Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Okay okay, spent my weekend gooning around learning GRPO math. Here's some takes. Essentially, this is me yapping through a recap of smaller details on how GRPO is implemented, what Dr. GRPO changes, why, DAPO, connections to PPO, aggregating batches... Reading list below.

Nathan Lambert

94,765 subscribers

123,113 просмотров • 1 год назад •via X (Twitter)

Образование

Anya Rossi• Live Now

Private livecam show

Комментарии: 10

Фото профиля Nathan Lambert

Nathan Lambert1 год назад

More coherent version coming to @interconnectsai this week. I know this format won't be for everyone, but I hope some of you love it! RLHF Book: DeepSeekMath paper: Where does ratio come from in PPO? DAPO: DAPO announcement: My DAPO recap: Dr. GRPO: Dr. GRPO announcement: TRL GRPO implementation: Unbiased GRPO implementation: Thread on GRPO implementation on x:

Фото профиля Nathan Lambert

Nathan Lambert1 год назад

Watch on YouTube:

Фото профиля Nathan Lambert

Nathan Lambert1 год назад

Thanks to many authors and folks for discussing / proposing questions, @QGallouedec , @ethayarajh , @zzlccc, @hamishivi @vwxyzjn @danielhanchen -- have distilled a lot from y'all in the last 72hours. No, this content isn't really meant for you, you already know this shit :)

Фото профиля Nathan Lambert

Nathan Lambert1 год назад

I know this is B tier production quality but A tier nerding out.

Фото профиля Filip Brnadic

Filip Brnadic1 год назад

What's your crypto exit strategy? I track 30+ indicators that have successfully marked the top of a bull run in prior cycles. My logic is simple. ✅ Hold blue-chip assets during the bull run ✅ Avoid a -70% drawdown by selling near the top. Check it out 👇

Фото профиля Daniel Han

Daniel Han1 год назад

I actually thought Dr GRPO's 1/max_length for eg 1/4096 was to counteract gradient accumulation causing imbalanced losses (like when doing CE masked mean) It makes sense to remove the std() since 1/large(std) reduces impacts of super hard problems - 1/100 (hard) vs 1/1 (easy)

Фото профиля xlr8harder

xlr8harder1 год назад

is phrasing still a thing?

Фото профиля tom

tom1 год назад

you're not the first to goon over RL

Фото профиля Daniel Han

Daniel Han1 год назад

Awesome video - not 2nd tier quality, but 1st tier :)

Фото профиля Itsram10

Itsram101 год назад

That's the gooning I like, that all smart people must do

Похожие видео

Karpathy's prediction about RL is coming true now! He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a higher-dimensional feedback channel. Every major AI lab trains models with RL today (OpenAI, Anthropic, DeepSeek). And their key bottleneck has always been the reward functions. GRPO by DeepSeek worked well for math and code because the environment gave a binary signal. But for real agent tasks, someone still has to hand-code the scoring function. That takes days and breaks every time the pipeline changes. RULER (implemented in OpenPipe ART, 10k stars) addresses the exact problem Karpathy identified. The reward criteria are defined in plain English, and an LLM evaluates each trajectory against that description to provide feedback for training. I trained a Qwen3 1.4B agent that plays 2048 using GRPO with this exact workflow. In this case, the agent saw the board, picked a direction, and RULER evaluated the outcome, all from this natural language definition. You can see the full implementation on GitHub and try it yourself. Here's the ART Repo: (don't forget to star it ⭐ ) Just like RLHF replaced manual rankings and GRPO replaced the critic model, natural language rewards are replacing hand-coded scoring functions. RL reward engineering is now prompt engineering. I wrote a full walkthrough covering RL for LLM agents, from RLHF to GRPO to RULER, in the article below.

Karpathy's prediction about RL is coming true now! He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a higher-dimensional feedback channel. Every major AI lab trains models with RL today (OpenAI, Anthropic, DeepSeek). And their key bottleneck has always been the reward functions. GRPO by DeepSeek worked well for math and code because the environment gave a binary signal. But for real agent tasks, someone still has to hand-code the scoring function. That takes days and breaks every time the pipeline changes. RULER (implemented in OpenPipe ART, 10k stars) addresses the exact problem Karpathy identified. The reward criteria are defined in plain English, and an LLM evaluates each trajectory against that description to provide feedback for training. I trained a Qwen3 1.4B agent that plays 2048 using GRPO with this exact workflow. In this case, the agent saw the board, picked a direction, and RULER evaluated the outcome, all from this natural language definition. You can see the full implementation on GitHub and try it yourself. Here's the ART Repo: (don't forget to star it ⭐ ) Just like RLHF replaced manual rankings and GRPO replaced the critic model, natural language rewards are replacing hand-coded scoring functions. RL reward engineering is now prompt engineering. I wrote a full walkthrough covering RL for LLM agents, from RLHF to GRPO to RULER, in the article below.

