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.

9. Google unveils PaLM 2. It has powerful multilingual, reasoning, and coding capabilities trained on 100+ languages. PaLM 2 comes in 4 sizes: - Gecko (10 billion parameters) - Otter (100 billion parameters) - Bison (1 trillion parameters) - Unicorn (10 trillion parameters)

Barsee 🐶

269,826 subscribers

123,288 views • 3 years ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

18 Comments

Barsee 🐶3 years ago

Google wins the internet with its biggest breakthroughs in AI. It takes their generative AI capabilities to a whole new level. Here are the 10 revolutionary innovations they unveiled today:

Barsee 🐶3 years ago

1. Bard Google Bard's new features: -Dark theme -Visual features to search from photos -Support 20+ programming languages -Works with other apps -Write in Gmail, Export into Docs -Generate spreadsheet

Barsee 🐶3 years ago

2. Google unveils a list of future plugins for Bard.

Barsee 🐶3 years ago

3. Google releases "Help me write" in Gmail. Email has changed forever. Just write one-line prompt, AI will automatically generates the whole email for you in seconds. There are also many options to refine your emails.

Barsee 🐶3 years ago

4. Bard + Adobe Firefly announced Bard will be able to generate completely new images using Adobe Firefly.

Barsee 🐶3 years ago

5. Google Unveils Gemini This is Google's next-generation multimodal foundation model. It will directly rival with @OpenAI GPT. Bard will transition to the Google Deepmind Gemini model.

Barsee 🐶3 years ago

6. Google X Adobe Google and Adobe have teamed up to power up Adobe Aero, a tool for AR creators! This partnership brings Google's real-world modeling into Adobe's creative design tools.

Barsee 🐶3 years ago

7. Google Photos' new "Magic Editor" This new feature lets you change: -The lighting of a scene -Get rid of elements you don't like -Adjust your position in the photo.

Barsee 🐶3 years ago

8. Google just announced Immersive View for Maps It helps you: -Know the actual weather and traffic on your route -It calculates the fastest path in less than 2 sec.

Barsee 🐶3 years ago

10. This is how many times Google's CEO says AI

Barsee 🐶3 years ago

If you found this helpful or interesting. Please retweet the tweet below to share it with others:

Barsee 🐶3 years ago

I will cover more about its use cases in my free newsletter today.

Dickson Pau3 years ago

These parameters count cannot be correct. Please do better.

Marco Pinnisi3 years ago

@heyBarsee Source for these model sizes?

APOideas3 years ago

How can I use PaLM2?

Rafa is a northern father ✊🏻3 years ago

@DotCSV

Adam Lieziert3 years ago

Nice- I’ve waited a long time for this upgrade!

Chris Vanderloo3 years ago

A 10bil parameter model isn’t light enough to run on a mobile device. Gecko is probably 2 orders of magnitude smaller

Related Videos

A new milestone in Bharat’s foundational AI model journey. ✅ 8 more teams selected to build indigenous foundational models ✅ AI models being developed range from SLMs with < 10 billion parameters to LLMs with 1 trillion parameters - cutting across sectors

A new milestone in Bharat’s foundational AI model journey. ✅ 8 more teams selected to build indigenous foundational models ✅ AI models being developed range from SLMs with < 10 billion parameters to LLMs with 1 trillion parameters - cutting across sectors

Ashwini Vaishnaw

46,066 views • 1 year ago

DeepSeek V4 just dropped and it's 40x cheaper than GPT-5.5 Pro 1.6 trillion parameters, trained on limited GPUs, still matches the top labs I built 4 full apps with it to see if the hype is real

DeepSeek V4 just dropped and it's 40x cheaper than GPT-5.5 Pro 1.6 trillion parameters, trained on limited GPUs, still matches the top labs I built 4 full apps with it to see if the hype is real

David Ondrej

130,550 views • 2 months ago

𝗞𝗶𝗺𝗶 𝗞𝟮.𝟲 𝗶𝘀 𝗖𝗵𝗶𝗻𝗮'𝘀 𝗻𝗲𝘄 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲 𝗮𝗻𝘀𝘄𝗲𝗿 𝘁𝗼 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲. It has 1 trillion parameters but only uses 32 billion at a time. Context window holds 256,000 tokens. It scored 85% on Live Code Bench. Claude hit 64%. It plugs into OpenClaw, Cursor, and Cline with no restrictions. Kimi Claw deploys in the cloud in 2 minutes. It's way cheaper to run than Claude day to day. Save this. Your coding stack just got a real rival.

