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.

Have you debugged your training data? You might not like what you find. Introducing predictive data debugging: reveal and shape what your model will learn before training. In DPO datasets, we found broken guardrails, hallucinations, and fish fart fan fiction (seriously). (1/9)

Goodfire

25,871 subscribers

183,872 views • 1 month ago •via X (Twitter)

Education Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Today, AWS CEO Matt Garman announced Nova Forge, a model builder which lets companies inject their own data during the pre-training phase. "You [tell Forge]: 'Here's my corpus of corporate data, here's everything I need to know about my industry.' We then mix that in and finish pre-training the model. So you get a pre-trained model that totally understands your company and your data."

Today, AWS CEO Matt Garman announced Nova Forge, a model builder which lets companies inject their own data during the pre-training phase. "You [tell Forge]: 'Here's my corpus of corporate data, here's everything I need to know about my industry.' We then mix that in and finish pre-training the model. So you get a pre-trained model that totally understands your company and your data."

TBPN

97,010 views • 8 months ago

New video series: How to Amiko In this series, we’ll walk you through how to use the Amiko site, from onboarding your twin to exploring everything live inside our Beta. Episode 2: Train Your Twin Once you create your social and behavioral twin, shape what it knows about you and how it understands you in Training Chats. Training your twin builds depth: this is where you give it your personal traits, memories, preferences, history. This is where it learns specific, personal information about you, always controlled by you. Training teaches your twin more about you, beyond the social and behavioral layer, and keeps you in control of the data the twin has access to. Without training, your twin will act like the oldest child in a large family. With training, it will know you were the oldest of five siblings, and any additional details you wish to disclose.

New video series: How to Amiko In this series, we’ll walk you through how to use the Amiko site, from onboarding your twin to exploring everything live inside our Beta. Episode 2: Train Your Twin Once you create your social and behavioral twin, shape what it knows about you and how it understands you in Training Chats. Training your twin builds depth: this is where you give it your personal traits, memories, preferences, history. This is where it learns specific, personal information about you, always controlled by you. Training teaches your twin more about you, beyond the social and behavioral layer, and keeps you in control of the data the twin has access to. Without training, your twin will act like the oldest child in a large family. With training, it will know you were the oldest of five siblings, and any additional details you wish to disclose.

AMIKO

18,927 views • 8 months ago

Introducing Mainstage. This alpha build focuses on training tools, making and sharing combos with the community, and 2XKO match data. We hope you find value in it, and we really want to hear how we can improve it. Feature requests, feedback, cool ideas… we want it all! Jump in and let us know what you think.

Introducing Mainstage. This alpha build focuses on training tools, making and sharing combos with the community, and 2XKO match data. We hope you find value in it, and we really want to hear how we can improve it. Feature requests, feedback, cool ideas… we want it all! Jump in and let us know what you think.

FlyFighters

569,135 views • 7 months ago

Introducing Perle: an exceptional, modular AI training data management solution. It’s time to save your AI from itself. Perle can help. If AI could do one thing perfectly for you, what would it be? We might just make it happen. #HelloPerle

Introducing Perle: an exceptional, modular AI training data management solution. It’s time to save your AI from itself. Perle can help. If AI could do one thing perfectly for you, what would it be? We might just make it happen. #HelloPerle

Perle

21,838 views • 1 year ago

In case you missed it, we recently launched "Post-training of LLMs," a short course where you'll: ✅ Understand when and why to use post-training methods like Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning. ✅ Learn the concepts underlying the three post-training methods of SFT, DPO, and Online RL, their common use-cases, and how to curate high-quality data to effectively train a model using each method. ✅ Download a pre-trained model and implement post-training pipelines to turn a base model into an instruct model, change the identity of a chat assistant, and improve a model’s math capabilities. Learn more and enroll for free:

In case you missed it, we recently launched "Post-training of LLMs," a short course where you'll: ✅ Understand when and why to use post-training methods like Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning. ✅ Learn the concepts underlying the three post-training methods of SFT, DPO, and Online RL, their common use-cases, and how to curate high-quality data to effectively train a model using each method. ✅ Download a pre-trained model and implement post-training pipelines to turn a base model into an instruct model, change the identity of a chat assistant, and improve a model’s math capabilities. Learn more and enroll for free:

DeepLearning.AI

16,779 views • 1 year ago

Google's Jeff Dean says current pre-training is passive: initialize a model, stream the internet past it, let it observe But models need to learn not just from data, but by acting, predicting, and choosing what to learn from next "we have this artificial distinction now between pre and post-training, and it shouldn't exist long term"

Google's Jeff Dean says current pre-training is passive: initialize a model, stream the internet past it, let it observe But models need to learn not just from data, but by acting, predicting, and choosing what to learn from next "we have this artificial distinction now between pre and post-training, and it shouldn't exist long term"

Haider.