Avi Chawla

350,213 просмотров • 2 месяцев назад

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

Andrew Ng

86,457 просмотров • 1 год назад

I had an awesome time interviewing idan shenfeld and Jonas Hübotter from MIT and ETH Zurich about self-distillation. this very promising post-training paradigm where the model acts as its own teacher by conditioning on environment feedback or demonstrations. we cover the SDPO algo for reinforcement learning with rich feedback and SDFT for continual learning without forgetting along with many applications. we dig into how it works, why it's simpler and faster than GRPO, and where this is already showing up in production systems. table of content: 0:00 - what is self distillation 2:50 - idan (MIT) and jonas (ETH Zurich) introduction and motivation 18:40 - different perspective of on-policy self-distillation (presentation) 36:00 - metacognition and specificity in self-distillation 37:24 - very long hard task and self-distillation 42:00 - continual learning with self-distillation (presentation) 1:16:50 - what is next in this research direction? 1:20:00 - is there any experience with subjective feedbacks? 1:22:50 - quality vs number of feedbacks? 1:26:40 - what setting would self-distillation struggle vs GRPO? the slides were super crisp really cool of them to share! enjoy my guys 🌹

I had an awesome time interviewing idan shenfeld and Jonas Hübotter from MIT and ETH Zurich about self-distillation. this very promising post-training paradigm where the model acts as its own teacher by conditioning on environment feedback or demonstrations. we cover the SDPO algo for reinforcement learning with rich feedback and SDFT for continual learning without forgetting along with many applications. we dig into how it works, why it's simpler and faster than GRPO, and where this is already showing up in production systems. table of content: 0:00 - what is self distillation 2:50 - idan (MIT) and jonas (ETH Zurich) introduction and motivation 18:40 - different perspective of on-policy self-distillation (presentation) 36:00 - metacognition and specificity in self-distillation 37:24 - very long hard task and self-distillation 42:00 - continual learning with self-distillation (presentation) 1:16:50 - what is next in this research direction? 1:20:00 - is there any experience with subjective feedbacks? 1:22:50 - quality vs number of feedbacks? 1:26:40 - what setting would self-distillation struggle vs GRPO? the slides were super crisp really cool of them to share! enjoy my guys 🌹

Yacine Mahdid

13,023 просмотров • 3 месяцев назад

This is my annual attempt at learning how to code as a designer lol it’s always on my resolution list But hey… my code is kind of working?? It actually takes real photos?? Progress?? v0

This is my annual attempt at learning how to code as a designer lol it’s always on my resolution list But hey… my code is kind of working?? It actually takes real photos?? Progress?? v0

marisa

107,574 просмотров • 1 год назад

An exciting new course: Fine-tuning and Reinforcement Learning for LLMs: Intro to Post-training, taught by Sharon Zhou, VP of AI at AMD. Available now at Post-training is the key technique used by frontier labs to turn a base LLM--a model trained on massive unlabeled text to predict the next word/token--into a helpful, reliable assistant that can follow instructions. I've also seen many applications where post-training is what turns a demo application that works only 80% of the time into a reliable system that consistently performs. This course will teach you the most important post-training techniques! In this 5 module course, Sharon walks you through the complete post-training pipeline: supervised fine-tuning, reward modeling, RLHF, and techniques like PPO and GRPO. You'll also learn to use LoRA for efficient training, and to design evals that catch problems before and after deployment. Skills you'll gain: - Apply supervised fine-tuning and reinforcement learning (RLHF, PPO, GRPO) to align models to desired behaviors - Use LoRA for efficient fine-tuning without retraining entire models - Prepare datasets and generate synthetic data for post-training - Understand how to operate LLM production pipelines, with go/no-go decision points and feedback loops These advanced methods aren’t limited to frontier AI labs anymore, and you can now use them in your own applications. Learn here:

An exciting new course: Fine-tuning and Reinforcement Learning for LLMs: Intro to Post-training, taught by Sharon Zhou, VP of AI at AMD. Available now at Post-training is the key technique used by frontier labs to turn a base LLM--a model trained on massive unlabeled text to predict the next word/token--into a helpful, reliable assistant that can follow instructions. I've also seen many applications where post-training is what turns a demo application that works only 80% of the time into a reliable system that consistently performs. This course will teach you the most important post-training techniques! In this 5 module course, Sharon walks you through the complete post-training pipeline: supervised fine-tuning, reward modeling, RLHF, and techniques like PPO and GRPO. You'll also learn to use LoRA for efficient training, and to design evals that catch problems before and after deployment. Skills you'll gain: - Apply supervised fine-tuning and reinforcement learning (RLHF, PPO, GRPO) to align models to desired behaviors - Use LoRA for efficient fine-tuning without retraining entire models - Prepare datasets and generate synthetic data for post-training - Understand how to operate LLM production pipelines, with go/no-go decision points and feedback loops These advanced methods aren’t limited to frontier AI labs anymore, and you can now use them in your own applications. Learn here:

Andrew Ng

132,304 просмотров • 9 месяцев назад

🐈‍⬛ some of you just keep on saying like 6 7.. 6 or 7, 6 or 7. enhypen is what. like, i’m okay. i actually don’t care so i’m okay. enhypen is enhypen. i actually don’t care so… i always try my best to ignore that so i’m okay but don’t do that to other members especially on their birthdays. like, i’m okay but please don’t do that to others, okay? 🐈‍⬛ i really don’t want them to feel bad on their birthday, okay? if you’re saying that on someone’s birthday… like i’m okay, i feel good, i feel good with all my fans. today was such a happy birthday for me but i’m really nervous that what if other members would feel depressed in this situation. i’m really scared for that so hopefully i don’t want that happening in the future, on the other member’s birthday, like this kind of things so… the only thing i wanna say is, let’s see the situation. let’s look around, okay? let’s just look around. like, what is going on. 🐈‍⬛ i’m fine, i’m okay. i don’t actually care. i’m fine but i just want other members to be happy with no risks, okay? that’s the reason

🐈‍⬛ some of you just keep on saying like 6 7.. 6 or 7, 6 or 7. enhypen is what. like, i’m okay. i actually don’t care so i’m okay. enhypen is enhypen. i actually don’t care so… i always try my best to ignore that so i’m okay but don’t do that to other members especially on their birthdays. like, i’m okay but please don’t do that to others, okay? 🐈‍⬛ i really don’t want them to feel bad on their birthday, okay? if you’re saying that on someone’s birthday… like i’m okay, i feel good, i feel good with all my fans. today was such a happy birthday for me but i’m really nervous that what if other members would feel depressed in this situation. i’m really scared for that so hopefully i don’t want that happening in the future, on the other member’s birthday, like this kind of things so… the only thing i wanna say is, let’s see the situation. let’s look around, okay? let’s just look around. like, what is going on. 🐈‍⬛ i’m fine, i’m okay. i don’t actually care. i’m fine but i just want other members to be happy with no risks, okay? that’s the reason

ᰔ

127,797 просмотров • 3 месяцев назад

Chamath: US AI Startups Can Learn A Lot from DeepSeek "This is a case where necessity was the mother of invention." On E213, Chamath Palihapitiya explained how, even if the $6M number is inaccurate, DeepSeek still had some impressive breakthroughs: GRPO > PPO for reinforcement learning "These guys were like, 'Well, how am I going to do this reinforcement learning thing?' They invented a totally different algorithm." "It uses a lot less computer memory and it's highly performant." PTX > CUDA to build the model "And then the second thing that was crazy is everybody is used to building models and compiling through CUDA, which is NVIDIA's proprietary language." "And these guys worked totally around CUDA and they did something called PTX, which goes right to the bare metal." What US AI startups can learn from this: "We, meaning the West, with all the money that we've had, didn't come up with these ideas." "And I think part of why we didn't come up is not that we're not smart enough to do it, but we weren't forced to because the constraints didn't exist." "And so I just wonder how we make sure we learn this principle, meaning, when the AI company wakes up and rolls out of bed and some VC gives them $200M, maybe that's not the right answer for a Series A or a Seed." "And maybe the right answer is $2M so that they do these DeepSeek-like innovations."