𝗞𝗶𝗺𝗶 𝗞𝟮.𝟲 𝗶𝘀 𝗖𝗵𝗶𝗻𝗮'𝘀 𝗻𝗲𝘄 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲 𝗮𝗻𝘀𝘄𝗲𝗿 𝘁𝗼 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲. It has 1 trillion parameters but only uses 32 billion at a time. Context window holds 256,000 tokens. It scored 85% on Live Code Bench. Claude hit 64%. It plugs into OpenClaw, Cursor, and Cline with no restrictions. Kimi Claw deploys in the cloud in 2 minutes. It's way cheaper to run than Claude day to day. Save this. Your coding stack just got a real rival.

Julian Goldie SEO

102,245 views • 5 months ago

OpenAI has finally released GPT OSS 20B!! This is probably the most powerful model you can run locally on a laptop. → Only 3.6B active parameters → Adjustable "Reasoning Effort" → Works with 16GB of RAM! Quick steps to use it 100% free and offline:

OpenAI has finally released GPT OSS 20B!! This is probably the most powerful model you can run locally on a laptop. → Only 3.6B active parameters → Adjustable "Reasoning Effort" → Works with 16GB of RAM! Quick steps to use it 100% free and offline:

Paul Couvert

595,738 views • 1 year ago

NEW: Alibaba just released Qwen 3.5 Small — a family of powerful multimodal models available in a range of sizes (0.8B, 2B, 4B, and 9B parameters). Perfect for on-device applications! They can even run 100% locally in your browser on WebGPU, powered by Transformers.js! 🤯

NEW: Alibaba just released Qwen 3.5 Small — a family of powerful multimodal models available in a range of sizes (0.8B, 2B, 4B, and 9B parameters). Perfect for on-device applications! They can even run 100% locally in your browser on WebGPU, powered by Transformers.js! 🤯

Xenova

102,592 views • 6 months ago

WHY IS NO ONE TALKING ABOUT THIS?? Gemma 3n model was one of the best surprises for me. The fact that you can run it on edge devices even with just 2GB of RAM is impressive. A few weeks back, I was on holiday and used the Gemini Live feature a lot. But I kept running into issues whenever I was in a place where the network wasn’t reliable. Gemma 3n : > Multimodal: Supports text and image inputs; video/audio in development. > Context Window: Up to 128K tokens (32K for 1B model). > Multilingual: Trained on 140+ languages. > Privacy: Offline, on-device processing for data security. > Model Sizes: E2B (5B parameters, 2GB RAM) and E4B (10–12B parameters, 3GB RAM). Video credit: Google YT

WHY IS NO ONE TALKING ABOUT THIS?? Gemma 3n model was one of the best surprises for me. The fact that you can run it on edge devices even with just 2GB of RAM is impressive. A few weeks back, I was on holiday and used the Gemini Live feature a lot. But I kept running into issues whenever I was in a place where the network wasn’t reliable. Gemma 3n : > Multimodal: Supports text and image inputs; video/audio in development. > Context Window: Up to 128K tokens (32K for 1B model). > Multilingual: Trained on 140+ languages. > Privacy: Offline, on-device processing for data security. > Model Sizes: E2B (5B parameters, 2GB RAM) and E4B (10–12B parameters, 3GB RAM). Video credit: Google YT

AshutoshShrivastava

210,641 views • 1 year ago

Ron Baron on SpaceX’s Starlink in a new video: "It is the internet of the planet. We think in 10 years, that’s going to be a trillion dollar annual revenue business. 1 trillion in revenue will produce, we think roughly $700-800 billion in profits a year. The means, in 10 years, this company from Starlink alone will be worth somewhere around $14 trillion." Ron is the 🐐, since investing in SpaceX in 2017, Baron Capital has turned roughly $2 billion into a ~$25 billion position.

Ron Baron on SpaceX’s Starlink in a new video: "It is the internet of the planet. We think in 10 years, that’s going to be a trillion dollar annual revenue business. 1 trillion in revenue will produce, we think roughly $700-800 billion in profits a year. The means, in 10 years, this company from Starlink alone will be worth somewhere around $14 trillion." Ron is the 🐐, since investing in SpaceX in 2017, Baron Capital has turned roughly $2 billion into a ~$25 billion position.

