Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

LLMs sound homogeneous *because* feedback modalities like rankings, principles, and pairs cater to group-level preferences. Asking an individual to rank ~1K outputs or provide accurate principles takes effort. What if we relied on a few demos to elicit annotator preferences?

oshaikh13's profile picture

Omar Shaikh

2,007 subscribers

52,304 görüntüleme • 2 yıl önce •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

Johnson; The principle that we try to advance is that no one's ever asked or asked to compromise core principles.. But we have to compromise on occasion, our preferences, because that's the nature of a deliberative body like ours. So we're never going to get everything we want.
0:59

Johnson; The principle that we try to advance is that no one's ever asked or asked to compromise core principles.. But we have to compromise on occasion, our preferences, because that's the nature of a deliberative body like ours. So we're never going to get everything we want.

Acyn

116,446 görüntüleme • 1 yıl önce

New short course on Reinforcement Learning from Human Feedback! RLHF is one of the key techniques that led to the rise of modern LLMs. It is used to align LLMs with human preferences, to make them more honest, helpful and harmless, by (i) learning a reward function that mimics human preferences, as expressed in human-provided labels, then, (ii) tuning an LLM to generate outputs that receive a high reward. In this course, taught by Nikita Namjoshi, Developer Advocate for GenAI at Google Cloud, you'll learn the details of how RLHF works, including how to apply it to tune an LLM for your own applications. You'll also use an open source library to tune a base LLM to align with human preferences expressed in a training set, and evaluate the tuned model by comparing its responses before and after RLHF-tuning. Please sign up here!
3:26

New short course on Reinforcement Learning from Human Feedback! RLHF is one of the key techniques that led to the rise of modern LLMs. It is used to align LLMs with human preferences, to make them more honest, helpful and harmless, by (i) learning a reward function that mimics human preferences, as expressed in human-provided labels, then, (ii) tuning an LLM to generate outputs that receive a high reward. In this course, taught by Nikita Namjoshi, Developer Advocate for GenAI at Google Cloud, you'll learn the details of how RLHF works, including how to apply it to tune an LLM for your own applications. You'll also use an open source library to tune a base LLM to align with human preferences expressed in a training set, and evaluate the tuned model by comparing its responses before and after RLHF-tuning. Please sign up here!

Andrew Ng

205,569 görüntüleme • 2 yıl önce

Imagine standing up in the House or Commons and asking to immigrate people into Canada these who are tied to an Islamic extremist group. Jamaat e Islami whole purpose is to spread Islamic rule and live under Islamic principles. It's affiliated groups like the Hizbul Mujahideen is a designated terrorist group in Canada. I couldn't care less what 200 Bangladeshi want. We need none of it in Canada.
0:39

Imagine standing up in the House or Commons and asking to immigrate people into Canada these who are tied to an Islamic extremist group. Jamaat e Islami whole purpose is to spread Islamic rule and live under Islamic principles. It's affiliated groups like the Hizbul Mujahideen is a designated terrorist group in Canada. I couldn't care less what 200 Bangladeshi want. We need none of it in Canada.

Kirk Lubimov

80,986 görüntüleme • 10 ay önce

you might've heard about context engineering, but nobody has explained what it really is... that's why I made a full practical walkthrough on how to use it to make your AI outputs personalized and consistent in this walkthrough, I go over: – what context engineering actually is (without the jargon) – how to train AI on your own data and preferences – how to get outputs that sound like you wrote them (not generic AI) reply "context" and I'll DM it to you (must be following)
0:36

you might've heard about context engineering, but nobody has explained what it really is... that's why I made a full practical walkthrough on how to use it to make your AI outputs personalized and consistent in this walkthrough, I go over: – what context engineering actually is (without the jargon) – how to train AI on your own data and preferences – how to get outputs that sound like you wrote them (not generic AI) reply "context" and I'll DM it to you (must be following)