Chamath: US AI Startups Can Learn A Lot from DeepSeek "This is a case where necessity was the mother of invention." On E213, Chamath Palihapitiya explained how, even if the $6M number is inaccurate, DeepSeek still had some impressive breakthroughs: GRPO > PPO for reinforcement learning "These guys were like, 'Well, how am I going to do this reinforcement learning thing?' They invented a totally different algorithm." "It uses a lot less computer memory and it's highly performant." PTX > CUDA to build the model "And then the second thing that was crazy is everybody is used to building models and compiling through CUDA, which is NVIDIA's proprietary language." "And these guys worked totally around CUDA and they did something called PTX, which goes right to the bare metal." What US AI startups can learn from this: "We, meaning the West, with all the money that we've had, didn't come up with these ideas." "And I think part of why we didn't come up is not that we're not smart enough to do it, but we weren't forced to because the constraints didn't exist." "And so I just wonder how we make sure we learn this principle, meaning, when the AI company wakes up and rolls out of bed and some VC gives them $200M, maybe that's not the right answer for a Series A or a Seed." "And maybe the right answer is $2M so that they do these DeepSeek-like innovations."

The All-In Podcast

108,815 просмотров • 1 год назад

What some people miss while reading Acts is how the Apostle Paul essentially refused to save his own life. His "why" ended up leading to one of the most historically significant events in the history of Christianity and the world. Join me for a mini-Bible study 👇

What some people miss while reading Acts is how the Apostle Paul essentially refused to save his own life. His "why" ended up leading to one of the most historically significant events in the history of Christianity and the world. Join me for a mini-Bible study 👇

Josh Howerton

61,871 просмотров • 3 месяцев назад

i swear why would you do this to him.. on HIS BIRTHDAY?! 😢 🐈‍⬛ some of you guys are just like, you actually don't care about what the situation is going on, right? like, you guys keep like, saying like, 6, 7, 6 or 7, 6 or 7, enhypen is what, enhypen is what. like, i'm okay. i actually don't care, so i'm okay. enhyoen is enhypen. and like i actually don't care so i'm okay, i always try my best to ignore that so i'm okay but don't do that to other members especially at their birthday like i'm okay but please don't do that to other others. i really don't want them to feel that in their birthday okay just like um like thinking is totally like free but if you're saying that in someone's birthday like i'm okay, i'm saying this again. i'm okay but i feel good, i feel good with all my fans. today's was such a happy birthday for me, but i'm really like nervous that what if other members.. what would feel like depressed in this situation? like i'm really scared for that so hopefully i don't want that happen in the in the future in other members birthday like in this kind of things like so the only thing i want to say is like let's see the situation like um let's look around okay, let's just look around like what is going on, i hopefully like yeah.. that's it like i'm fine i'm okay. i'm okay. i don't actually care. like, i'm fine. but i just want other members to be happy with no risks, okay? that's the reason.

i swear why would you do this to him.. on HIS BIRTHDAY?! 😢 🐈‍⬛ some of you guys are just like, you actually don't care about what the situation is going on, right? like, you guys keep like, saying like, 6, 7, 6 or 7, 6 or 7, enhypen is what, enhypen is what. like, i'm okay. i actually don't care, so i'm okay. enhyoen is enhypen. and like i actually don't care so i'm okay, i always try my best to ignore that so i'm okay but don't do that to other members especially at their birthday like i'm okay but please don't do that to other others. i really don't want them to feel that in their birthday okay just like um like thinking is totally like free but if you're saying that in someone's birthday like i'm okay, i'm saying this again. i'm okay but i feel good, i feel good with all my fans. today's was such a happy birthday for me, but i'm really like nervous that what if other members.. what would feel like depressed in this situation? like i'm really scared for that so hopefully i don't want that happen in the in the future in other members birthday like in this kind of things like so the only thing i want to say is like let's see the situation like um let's look around okay, let's just look around like what is going on, i hopefully like yeah.. that's it like i'm fine i'm okay. i'm okay. i don't actually care. like, i'm fine. but i just want other members to be happy with no risks, okay? that's the reason.

