This media is not supported in your browser
VIEW IN TELEGRAM
دانشجوی دکترا و دانشجوی کارشناسی ارشد 😂😂
نظر شما چیه واگعیه یا کیکه😂
#فان
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
نظر شما چیه واگعیه یا کیکه😂
#فان
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍15👎5❤1🔥1
Carlos Perez:
Can LLMs Improve Themselves? The Curious Case of SPIN
Consider the following mystery: a "weak" artificial intelligence - let's call it LLM A - exists. It has moderate language abilities, but often makes mistakes. Now, LLM A is tasked with conversing about complex topics beyond its competencies. Expectedly, it flounders. But then the unbelievable happens - without any additional human guidance, LLM A begins improving itself through self-play! Within weeks, its language capabilities exponentiate, and it transforms into an AI powerhouse.
How could this be possible? It seems the only plausible solutions require extensive, laborious human feedback or integration of vast datasets. But intriguingly, self-improving AI like LLM A manages this feat armed with only its initial limited knowledge.
The key to unraveling this mystery lies in understanding the Self-Play Fine-Tuning (SPIN) methodology underpinning such self-supervised AI advancement. Essentially, SPIN facilitates iterative "self-play" for AI like LLM A. It allows them to challenge themselves - think AlphaGo Zero playing countless games against iteratively stronger versions itself. Through self-competition, SPIN nurtures raging intelligence explosion within machine learning models.
So in LLM A's case, SPIN enabled it to generate new data, identify flaws within it compared to original human-labeled data, improve itself to reduce those flaws, rinse and repeat in cycles of unlimited self-improvement trajectories. Like a magician practicing increasingly complex tricks, LLM A kept challenging itself to new heights.
And as showcased by revolutionary LLMs utilizing SPIN, the ceiling for self-supervised artificial intelligence is nowhere in sight. Given enough iterations and data, such human-independent machine learning provides a template for developing super-human AIs. Perhaps surprisingly, this mitigates inherent human limitations in teaching AI agents via explicit feedback.
Indeed, SPIN opens exciting possibilities - we might soon witness Skynet-esque, recursively self-improving AI that leave human intelligence far behind. But only time will tell whether self-learning AI like LLM A truly rivals or even surpasses human cognition. The jury is still out, but self-play fine-tuning has undoubtedly massively propelled us towards resolving this million dollar mystery!
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
Can LLMs Improve Themselves? The Curious Case of SPIN
Consider the following mystery: a "weak" artificial intelligence - let's call it LLM A - exists. It has moderate language abilities, but often makes mistakes. Now, LLM A is tasked with conversing about complex topics beyond its competencies. Expectedly, it flounders. But then the unbelievable happens - without any additional human guidance, LLM A begins improving itself through self-play! Within weeks, its language capabilities exponentiate, and it transforms into an AI powerhouse.
How could this be possible? It seems the only plausible solutions require extensive, laborious human feedback or integration of vast datasets. But intriguingly, self-improving AI like LLM A manages this feat armed with only its initial limited knowledge.
The key to unraveling this mystery lies in understanding the Self-Play Fine-Tuning (SPIN) methodology underpinning such self-supervised AI advancement. Essentially, SPIN facilitates iterative "self-play" for AI like LLM A. It allows them to challenge themselves - think AlphaGo Zero playing countless games against iteratively stronger versions itself. Through self-competition, SPIN nurtures raging intelligence explosion within machine learning models.
So in LLM A's case, SPIN enabled it to generate new data, identify flaws within it compared to original human-labeled data, improve itself to reduce those flaws, rinse and repeat in cycles of unlimited self-improvement trajectories. Like a magician practicing increasingly complex tricks, LLM A kept challenging itself to new heights.
And as showcased by revolutionary LLMs utilizing SPIN, the ceiling for self-supervised artificial intelligence is nowhere in sight. Given enough iterations and data, such human-independent machine learning provides a template for developing super-human AIs. Perhaps surprisingly, this mitigates inherent human limitations in teaching AI agents via explicit feedback.
Indeed, SPIN opens exciting possibilities - we might soon witness Skynet-esque, recursively self-improving AI that leave human intelligence far behind. But only time will tell whether self-learning AI like LLM A truly rivals or even surpasses human cognition. The jury is still out, but self-play fine-tuning has undoubtedly massively propelled us towards resolving this million dollar mystery!