Tyler

20,018 görüntüleme • 1 yıl önce

We've prototyped a million versions of sliders, prompts, etc. (Credit to Benji Taylor for this one) However, some issues we haven't reconciled: A. It's unclear that algo preferences can be expressed through toggles B. Even if we solved (A)—i.e., the user could perfectly express their preferences to us—could X ingest that to give them what they asked for? Timelines are complex C. Is there any context where we override a user's preferences? For example, if they opted out of news or politics but there's time-sensitive information about an earthquake or terrorist attack. Having said all of this, we launched Custom Timelines and Snooze Topics and they've been great features for power users.
0:22

We've prototyped a million versions of sliders, prompts, etc. (Credit to Benji Taylor for this one) However, some issues we haven't reconciled: A. It's unclear that algo preferences can be expressed through toggles B. Even if we solved (A)—i.e., the user could perfectly express their preferences to us—could X ingest that to give them what they asked for? Timelines are complex C. Is there any context where we override a user's preferences? For example, if they opted out of news or politics but there's time-sensitive information about an earthquake or terrorist attack. Having said all of this, we launched Custom Timelines and Snooze Topics and they've been great features for power users.

Nikita Bier

278,517 görüntüleme • 2 ay önce

Here's the thing that all politicians have to think about, including those who provide funding, like Ken. There is pre and post Mamdani, because what he did is change the rules forever in terms of what it takes to get elected. Doesn't matter if it's a mayoral race or a governor race, senator race, doesn't matter. He took data science to a new level and social media to a new level. I don't agree with his policies. I hate everything he's doing, but he is a winner when it comes to understanding what it takes to win in politics today.
1:33

Here's the thing that all politicians have to think about, including those who provide funding, like Ken. There is pre and post Mamdani, because what he did is change the rules forever in terms of what it takes to get elected. Doesn't matter if it's a mayoral race or a governor race, senator race, doesn't matter. He took data science to a new level and social media to a new level. I don't agree with his policies. I hate everything he's doing, but he is a winner when it comes to understanding what it takes to win in politics today.

Kevin O'Leary aka Mr. Wonderful

20,551 görüntüleme • 1 ay önce

We interview the first ever AI welfare researcher at an AI company (Kyle Fish) about: • Why arguing LLMs aren't conscious because they 'just predict the next token' is like saying humans can't be conscious because we 'just reproduce' • Why AI consciousness skeptics are overconfident • Whether overestimating or underestimating AI consciousness is the bigger risk • What studies of AI preferences indicate they might enjoy or dislike doing • Compensating AI models if they're one day found to have been suffering • Anthropic's first ever welfare assessment for a frontier AI model before release • Why Claude gets into a 'bliss attractor state' when left to talk to itself • If we train AIs to enjoy serving us, is that a win-win or morally suss? • Letting Claude hang up on you • His p(conscoiusness) On the 80,000 Hours Podcast on Youtube, Spotify, Apple, and everywhere else. Links below. Enjoy!
2:34:34

We interview the first ever AI welfare researcher at an AI company (Kyle Fish) about: • Why arguing LLMs aren't conscious because they 'just predict the next token' is like saying humans can't be conscious because we 'just reproduce' • Why AI consciousness skeptics are overconfident • Whether overestimating or underestimating AI consciousness is the bigger risk • What studies of AI preferences indicate they might enjoy or dislike doing • Compensating AI models if they're one day found to have been suffering • Anthropic's first ever welfare assessment for a frontier AI model before release • Why Claude gets into a 'bliss attractor state' when left to talk to itself • If we train AIs to enjoy serving us, is that a win-win or morally suss? • Letting Claude hang up on you • His p(conscoiusness) On the 80,000 Hours Podcast on Youtube, Spotify, Apple, and everywhere else. Links below. Enjoy!