ɪᴄɪᴇʟ. 🐚

79,503 просмотров • 3 месяцев назад

How can daeum leave when she knows she’s yet to receive so much love from the people around her? How is she supposed to act okay? “I want to live so badly” just put a knife through my heart it’ll hurt less this broke me #ourmovieep10 #ourmovie

How can daeum leave when she knows she’s yet to receive so much love from the people around her? How is she supposed to act okay? “I want to live so badly” just put a knife through my heart it’ll hurt less this broke me #ourmovieep10 #ourmovie

𓍯

11,481 просмотров • 1 год назад

Now that the Sallah weekend is over, I need to go on a diet and to lose the calories I gained eating meat around Ibadan. What a weekend it was. Spending time with constituents is one of my favorite parts of this job. That’s why I go back every weekend.

Now that the Sallah weekend is over, I need to go on a diet and to lose the calories I gained eating meat around Ibadan. What a weekend it was. Spending time with constituents is one of my favorite parts of this job. That’s why I go back every weekend.

Oloye Akin Alabi

21,061 просмотров • 1 год назад

A Wide Awake Message To Americans “Donald Trump is convicted before anyone on Jeffrey Epstein’s client list. Okay. We have politicians that go into office worth about $2, and now they're millionaires. Yet he's the one being funny with money? Okay. You charge him for allegedly paying people to hush when you have others who literally have unexplained bodies dropping around them. Okay. Some of you have been nothing but quiet until now, afraid to speak out, but somehow believe posting a flag upside down is gonna turn this whole thing around. Okay. I think most of us can agree what's going on is not okay, and no one is coming to save the day. If it wasn't gonna be before, unless they do something drastic, it's Trump 2024, for show. Take action.”

A Wide Awake Message To Americans “Donald Trump is convicted before anyone on Jeffrey Epstein’s client list. Okay. We have politicians that go into office worth about $2, and now they're millionaires. Yet he's the one being funny with money? Okay. You charge him for allegedly paying people to hush when you have others who literally have unexplained bodies dropping around them. Okay. Some of you have been nothing but quiet until now, afraid to speak out, but somehow believe posting a flag upside down is gonna turn this whole thing around. Okay. I think most of us can agree what's going on is not okay, and no one is coming to save the day. If it wasn't gonna be before, unless they do something drastic, it's Trump 2024, for show. Take action.”

Wall Street Apes

1,352,973 просмотров • 2 лет назад

this is the "brain of AI" Claude is made of billions of these artificial neurons. no one knows exactly how they work inside. that's why it's called a black box. you give it an input, it fires through billions of connections, something comes out the other side nobody fully understands why it works. but here's what we do know: the brain is the same for everyone what separates the top 1% isn't a better model it's the setup around it ClaudeKit gives Claude Code that setup. ( full breakdown in the article below

this is the "brain of AI" Claude is made of billions of these artificial neurons. no one knows exactly how they work inside. that's why it's called a black box. you give it an input, it fires through billions of connections, something comes out the other side nobody fully understands why it works. but here's what we do know: the brain is the same for everyone what separates the top 1% isn't a better model it's the setup around it ClaudeKit gives Claude Code that setup. ( full breakdown in the article below