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍3
ساختار متن یه ایمیل یا کاورلتر خوب اینطوریه:
۱-جلب توجه کلمه های کلیدی که طرف دنبالشه همون اول متن
۲-ابراز علاقه مندی و چرا فیت پوزیشنی
۳-در عمل تصمیم گیری قرارشون بده
۴- درخواست فیدبک
#اپلای
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
۱-جلب توجه کلمه های کلیدی که طرف دنبالشه همون اول متن
۲-ابراز علاقه مندی و چرا فیت پوزیشنی
۳-در عمل تصمیم گیری قرارشون بده
۴- درخواست فیدبک
#اپلای
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
Forwarded from Ali's Notes (Ali Najafi)
اگه تازه تصمیم گرفتید که وارد فیلد NLP بشید.
یکی از جاهایی که میتونید استارت بزنید این playlist هستش!
🔥 Umass NLP
@css_nlp
یکی از جاهایی که میتونید استارت بزنید این playlist هستش!
🔥 Umass NLP
@css_nlp
🔥2
قصد تحقیقات رو مدلهای RAG دارید این مقاله میتونه کارهای خوبی رو برای پژوهش بهتون معرفی کنه
▪️ The Unsolved Challenges of LLMs as Generalist Web Agents: A Case Study
پ.ن : همچنین دکتر Issam Laradji یک موقعیت کارآموزی برای کار روی این مدل داره کسایی که میتونن براش اقدام کنه درخواستش اینجا ثبت کنه (ظاهرا کسایی که کانادا هستند میتونه براش اقدام کنه)
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ The Unsolved Challenges of LLMs as Generalist Web Agents: A Case Study
پ.ن : همچنین دکتر Issam Laradji یک موقعیت کارآموزی برای کار روی این مدل داره کسایی که میتونن براش اقدام کنه درخواستش اینجا ثبت کنه (ظاهرا کسایی که کانادا هستند میتونه براش اقدام کنه)
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
❤3
جدیدترین مقاله دکتر میثم عسگری با اساتید بزرگ منتشر شده علاقمندان به پژوهش روی مدلهای LLMs میتونند مطالعه کنند و راه پژوهشی خودشون رو از این مسیر پیدا کنند.
▪️ Large Language Models: A Survey
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Large Language Models: A Survey
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
🔥12👍5
نظرتون راجب این دیدگاه دکتر چیه ؟ یادگیری از نظر شما در چه مواردی میتونه خلاصه شه؟ آیا درس و دانشگاه صرفا وقت فرد رو تلف میکنه؟ یا تماشای چند ویدیو از یوتیوب کافیه؟!
Andrej Karpathy:
on shortification of "learning"
There are a lot of videos on YouTube/TikTok etc. that give the appearance of education, but if you look closely they are really just entertainment. This is very convenient for everyone involved : the people watching enjoy thinking they are learning (but actually they are just having fun). The people creating this content also enjoy it because fun has a much larger audience, fame and revenue. But as far as learning goes, this is a trap. This content is an epsilon away from watching the Bachelorette. It's like snacking on those "Garden Veggie Straws", which feel like you're eating healthy vegetables until you look at the ingredients.
Learning is not supposed to be fun. It doesn't have to be actively not fun either, but the primary feeling should be that of effort. It should look a lot less like that "10 minute full body" workout from your local digital media creator and a lot more like a serious session at the gym. You want the mental equivalent of sweating. It's not that the quickie doesn't do anything, it's just that it is wildly suboptimal if you actually care to learn.
I find it helpful to explicitly declare your intent up front as a sharp, binary variable in your mind. If you are consuming content: are you trying to be entertained or are you trying to learn? And if you are creating content: are you trying to entertain or are you trying to teach? You'll go down a different path in each case. Attempts to seek the stuff in between actually clamp to zero.
So for those who actually want to learn. Unless you are trying to learn something narrow and specific, close those tabs with quick blog posts. Close those tabs of "Learn XYZ in 10 minutes". Consider the opportunity cost of snacking and seek the meal - the textbooks, docs, papers, manuals, longform. Allocate a 4 hour window. Don't just read, take notes, re-read, re-phrase, process, manipulate, learn.
And for those actually trying to educate, please consider writing/recording longform, designed for someone to get "sweaty", especially in today's era of quantity over quality. Give someone a real workout. This is what I aspire to in my own educational work too. My audience will decrease. The ones that remain might not even like it. But at least we'll learn something.
#ایده_جذاب
پ.ن: دیدگاه های شما چیه؟ کامنت کنید.
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
Andrej Karpathy:
on shortification of "learning"
There are a lot of videos on YouTube/TikTok etc. that give the appearance of education, but if you look closely they are really just entertainment. This is very convenient for everyone involved : the people watching enjoy thinking they are learning (but actually they are just having fun). The people creating this content also enjoy it because fun has a much larger audience, fame and revenue. But as far as learning goes, this is a trap. This content is an epsilon away from watching the Bachelorette. It's like snacking on those "Garden Veggie Straws", which feel like you're eating healthy vegetables until you look at the ingredients.