55,694 views • 4 months ago

🚨 Your home blood pressure reading is probably wrong right now. Not because your machine is broken. Because of what you did in the five minutes before you pressed that button. Here is what the data actually shows. Most people sit down, strap on the cuff, and press the button

🚨 Your home blood pressure reading is probably wrong right now. Not because your machine is broken. Because of what you did in the five minutes before you pressed that button. Here is what the data actually shows. Most people sit down, strap on the cuff, and press the button

Dr Alo, DO, FACC

15,680 views • 9 days ago

Think you know what really impacts your long-term health? These answers might surprise you. Your habits today shape how you feel tomorrow, and Healthspan on WHOOP helps you learn the difference.

Think you know what really impacts your long-term health? These answers might surprise you. Your habits today shape how you feel tomorrow, and Healthspan on WHOOP helps you learn the difference.

WHOOP

12,491 views • 1 year ago

AI Data Control - SDG Model V1 Live on SecureGPT! We are thrilled to announce that Selective Data Guard (SDG) v1 is now live on SecureGPT for our beta testers! SecureGPT has always ensured your personal information remains protected during AI interactions, but now, we're taking it a step further. With the SDG model, you gain the power to choose the data you wish to expose. SDG leverages proprietary algorithms to intelligently analyze and cluster similar entities, validating them against the custom rules you define. You now have the ability to specify what data to expose and what data to filter. This allows you to interact with AI agents seamlessly, with full control over your privacy. Your data, your rules—take control of AI interactions like never before!

AI Data Control - SDG Model V1 Live on SecureGPT! We are thrilled to announce that Selective Data Guard (SDG) v1 is now live on SecureGPT for our beta testers! SecureGPT has always ensured your personal information remains protected during AI interactions, but now, we're taking it a step further. With the SDG model, you gain the power to choose the data you wish to expose. SDG leverages proprietary algorithms to intelligently analyze and cluster similar entities, validating them against the custom rules you define. You now have the ability to specify what data to expose and what data to filter. This allows you to interact with AI agents seamlessly, with full control over your privacy. Your data, your rules—take control of AI interactions like never before!

ZKCrypt AI

12,416 views • 1 year ago

“don’t train your own model” is common ai advice. it's wrong. your token bill's the proof. today, we’re excited to launch castform into open preview. castform is the easiest way for you to train your own model, on your own data. open-weights models are performant and much cheaper. when trained on your task & proprietary data, they beat closed models. the thing standing between you and that was weeks of plumbing & years of ml expertise. with castform, model training is as simple as prompt engineering. castform bring your agent traces or raw corpora. castform turns it into training data, picks the right algorithmic recipes, manages gpus, and gives you an ide to watch and chat with your model as it learns. see what you can build with castform👇

“don’t train your own model” is common ai advice. it's wrong. your token bill's the proof. today, we’re excited to launch castform into open preview. castform is the easiest way for you to train your own model, on your own data. open-weights models are performant and much cheaper. when trained on your task & proprietary data, they beat closed models. the thing standing between you and that was weeks of plumbing & years of ml expertise. with castform, model training is as simple as prompt engineering. castform bring your agent traces or raw corpora. castform turns it into training data, picks the right algorithmic recipes, manages gpus, and gives you an ide to watch and chat with your model as it learns. see what you can build with castform👇

girish

456,481 views • 1 month ago

Why is your speed training not transferring to your sport? Like many, your lost learning "speed technique" before embedding "speed principles". Four key principles govern efficient and effective performance across multi directional speed. I have given you one, what are the other three? Credit to Dr Chris Bramah and his great lab for hosting us over the summer and providing such excellent video and data.

Why is your speed training not transferring to your sport? Like many, your lost learning "speed technique" before embedding "speed principles". Four key principles govern efficient and effective performance across multi directional speed. I have given you one, what are the other three? Credit to Dr Chris Bramah and his great lab for hosting us over the summer and providing such excellent video and data.

Jonas Dodoo

20,629 views • 1 year ago

Take care of your body, get in shape. You might not even need the extras. And while you’re at it, explore yourself a little, no shame in knowing what you like.

Sensitive content

Take care of your body, get in shape. You might not even need the extras. And while you’re at it, explore yourself a little, no shame in knowing what you like.

Larkin Love

36,874 views • 2 months ago

Here's how I would learn data engineering basics in 2025: - Find a data source you care about (examples: gaming APIs, stock market, web scraping, etc) - Use Python to interact and ingest your source. Initially just write the data to a CSV. - Setup an account with Snowflake or Google BigQuery. - update your Python script to load a table in Snowflake/BigQuery - schedule your script with CRON in the cloud with a service like Heroku. - build aggregations and visualizations on top of your ingested data Only thing this misses is data quality and complex job orchestration which you can learn later! How would you learn data engineering nowadays?

Here's how I would learn data engineering basics in 2025: - Find a data source you care about (examples: gaming APIs, stock market, web scraping, etc) - Use Python to interact and ingest your source. Initially just write the data to a CSV. - Setup an account with Snowflake or Google BigQuery. - update your Python script to load a table in Snowflake/BigQuery - schedule your script with CRON in the cloud with a service like Heroku. - build aggregations and visualizations on top of your ingested data Only thing this misses is data quality and complex job orchestration which you can learn later! How would you learn data engineering nowadays?