Hamza Khalid

175,865 просмотров • 1 месяц назад

My Girlfriend Gave Me An HIV; Not Sure Where To Go From Here ( Should We Stay Together?) Caller (Tim):so my question is, my girlfriend gave me a... a quite serious STD, and my parents found out about this. They are now against our relationship saying that it's a way that God is showing that this relationship was not meant to be. Um, I've been fighting this like against my parents. And my question is, has... is there too much that has happened for this relationship to work? Uh, or am I trying too hard to make it work? Dr. John Delony: Bro, there's a lot going on here, man. Whole lot. Um, before I dig in, let's... let me just... let's just have a conversation so I can get to know you for a second. Is that cool? Caller: Okay, yeah. Dr. John Delony: Um, you sound a little bit nervous. Are you nervous, or is this just weird? Or, I realize you're calling a stranger being like, "Yeah, I got an STI from this woman that I love." This whole thing's kind of weird. Caller: Well, no. This is basically the first person outside of this situation that I've talked to, so... Dr. John Delony: Gotcha. Well, dude, I'm... I'm grateful for the trust. It sounds like there's a lot, lot going on here. Um, tell me about you. How old are you? What do you do for a living? All that kind of stuff. Caller: Uh, 27. Um, uh, going to... uh, studying nursing. Dr. John Delony: Okay. Caller: And, uh, just about to graduate here. Dr. John Delony: So, congratulations on that. Um, do you live at home? Caller: No, I don't. Dr. John Delony: Okay. So, what kind of STI did you get? Caller: Uh, HIV. Dr. John Delony: Okay. So, you've got HIV. Okay, so I was wondering if this was... The... my first question was why in the world do your parents know? Um, if I'm 27, the last people on planet Earth I would call and tell I have an STI is my parents. But, um, HIV is super serious. What is your, um, health prognosis? Caller: it's... so, I guess the situation gets a little bit more difficult because, my dad does computer diagnostics, and he was able to accidentally kind of find out I have this, and so I am... Dr. John Delony: Accidentally kind of? What does that mean? Caller: Meaning he wasn't expecting it. He wasn't looking for it. It was just kind of came up when, uh, he was doing the... I regularly do checkups. He's a... a doctor. Dr. John Delony: Oh, so he's a doctor at the place where you went to get checked out, and he discovered through the computer system that his son has HIV? Caller: Yes. it's a... alternative medicine that he does, and so it's... it's a very different kind from modern medicine, so it's just the way that he found out. Um, I just go to my dad for all his, checkups. Dr. John Delony: Yeah. What a mess. Okay, so tell me about this person that you're dating. Caller: Um, so... um... I guess what... uh, what would you like to know about her? Dr. John Delony: Well, just tell me about her. Caller: Well, uh, she's a... she's a nurse. Um, and, uh, Illinois, and, um... This is the second, uh, serious relationship she's had. Dr. John Delony: Okay. Caller: Uh, for me, it's... um, maybe the first serious relationship for me. Dr. John Delony: Okay. Caller: Um, and so, she's... I mean, she grew up in a Christian family. She's... um... Dr. John Delony: Did she knowingly... knowingly give you HIV? Caller: No, she didn't know about it. Dr. John Delony: Okay. Is she the first person you've ever slept with? Caller: Yes. Dr. John Delony: Okay. How long have y'all been together?

My Girlfriend Gave Me An HIV; Not Sure Where To Go From Here ( Should We Stay Together?) Caller (Tim):so my question is, my girlfriend gave me a... a quite serious STD, and my parents found out about this. They are now against our relationship saying that it's a way that God is showing that this relationship was not meant to be. Um, I've been fighting this like against my parents. And my question is, has... is there too much that has happened for this relationship to work? Uh, or am I trying too hard to make it work? Dr. John Delony: Bro, there's a lot going on here, man. Whole lot. Um, before I dig in, let's... let me just... let's just have a conversation so I can get to know you for a second. Is that cool? Caller: Okay, yeah. Dr. John Delony: Um, you sound a little bit nervous. Are you nervous, or is this just weird? Or, I realize you're calling a stranger being like, "Yeah, I got an STI from this woman that I love." This whole thing's kind of weird. Caller: Well, no. This is basically the first person outside of this situation that I've talked to, so... Dr. John Delony: Gotcha. Well, dude, I'm... I'm grateful for the trust. It sounds like there's a lot, lot going on here. Um, tell me about you. How old are you? What do you do for a living? All that kind of stuff. Caller: Uh, 27. Um, uh, going to... uh, studying nursing. Dr. John Delony: Okay. Caller: And, uh, just about to graduate here. Dr. John Delony: So, congratulations on that. Um, do you live at home? Caller: No, I don't. Dr. John Delony: Okay. So, what kind of STI did you get? Caller: Uh, HIV. Dr. John Delony: Okay. So, you've got HIV. Okay, so I was wondering if this was... The... my first question was why in the world do your parents know? Um, if I'm 27, the last people on planet Earth I would call and tell I have an STI is my parents. But, um, HIV is super serious. What is your, um, health prognosis? Caller: it's... so, I guess the situation gets a little bit more difficult because, my dad does computer diagnostics, and he was able to accidentally kind of find out I have this, and so I am... Dr. John Delony: Accidentally kind of? What does that mean? Caller: Meaning he wasn't expecting it. He wasn't looking for it. It was just kind of came up when, uh, he was doing the... I regularly do checkups. He's a... a doctor. Dr. John Delony: Oh, so he's a doctor at the place where you went to get checked out, and he discovered through the computer system that his son has HIV? Caller: Yes. it's a... alternative medicine that he does, and so it's... it's a very different kind from modern medicine, so it's just the way that he found out. Um, I just go to my dad for all his, checkups. Dr. John Delony: Yeah. What a mess. Okay, so tell me about this person that you're dating. Caller: Um, so... um... I guess what... uh, what would you like to know about her? Dr. John Delony: Well, just tell me about her. Caller: Well, uh, she's a... she's a nurse. Um, and, uh, Illinois, and, um... This is the second, uh, serious relationship she's had. Dr. John Delony: Okay. Caller: Uh, for me, it's... um, maybe the first serious relationship for me. Dr. John Delony: Okay. Caller: Um, and so, she's... I mean, she grew up in a Christian family. She's... um... Dr. John Delony: Did she knowingly... knowingly give you HIV? Caller: No, she didn't know about it. Dr. John Delony: Okay. Is she the first person you've ever slept with? Caller: Yes. Dr. John Delony: Okay. How long have y'all been together?