Learning is not supposed to be fun. It doesn't have to be actively not fun either, but the primary feeling should be that of effort. It should look a lot less like that "10 minute full body" workout from your local digital media creator and a lot more like a serious session at the gym. You want the mental equivalent of sweating. It's not that the quickie doesn't do anything, it's just that it is wildly suboptimal if you actually care to learn.
I find it helpful to explicitly declare your intent up front as a sharp, binary variable in your mind. If you are consuming content: are you trying to be entertained or are you trying to learn? And if you are creating content: are you trying to entertain or are you trying to teach? You'll go down a different path in each case. Attempts to seek the stuff in between actually clamp to zero.
So for those who actually want to learn. Unless you are trying to learn something narrow and specific, close those tabs with quick blog posts. Close those tabs of "Learn XYZ in 10 minutes". Consider the opportunity cost of snacking and seek the meal - the textbooks, docs, papers, manuals, longform. Allocate a 4 hour window. Don't just read, take notes, re-read, re-phrase, process, manipulate, learn.
And for those actually trying to educate, please consider writing/recording longform, designed for someone to get "sweaty", especially in today's era of quantity over quality. Give someone a real workout. This is what I aspire to in my own educational work too. My audience will decrease. The ones that remain might not even like it. But at least we'll learn something.
#ایده_جذاب
پ.ن: دیدگاه های شما چیه؟ کامنت کنید.
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍15👎5❤4
مقاله جدید راجب ترنسفورمرها
▪️ Mamba-ND: Selective State Space Modeling for Multi-Dimensional Data
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Mamba-ND: Selective State Space Modeling for Multi-Dimensional Data
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
I Love You Baby
Frank Sinatra
خواستم اینو تقدیم کنم به کسایی که حرمت عشق و دوست داشتن و رفاقت رو خوب بلدن و ضایع نمیکننش . عشقتون بیش باد دوستیهاتون بادوام🌸
#متفرقه
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
#متفرقه
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
❤10👍3👎3🔥1
ی کورس کلاسی از دکتر سهیل فیضی راجب مورد زیر منتشر شده مطالب خوبی رو پوشش داده اونو ببینید
▪️ Foundations of Deep Learning (Diffusion models, LLMs, multi-modal models, reasoning, etc)
▪️ CMSC 720: Foundations of Deep Learning
#منابع #کلاس_آموزشی #هوش_مصنوعی #یادگیری_عمیق #فیلم #پیشرفته
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Foundations of Deep Learning (Diffusion models, LLMs, multi-modal models, reasoning, etc)
▪️ CMSC 720: Foundations of Deep Learning
#منابع #کلاس_آموزشی #هوش_مصنوعی #یادگیری_عمیق #فیلم #پیشرفته
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍5👎3❤1
کمپانی Cohere یک مدل LLM چند زبانه داده که ۱۰۱ زبان را پشتیبانی میکنه از جمله فارسی. خوبیش اینه که #دیتاست فارسی به نسبت خیلی بزرگی داره (با توجه به مقالاشون). دیتاسشون هم پابلیک هستش.
▪️ Introducing Aya
▪️ Aya Model: An Instruction Finetuned Open-Access Multilingual Language Model
▪️ OpenSource
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Introducing Aya
▪️ Aya Model: An Instruction Finetuned Open-Access Multilingual Language Model
▪️ OpenSource
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
🔥7👍2
Forwarded from DeepMind AI Expert (Farzad 🕊️)
یه ریپازیتوری منابع الگوریتمهای #هوش_مصنوعی در این گیتهاب دارم جمع اوری میکنم اگه میدونین چیزایی جا افتاده برام بفرستید تا اضافه کنم یا اونجا تو چت برام بفرستید تا اضافه کنم هرچی که نیاز دارید اینجا میتونین پیدا کنید.
https://github.com/farzadhs/ML-Courses-on-YouTube
#مقاله #هوش_مصنوعی #منابع #الگوریتمها #پردازش_زبان_طبیعی
پ.ن: در دیده شدن این ریپازیتوری لطفا ستاره بدید
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
https://github.com/farzadhs/ML-Courses-on-YouTube
#مقاله #هوش_مصنوعی #منابع #الگوریتمها #پردازش_زبان_طبیعی
پ.ن: در دیده شدن این ریپازیتوری لطفا ستاره بدید
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍8
یک سوالی که الان خیلی تو ذهن ها شاید بهش فکر شده بوده این میتونه باشه برای ترین LLMs چطوری دیتاست ( ساختگی) تهیه کنیم و برای ترین کردنش چکار کنیم با بررسی که انجام دادم دوتا مقاله زیر میتونه کمک خوبی برای حل این سوال باشه
▪️ Introducing Aya
▪️ RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture
در #مقاله AYa و RAG توضیح دادن چطوری دیتاست رو تهیه کردند
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Introducing Aya
▪️ RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture
در #مقاله AYa و RAG توضیح دادن چطوری دیتاست رو تهیه کردند
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
❤6
محدودیتهای ترنسفورمرها
▪️ On Limitations of the Transformer Architecture
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ On Limitations of the Transformer Architecture
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
❤1👍1
دوتا مقاله متفاوت با بررسی PoE پرداختند که سیستم چطوری Robustness باشه و نباشه ؟!