Rob Wiblin

33,894 görüntüleme • 1 yıl önce

We asked nico (Co-Founder of Fundamental) about hyperscaler competition and principles in AI development. "I think they would if they could (copy Shortcut)." “Even if you made an effort to copy me, you'd have to believe the things I believe in wholeheartedly.” "I think most of the people at frontier labs don't care about excel or they think that we're going to an input/output model anyway." “I think there's like a bajillion dollars to make in between those two states...”
1:13

We asked nico (Co-Founder of Fundamental) about hyperscaler competition and principles in AI development. "I think they would if they could (copy Shortcut)." “Even if you made an effort to copy me, you'd have to believe the things I believe in wholeheartedly.” "I think most of the people at frontier labs don't care about excel or they think that we're going to an input/output model anyway." “I think there's like a bajillion dollars to make in between those two states...”

TBPN

17,165 görüntüleme • 1 yıl önce

Introducing Compound Skills 🪄 You can now train the AI analyst in your way of working, just like you would a human. Ask Compound to create a Skill for any task (e.g. DCFs, waterfall modeling). Compound remembers your preferences and will apply it the next time it works on a similar task. You can also ask Compound to update skills, if you’ve provided more feedback during a task, or simply if you want to iterate on a skill. With Skills, Compound has become more powerful than ever - continually learning and adapting to your preferences. Try it now today at P.S For new users, get a one-time signup bonus of 40K free credits. You can also refer a friend (check settings for a link), and you’ll both get 10K free credits.
0:42

Introducing Compound Skills 🪄 You can now train the AI analyst in your way of working, just like you would a human. Ask Compound to create a Skill for any task (e.g. DCFs, waterfall modeling). Compound remembers your preferences and will apply it the next time it works on a similar task. You can also ask Compound to update skills, if you’ve provided more feedback during a task, or simply if you want to iterate on a skill. With Skills, Compound has become more powerful than ever - continually learning and adapting to your preferences. Try it now today at P.S For new users, get a one-time signup bonus of 40K free credits. You can also refer a friend (check settings for a link), and you’ll both get 10K free credits.

Compound AI

16,746 görüntüleme • 7 ay önce

Gaelic Football Coaching, South of Ireland Conference 6 days to go!! Tickets selling out(club tickets will fill our limits very fast) We are hosting a very exciting event for gaelic Football coaches in the South of Ireland, and we’d love to see you there. Join us for Coaching the principles of Gaelic Football, South of Ireland Conference, 22 November 2025 at 9:30 AM at Fethard Town Park Sporting Facility, Co. Tipperary. The event is being run by Kevin and is going to be the opening presenter as well. We have an excellent and highly experienced crew of presenters that are going to bring us through what are "principles of play" to coaching the principles of the game at the various ages and levels. The presenters will detail how they develop skills, individuals and principles of play with their teams in a practical applied setting. We have All Ireland winning players and coaches and a wide group of vastly experienced and highly qualified people. There is simply nothing like this for Football lovers and coaches in the South. We want to change this and have an annual flagship event. See below a look into what The Lavin brothers (GameSenseCoaching ) where Stephen outlines what they are going to bring to the conference, delving into why they use Small Sided Games. Check out the link for more details on the day and the presenters. Register soon because space is limited. Space for approx 20-30 individuals or 3/4 clubs at present. We hope you’re able to join us!
0:56

Gaelic Football Coaching, South of Ireland Conference 6 days to go!! Tickets selling out(club tickets will fill our limits very fast) We are hosting a very exciting event for gaelic Football coaches in the South of Ireland, and we’d love to see you there. Join us for Coaching the principles of Gaelic Football, South of Ireland Conference, 22 November 2025 at 9:30 AM at Fethard Town Park Sporting Facility, Co. Tipperary. The event is being run by Kevin and is going to be the opening presenter as well. We have an excellent and highly experienced crew of presenters that are going to bring us through what are "principles of play" to coaching the principles of the game at the various ages and levels. The presenters will detail how they develop skills, individuals and principles of play with their teams in a practical applied setting. We have All Ireland winning players and coaches and a wide group of vastly experienced and highly qualified people. There is simply nothing like this for Football lovers and coaches in the South. We want to change this and have an annual flagship event. See below a look into what The Lavin brothers (GameSenseCoaching ) where Stephen outlines what they are going to bring to the conference, delving into why they use Small Sided Games. Check out the link for more details on the day and the presenters. Register soon because space is limited. Space for approx 20-30 individuals or 3/4 clubs at present. We hope you’re able to join us!