Hecto Crypto | NetLink ⛓

14,886 просмотров • 17 дней назад

Another piece from the Sasha Rush conversation, this time on rewards for coding RL. He said Cursor uses a mix. Some rewards look at the tool calls themselves, some only at the final output. Everything end-to-end, no process rewards guessing what happens in the middle. I agree with him that the next push (and the current one) is going past verifiers. Coding is a good example where you get both. You have the 0/1 signals, but also a ton of soft ones. Does the diff read well? Did the agent burn context for no reason? Is the PR the right size? Did it rewrite a file it didn't need to touch? So many signals sitting there, and we need to figure out how to actually train on them. This is also why I think the environment (and the data) is becoming the bottleneck, not the algorithm. GRPO works. PPO works. The hard part is building a sandbox that exposes enough signal to learn from. We're already seeing this, and it'll only accelerate: a whole category of companies selling environments is emerging, and they have a good reward to do so…

Another piece from the Sasha Rush conversation, this time on rewards for coding RL. He said Cursor uses a mix. Some rewards look at the tool calls themselves, some only at the final output. Everything end-to-end, no process rewards guessing what happens in the middle. I agree with him that the next push (and the current one) is going past verifiers. Coding is a good example where you get both. You have the 0/1 signals, but also a ton of soft ones. Does the diff read well? Did the agent burn context for no reason? Is the PR the right size? Did it rewrite a file it didn't need to touch? So many signals sitting there, and we need to figure out how to actually train on them. This is also why I think the environment (and the data) is becoming the bottleneck, not the algorithm. GRPO works. PPO works. The hard part is building a sandbox that exposes enough signal to learn from. We're already seeing this, and it'll only accelerate: a whole category of companies selling environments is emerging, and they have a good reward to do so…

Ravid Shwartz Ziv

14,826 просмотров • 3 месяцев назад

🐈‍⬛ like.. what should i do? what should i do, hmm? 🐈‍⬛ some of you guys just.. you actually don’t care about what’s.. the situation is going on, right? 🐈‍⬛ like.. you guys keep… some of you just keep saying six seven, six or seven, six or seven, enhypen is what, enhypen is what 🐈‍⬛ i’m okay, i actually don’t care so i’m okay. enhypen is enhypen. i’ll always try my best to ignore that, so i’m okay, but don’t do that to other members, especially at their birthday 🐈‍⬛ like.. i’m okay but please don’t do that to other others, okay? i really don’t want them to feel that in their birthday, okay? thinking is totally free, but if you’re saying that in someone’s birthday.. like… 🐈‍⬛ i’m okay, i’m saying this again, i’m okay but.. i feel good, i feel good with all my fans. today’s such a happy birthday for me but i’m really nervous that.. what if other members would feel depressed in this situations? i’m really scared for that, so hopefully i don’t want that happen in the future in other members’ birthday

🐈‍⬛ like.. what should i do? what should i do, hmm? 🐈‍⬛ some of you guys just.. you actually don’t care about what’s.. the situation is going on, right? 🐈‍⬛ like.. you guys keep… some of you just keep saying six seven, six or seven, six or seven, enhypen is what, enhypen is what 🐈‍⬛ i’m okay, i actually don’t care so i’m okay. enhypen is enhypen. i’ll always try my best to ignore that, so i’m okay, but don’t do that to other members, especially at their birthday 🐈‍⬛ like.. i’m okay but please don’t do that to other others, okay? i really don’t want them to feel that in their birthday, okay? thinking is totally free, but if you’re saying that in someone’s birthday.. like… 🐈‍⬛ i’m okay, i’m saying this again, i’m okay but.. i feel good, i feel good with all my fans. today’s such a happy birthday for me but i’m really nervous that.. what if other members would feel depressed in this situations? i’m really scared for that, so hopefully i don’t want that happen in the future in other members’ birthday

peanut..