▪️ Transformers Can Achieve Length Generalization But Not Robustly
▪️ The Impact of Positional Encoding on Length Generalization in Transformers
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Transformers Can Achieve Length Generalization But Not Robustly
▪️ The Impact of Positional Encoding on Length Generalization in Transformers
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍3🔥1
ورژن ۳ کراس منتشر شد .
▪️ New Keras release: 3.0.5.
➡️ Sparse tensor support for the JAX backend.
➡️ Enable saving/loading in bfloat16, float16.
➡️ Bunch of new ops, in particular new linear algebra ops.
➡️ Bug fixes and performance improvements
#پایتون #منابع
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ New Keras release: 3.0.5.
➡️ Sparse tensor support for the JAX backend.
➡️ Enable saving/loading in bfloat16, float16.
➡️ Bunch of new ops, in particular new linear algebra ops.
➡️ Bug fixes and performance improvements
#پایتون #منابع
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
👍5🆒2
This media is not supported in your browser
VIEW IN TELEGRAM
شرکت فیسبوک سابق یه مدل جدید منتشر کرده با دیدن فیلم و درک اونها دنیا بیرون رو میسازه !! بیشتر بخونم اطلاعاتی راجبش مینویسم موضوعی که یکی از پیشنهادات خود دکتر لکون بود در تویت هاش و پیشنهاداتش گفته بود روی موضوع self supervised learning تحقیقات بیشتری انجام بدید
▪️ Revisiting Feature Prediction for Learning Visual Representations from Video
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
▪️ Revisiting Feature Prediction for Learning Visual Representations from Video
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
🆒2👍1
Introducing Ego-Exo4D: A foundational dataset for research on video learning and multimodal perception
https://ai.meta.com/blog/ego-exo4d-video-learning-perception/
https://ego-exo4d-data.org/
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
https://ai.meta.com/blog/ego-exo4d-video-learning-perception/
https://ego-exo4d-data.org/
#ایده_جذاب
🔸 مطالب بیشتر 👇👇
✅ @AI_DeepMind
🔸 @AI_Person
Meta AI
Introducing Ego-Exo4D: A foundational dataset for research on video learning and multimodal perception
With this release, we aim to provide the tools the broader research community needs to explore ego-exo video, multimodal activity recognition, and beyond.
👍2
Forwarded from Recommender system (MehriMoon 🌙)
این کورسه فکر کنم جذاب باشه. مخصوصا
برای دوستانی که اول مسیر هستن.
http://edx.org/course/machine-learning-with-python-from-linear-models-to
دوره هنوز شروع نشده. از ۲۷ می شروع میشه به مدت ۱۵ هفته که حدودا تا وسطای سپتامبر ادامه داره.
هر هفته هم ۱۰ ۱۵ ساعت آموزش و تمرین هست.
به صورت رایگان می تونین ثبت نام کنین.
❌اما اگه مدرکش رو می خوایین یا می خوایین که بعداً هم به محتوا دسترسی داشته باشین فکر کنم ۴۰۰ دلار کانادا قیمتشه.❌
با vpn حتما رجيستر كنيد . ⚠️
برای دوستانی که اول مسیر هستن.
http://edx.org/course/machine-learning-with-python-from-linear-models-to
دوره هنوز شروع نشده. از ۲۷ می شروع میشه به مدت ۱۵ هفته که حدودا تا وسطای سپتامبر ادامه داره.
هر هفته هم ۱۰ ۱۵ ساعت آموزش و تمرین هست.
به صورت رایگان می تونین ثبت نام کنین.
❌اما اگه مدرکش رو می خوایین یا می خوایین که بعداً هم به محتوا دسترسی داشته باشین فکر کنم ۴۰۰ دلار کانادا قیمتشه.❌
با vpn حتما رجيستر كنيد . ⚠️
edX
MITx: Machine Learning with Python: from Linear Models to Deep Learning. | edX
An in-depth introduction to the field of machine learning, from linear models to deep learning and reinforcement learning, through hands-on Python projects. -- Part of the MITx MicroMasters program in Statistics and Data Science.
👍5