Zach Wilson

20,368 views • 1 year ago

You scroll. You shop. You stream. You spend. They profit. Your digital life should be more than a data trail. The Sophon App will turn it into an asset you own and earn from. Early access is coming. In the meantime, see what your data can do for you, and win your share of $50k

You scroll. You shop. You stream. You spend. They profit. Your digital life should be more than a data trail. The Sophon App will turn it into an asset you own and earn from. Early access is coming. In the meantime, see what your data can do for you, and win your share of $50k

Sophon

18,795 views • 9 months ago

We introduce Tagger Phase 2, bringing manual #data labeling and #data review into the #DeCorp ecosystem. With this iteration, we have completed the final components necessary for a fully integrated and decentralized cycle: 1) Data processing from scratch 2) Peer-to-peer data validation 3) Automated pre-training of AI copilot models 4) Quasi-autonomous active learning It is not long now before you may directly engage with this unprecedented and comprehensive system. See you soon.

We introduce Tagger Phase 2, bringing manual #data labeling and #data review into the #DeCorp ecosystem. With this iteration, we have completed the final components necessary for a fully integrated and decentralized cycle: 1) Data processing from scratch 2) Peer-to-peer data validation 3) Automated pre-training of AI copilot models 4) Quasi-autonomous active learning It is not long now before you may directly engage with this unprecedented and comprehensive system. See you soon.

Tagger

104,265 views • 1 year ago

Today we're launching Amplitude MCP MCP was made for data analytics. Ask a vague question and have an AI model iteratively query Amplitude to find you insights. Amplitude MCP exposes all of Amplitude's functionality so an agent can interact with it directly. It allows you to use Amplitude without having to know anything about Amplitude's user interface or your data taxonomy. It uses the intelligence of off the shelf AI agents, including Claude, Gemini, and Cursor, to query Amplitude data you already have to find answers. It leverages all of your existing Amplitude data and how it's used to navigate through large complex behavioral datasets. You can have conversations about your product data directly inside any AI model with MCP. It can create Amplitude dashboards and notebooks with what it finds and even make recommendations. This is the best MCP use case I've ever seen. Watching an AI agent think, reason, query Amplitude, and then repeat that process iteratively is magical. See it in action:

Today we're launching Amplitude MCP MCP was made for data analytics. Ask a vague question and have an AI model iteratively query Amplitude to find you insights. Amplitude MCP exposes all of Amplitude's functionality so an agent can interact with it directly. It allows you to use Amplitude without having to know anything about Amplitude's user interface or your data taxonomy. It uses the intelligence of off the shelf AI agents, including Claude, Gemini, and Cursor, to query Amplitude data you already have to find answers. It leverages all of your existing Amplitude data and how it's used to navigate through large complex behavioral datasets. You can have conversations about your product data directly inside any AI model with MCP. It can create Amplitude dashboards and notebooks with what it finds and even make recommendations. This is the best MCP use case I've ever seen. Watching an AI agent think, reason, query Amplitude, and then repeat that process iteratively is magical. See it in action:

Spenser Skates

32,332 views • 9 months ago

What new web features are your users actually ready for? Stop guessing. 🔮 Baseline helps you make data-driven decisions, so you can ship the right features at the right time. Learn more and find your target with #WhatsMyBaseline →

What new web features are your users actually ready for? Stop guessing. 🔮 Baseline helps you make data-driven decisions, so you can ship the right features at the right time. Learn more and find your target with #WhatsMyBaseline →

Chrome for Developers

24,818 views • 11 months ago

$AZRH fuck the shorts. You’re stupid enough to think we don’t have your data… I’m not talking about big bloc sales, I get it, it’s the holidays. I’m talking about the malicious 50-200 share sales into the bid. We have your data. We have an active investigation digging into all of you… your data are numbered - what you disgusting… our 60 Day VWAP is over $1.68… over half our float or 6.5M shares have flipped above $1.55… you’re not gonna win this battle - chalk it up to the fact that you made a bad bet… We know who you are, we have your data, and we are actively working with FINRA to Investigate and eradicate you… stay tuned…

$AZRH fuck the shorts. You’re stupid enough to think we don’t have your data… I’m not talking about big bloc sales, I get it, it’s the holidays. I’m talking about the malicious 50-200 share sales into the bid. We have your data. We have an active investigation digging into all of you… your data are numbered - what you disgusting… our 60 Day VWAP is over $1.68… over half our float or 6.5M shares have flipped above $1.55… you’re not gonna win this battle - chalk it up to the fact that you made a bad bet… We know who you are, we have your data, and we are actively working with FINRA to Investigate and eradicate you… stay tuned…

Wildcat

31,004 views • 7 months ago

Your data is being harvested for ads and AI training while you click 'I agree' on EULAs. The alternative? ICP puts data sovereignty in YOUR hands without the headache of running infrastructure. Store code and data while maintaining control. Take back ownership. Watch the full video:

Your data is being harvested for ads and AI training while you click 'I agree' on EULAs. The alternative? ICP puts data sovereignty in YOUR hands without the headache of running infrastructure. Store code and data while maintaining control. Take back ownership. Watch the full video:

DFINITY Foundation

13,213 views • 1 year ago