Kevin Mulcahy

43,098 görüntüleme • 9 ay önce

Elon Musk: To do something truly new, you can't reason by analogy. You have to reason from first principles. “I'm a big believer in the science or physics framework of trying to analyze a problem, particularly if it's new. If something's new, then analogies don't work very well, because it's new. The way we get through daily life is mostly by analogy, by copying things with minor variations, because it's a computational shortcut. But if you want to do something that's fundamentally new, or particularly if it's counterintuitive, then you have to do a first principles analysis. You take things from what appear to be the fundamental, simple truths, and then reason up from there. And that requires a lot more thinking, and you can't do that for everything. But if you're trying to figure out something new, I think it's really the only way to go.” Source: American Geophysical Union (AGU) Fall Meeting, San Francisco, December 16, 2015.
1:02

Elon Musk: To do something truly new, you can't reason by analogy. You have to reason from first principles. “I'm a big believer in the science or physics framework of trying to analyze a problem, particularly if it's new. If something's new, then analogies don't work very well, because it's new. The way we get through daily life is mostly by analogy, by copying things with minor variations, because it's a computational shortcut. But if you want to do something that's fundamentally new, or particularly if it's counterintuitive, then you have to do a first principles analysis. You take things from what appear to be the fundamental, simple truths, and then reason up from there. And that requires a lot more thinking, and you can't do that for everything. But if you're trying to figure out something new, I think it's really the only way to go.” Source: American Geophysical Union (AGU) Fall Meeting, San Francisco, December 16, 2015.

ELON CLIPS

34,638 görüntüleme • 2 ay önce

The key to installing a PBO is to keep the main thing the main thing. Explain to your players the most important principles of play and continuously emphasize them throughout the season. These principles will become a constraint or a guide that will get their attention and intentions. Players will self-organize with tasks (constraints) and if you allow them to complete these tasks, you will see them playing great team offense. You don’t need to explain every last detail, just get them started with the basics of shot selection, spacing, ball movement, creating advantages with matchups or 2-man game, and off-ball picking. Using constraints like an 8-9 second shot clock on a side will get them to flip the field. Or, if you’re not picking off-ball it’s a turnover. Or if you hang out on the crease it’s a turnover. These are simple ways to get your team to play incredible and unpredictable team offense. Obviously, installing a PBO in one minute isn’t going to be a time constraint for your average coach, but the point of this reel is you would be amazed at how well your team will play when you take off the reigns and focus on the principles. Throughout the season there will be glaring strengths and weaknesses that you will have to address, but again, use constraints to do the teaching and keep the focus on the Principles.
0:56

The key to installing a PBO is to keep the main thing the main thing. Explain to your players the most important principles of play and continuously emphasize them throughout the season. These principles will become a constraint or a guide that will get their attention and intentions. Players will self-organize with tasks (constraints) and if you allow them to complete these tasks, you will see them playing great team offense. You don’t need to explain every last detail, just get them started with the basics of shot selection, spacing, ball movement, creating advantages with matchups or 2-man game, and off-ball picking. Using constraints like an 8-9 second shot clock on a side will get them to flip the field. Or, if you’re not picking off-ball it’s a turnover. Or if you hang out on the crease it’s a turnover. These are simple ways to get your team to play incredible and unpredictable team offense. Obviously, installing a PBO in one minute isn’t going to be a time constraint for your average coach, but the point of this reel is you would be amazed at how well your team will play when you take off the reigns and focus on the principles. Throughout the season there will be glaring strengths and weaknesses that you will have to address, but again, use constraints to do the teaching and keep the focus on the Principles.