181,625 просмотров • 3 месяцев назад

Incase anyone didn't want to read through my screenshots here is my video of my tab in the doc plus some more information I couldn't say before. Thank you to everyone who made me feel okay to speak out and make a video on this 😭 (Sorry it's not edited fancy I am tired)

Incase anyone didn't want to read through my screenshots here is my video of my tab in the doc plus some more information I couldn't say before. Thank you to everyone who made me feel okay to speak out and make a video on this 😭 (Sorry it's not edited fancy I am tired)

🔪🐋 Orca | 🌋Soon | VHorrorS

14,254 просмотров • 1 год назад

🚨 Heads up, Utah — this is not okay. 🚨 The UEA is creating an “Equity Ambassador” program for teachers to “advance racial and social justice work in public schools.” They are pushing to lead "systemic change" on how and what our kids are being taught. Our children deserve classrooms devoted to learning — math, science, reading, writing, history, etc. — not political indoctrination. And this is just part of a larger national issue. Watch the video below to see what I’m talking about. As a father of an educator, I know this is not who our teachers are, so why are we letting the union take control? ❗️Let me be blunt: Schools are for education, not agendas. If you feel the same, share this post. Ask your district and school board: 1. Why is the UEA trying to advance racial and social justice work in public schools? 2. And how will that affect what our children are taught? Our kids deserve a real education. Let’s defend that.💪

🚨 Heads up, Utah — this is not okay. 🚨 The UEA is creating an “Equity Ambassador” program for teachers to “advance racial and social justice work in public schools.” They are pushing to lead "systemic change" on how and what our kids are being taught. Our children deserve classrooms devoted to learning — math, science, reading, writing, history, etc. — not political indoctrination. And this is just part of a larger national issue. Watch the video below to see what I’m talking about. As a father of an educator, I know this is not who our teachers are, so why are we letting the union take control? ❗️Let me be blunt: Schools are for education, not agendas. If you feel the same, share this post. Ask your district and school board: 1. Why is the UEA trying to advance racial and social justice work in public schools? 2. And how will that affect what our children are taught? Our kids deserve a real education. Let’s defend that.💪

Mike Schultz

64,607 просмотров • 10 месяцев назад

#RHOBH after show! Boz says Amanda operates in theory, while she operates in practicality. Boz adds that Amanda is “selling an idea,” whereas she is “actually giving practical advice” based on her experience. Amanda was asked why she thinks Boz is so interested in her business and responded, “I think [Boz] had a hard time selling courses… I’ve spent my entire life learning how to do this. So it’s okay if she doesn’t know how to do it. I know how to do it, but to assume I’m illegitimate because her courses aren’t doing what mine do is not fair.”

#RHOBH after show! Boz says Amanda operates in theory, while she operates in practicality. Boz adds that Amanda is “selling an idea,” whereas she is “actually giving practical advice” based on her experience. Amanda was asked why she thinks Boz is so interested in her business and responded, “I think [Boz] had a hard time selling courses… I’ve spent my entire life learning how to do this. So it’s okay if she doesn’t know how to do it. I know how to do it, but to assume I’m illegitimate because her courses aren’t doing what mine do is not fair.”

Bye Wig Hello Drama

17,197 просмотров • 4 месяцев назад

What Hani typed ⌨️ 📎 Hello, my name is Seunghan. I'm 23 this year and 24 next year. How are you spending this time, whether it's fast or slow? I hope you all have a comfortable rest this weekend and continue to be happy. This is my small wish. Meditation and reading are good ways to find peace of mind, but I recommend listening to music. I find peace of mind by listening to music. If you've had a hard day, put on your favorite song, don't think about anything else, and focus on the song. Then everything will be okay.

What Hani typed ⌨️ 📎 Hello, my name is Seunghan. I'm 23 this year and 24 next year. How are you spending this time, whether it's fast or slow? I hope you all have a comfortable rest this weekend and continue to be happy. This is my small wish. Meditation and reading are good ways to find peace of mind, but I recommend listening to music. I find peace of mind by listening to music. If you've had a hard day, put on your favorite song, don't think about anything else, and focus on the song. Then everything will be okay.

아란

17,001 просмотров • 8 месяцев назад