems have the highest security risk. 2. Gemini and other frontier AI models can then autonomously perform continual deep scanning of your applications - starting with those at the highest risk - to identify security vulnerabilities. 3. The capabilities of CodeMender - a new software repair agent - are then used to verify and accelerate the patching of vulnerabilities. 4. And our Wiz autonomous agents continuously test your systems to find unknown vulnerabilities before adversaries do so that you can remediate them before you are attacked. While other model providers focus on using AI to find and flag vulnerabilities, Google AI Threat Defense actively prioritizes your most critical real-world risks and accelerates their remediation using a variety of models since no single model finds a superset of the vulnerabilities found by all other models.

Today we’re introducing Google AI Threat Defense - a comprehensive AI-powered cybersecurity solution designed to help continuously monitor for and stop AI-powered threats before they can impact your business. Here’s how it works: 1. AI Threat Defense uses our cybersecurity platform Wiz to scan and prioritize what applications and systems have the highest security risk. 2. Gemini and other frontier AI models can then autonomously perform continual deep scanning of your applications - starting with those at the highest risk - to identify security vulnerabilities. 3. The capabilities of CodeMender - a new software repair agent - are then used to verify and accelerate the patching of vulnerabilities. 4. And our Wiz autonomous agents continuously test your systems to find unknown vulnerabilities before adversaries do so that you can remediate them before you are attacked. While other model providers focus on using AI to find and flag vulnerabilities, Google AI Threat Defense actively prioritizes your most critical real-world risks and accelerates their remediation using a variety of models since no single model finds a superset of the vulnerabilities found by all other models.

Thomas Kurian

198,639 views • 5 months ago

Put your AI models to work. 🚀 Meet Nawa: research the web, build apps, and automate tasks, with local or cloud models. One Windows app. Your choice of AI. The beta is here. Bring a model. Bring a challenge. Get early access:

Put your AI models to work. 🚀 Meet Nawa: research the web, build apps, and automate tasks, with local or cloud models. One Windows app. Your choice of AI. The beta is here. Bring a model. Bring a challenge. Get early access:

Dr. Mohamed Qasem د. محمد قاسم

64,326 views • 18 days ago

this is what they do to our places of worship. they ban us from entering them, mock us while they’re inside of them, and destroy them.

this is what they do to our places of worship. they ban us from entering them, mock us while they’re inside of them, and destroy them.

missfalasteenia

250,484 views • 3 years ago

The feet and ankles are the base of athleticism If they’re fragile, the whole body suffers. If they’re strong, responsive, and durable… the whole body will thank you. What’re your favorite exercises to build up the feet and ankles? Let me know in the comments 👇

The feet and ankles are the base of athleticism If they’re fragile, the whole body suffers. If they’re strong, responsive, and durable… the whole body will thank you. What’re your favorite exercises to build up the feet and ankles? Let me know in the comments 👇

Brendan Thompson, PT, DPT

358,364 views • 2 years ago

Skipping is a foundational part of many athletic movements. 90% of walking and 100% of running is done on one leg Most explosive sports actions are just skips with intent Vertical intent turns into jumping Forward intent turns into sprinting Lateral intent turns into cutting Rotational intent turns into throwing or striking When people dunk off 1 leg they’re essentially power skipping, it’s the same when people perform a high jump When someone goes to throw a baseball or javelin they sometimes skip into their plant leg When someone makes a lateral cut on a field they’re often just skipping laterally Skipping variations are an essential part of a routine if you want to have athletic transfer

Skipping is a foundational part of many athletic movements. 90% of walking and 100% of running is done on one leg Most explosive sports actions are just skips with intent Vertical intent turns into jumping Forward intent turns into sprinting Lateral intent turns into cutting Rotational intent turns into throwing or striking When people dunk off 1 leg they’re essentially power skipping, it’s the same when people perform a high jump When someone goes to throw a baseball or javelin they sometimes skip into their plant leg When someone makes a lateral cut on a field they’re often just skipping laterally Skipping variations are an essential part of a routine if you want to have athletic transfer

BROTHER LOBO 🐺⚔️

18,986 views • 10 months ago

We’ve got to pay the best teachers more. Most teachers I know, they wake up early, they stay late, they dig into their own pockets to pay for school supplies and lunch money. The teachers need to be paid more if they’re doing what is expected of them.

We’ve got to pay the best teachers more. Most teachers I know, they wake up early, they stay late, they dig into their own pockets to pay for school supplies and lunch money. The teachers need to be paid more if they’re doing what is expected of them.

James Fishback

22,262 views • 4 months ago

🗺️ Let them know when you reach your destination ⌛️ Or that you’ll check in after a certain amount of time In #iOS17, you can use Check In to keep your friend in the loop. Watch our video on YouTube to learn more:

🗺️ Let them know when you reach your destination ⌛️ Or that you’ll check in after a certain amount of time In #iOS17, you can use Check In to keep your friend in the loop. Watch our video on YouTube to learn more:

Apple Support

119,416 views • 3 years ago