JM3 SPORTS

30,964 görüntüleme • 1 yıl önce

THE ABYSS AHEAD Are we still focused on the ideas that matter—or have we lost our way? In this clip, I talk about what I see as a dangerous shift away from discussing conservatism itself and toward distractions, personalities, and constant reaction. These are serious times. More extreme, more volatile, and more uncertain than in years past. And yet, instead of teaching, debating, and reinforcing core principles, too much of the conversation is spent on everything but the ideas that define a movement. That’s the real risk. Because if we’re not grounded in what we believe—if we’re not actively learning, teaching, and articulating those beliefs—we’re not prepared to meet the challenges ahead. And those challenges are real. This is a call to refocus. To stop playing defense. To get serious about ideas again—and to rally around the principles that matter. To watch the entire episode: YouTube: Rumble:
0:44

THE ABYSS AHEAD Are we still focused on the ideas that matter—or have we lost our way? In this clip, I talk about what I see as a dangerous shift away from discussing conservatism itself and toward distractions, personalities, and constant reaction. These are serious times. More extreme, more volatile, and more uncertain than in years past. And yet, instead of teaching, debating, and reinforcing core principles, too much of the conversation is spent on everything but the ideas that define a movement. That’s the real risk. Because if we’re not grounded in what we believe—if we’re not actively learning, teaching, and articulating those beliefs—we’re not prepared to meet the challenges ahead. And those challenges are real. This is a call to refocus. To stop playing defense. To get serious about ideas again—and to rally around the principles that matter. To watch the entire episode: YouTube: Rumble:

Mark R. Levin

24,227 görüntüleme • 3 ay önce

$PLTR CEO Alex Karp is ripping analysts to SHREDS right now on CNBC 🔮🔥 “Intelligent people learn from their mistakes… everyone else becomes an analyst or an advisor. “Do you want to believe what the librarian says about a book or do you want to go out in the world and see if it works? “They have a huge opinion about how to handle LLMs. It’s completely theoretical and idiotic. If you want to manage LLMs or AI, the infrastructure that allows you to manage them is where the actual value is. “People have been saying we are overvalued for 20 years. People have been saying our products wouldn’t make us profitable. That we would not be able to become a juggernaut. That we would not become GAAP profitable. That we wouldn’t get on the S&P. Keep saying that about us. We love it. It discourages people from doing anything like what we’re doing. And we are winning.”
2:26

$PLTR CEO Alex Karp is ripping analysts to SHREDS right now on CNBC 🔮🔥 “Intelligent people learn from their mistakes… everyone else becomes an analyst or an advisor. “Do you want to believe what the librarian says about a book or do you want to go out in the world and see if it works? “They have a huge opinion about how to handle LLMs. It’s completely theoretical and idiotic. If you want to manage LLMs or AI, the infrastructure that allows you to manage them is where the actual value is. “People have been saying we are overvalued for 20 years. People have been saying our products wouldn’t make us profitable. That we would not be able to become a juggernaut. That we would not become GAAP profitable. That we wouldn’t get on the S&P. Keep saying that about us. We love it. It discourages people from doing anything like what we’re doing. And we are winning.”

Jawwwn

1,198,605 görüntüleme • 1 yıl önce

Islamofascists and gender fluid Marxists forming a red green alliance against the preaching of Jesus Christ in Northern Ireland. We are a Christian country, founded on the principles of Christianity. If you don't like it..... Fuck off to an Islamic country where you truly belong. Support the country you live in or fuck off to a country you support. So sick of this bullshit. #EnoughIsEnough #GoHome #FuckIslam #Palestine #LGBTQ
4:01

Islamofascists and gender fluid Marxists forming a red green alliance against the preaching of Jesus Christ in Northern Ireland. We are a Christian country, founded on the principles of Christianity. If you don't like it..... Fuck off to an Islamic country where you truly belong. Support the country you live in or fuck off to a country you support. So sick of this bullshit. #EnoughIsEnough #GoHome #FuckIslam #Palestine #LGBTQ

Tommy Robinson 🇬🇧

152,696 görüntüleme • 2 ay önce