Nic Cruz Patane

17,465 views • 1 month ago

Andrej Karpathy just made one of the most interesting arguments about AI model design that most people are completely missing. His take is that frontier AI models are not too big because the technology is complex and too big because the training data is garbage. When you or I think of the internet, we picture Wall Street Journal articles, Wikipedia entries, serious writing. That is not what a pretraining dataset looks like. When researchers at frontier labs look at random documents from the actual training corpus, it is stock ticker symbols, broken HTML, spam, gibberish. One estimate puts Llama 3's information compression at just 0.07 bits per token meaning the model has only a hazy recollection of most of what it trained on. So we build trillion parameter models not because we need a trillion parameter brain but because we need a trillion-parameter compression engine to squeeze some intelligence out of a firehose of noise. Most of those parameters are doing memory work, not cognitive work. Karpathy's prediction is separate the two entirely. Build a cognitive core, a model that contains only the algorithms for reasoning and problem-solving, stripped of encyclopedic memorization and pair it with external memory that it can query when it needs facts. He thinks a cognitive core trained on high-quality data could hit genuine intelligence at around one billion parameters. For reference, today's flagship models run between 200 billion and 1.8 trillion parameters with most of that weight dedicated to remembering the internet's slop. The trend is already moving his direction. GPT-4o operates at roughly 200 billion parameters and outperforms the original 1.8 trillion-parameter GPT-4. Inference costs for GPT-3.5-level performance dropped 280-fold between 2022 and 2024 driven almost entirely by smaller, cleaner, better-architected models. The real bottleneck in AI right now is not compute but rather data quality.

Andrej Karpathy just made one of the most interesting arguments about AI model design that most people are completely missing. His take is that frontier AI models are not too big because the technology is complex and too big because the training data is garbage. When you or I think of the internet, we picture Wall Street Journal articles, Wikipedia entries, serious writing. That is not what a pretraining dataset looks like. When researchers at frontier labs look at random documents from the actual training corpus, it is stock ticker symbols, broken HTML, spam, gibberish. One estimate puts Llama 3's information compression at just 0.07 bits per token meaning the model has only a hazy recollection of most of what it trained on. So we build trillion parameter models not because we need a trillion parameter brain but because we need a trillion-parameter compression engine to squeeze some intelligence out of a firehose of noise. Most of those parameters are doing memory work, not cognitive work. Karpathy's prediction is separate the two entirely. Build a cognitive core, a model that contains only the algorithms for reasoning and problem-solving, stripped of encyclopedic memorization and pair it with external memory that it can query when it needs facts. He thinks a cognitive core trained on high-quality data could hit genuine intelligence at around one billion parameters. For reference, today's flagship models run between 200 billion and 1.8 trillion parameters with most of that weight dedicated to remembering the internet's slop. The trend is already moving his direction. GPT-4o operates at roughly 200 billion parameters and outperforms the original 1.8 trillion-parameter GPT-4. Inference costs for GPT-3.5-level performance dropped 280-fold between 2022 and 2024 driven almost entirely by smaller, cleaner, better-architected models. The real bottleneck in AI right now is not compute but rather data quality.

Milk Road AI

200,719 views • 5 months ago

AI is entering its physics era guys.. A duo refused Bezos' multi-billion-dollar offer, then unveiled an AI model that no PC can run. "They unveiled Accelerated Understanding (AU), a new physics-first AI startup that takes a different approach from traditional language models. Its models are designed to predict how physical systems evolve across space and time." "In testing, the model processed 5 trillion pieces of physical data in a single prompt — roughly 5 million times more data than Google Gemini or Anthropic Claude can handle at once. This isn't 5 trillion words or tokens. It's massive 3D + time (4D) physical data." 👀 "The model uses neural operators instead of conventional Transformer architectures, and it's designed to simulate phenomena such as weather, fluid dynamics, materials, geology and semiconductor behavior." "AU says it has trained models with up to 1 trillion parameters, with scaling experiments reaching 35 trillion parameters. A 5-trillion-point inference can reportedly generate around 22 TB of output." Now, AI is moving from predicting what comes next in a sentence to predicting what happens next in the physical world. Acceleration is everywhere! 👀

AI is entering its physics era guys.. A duo refused Bezos' multi-billion-dollar offer, then unveiled an AI model that no PC can run. "They unveiled Accelerated Understanding (AU), a new physics-first AI startup that takes a different approach from traditional language models. Its models are designed to predict how physical systems evolve across space and time." "In testing, the model processed 5 trillion pieces of physical data in a single prompt — roughly 5 million times more data than Google Gemini or Anthropic Claude can handle at once. This isn't 5 trillion words or tokens. It's massive 3D + time (4D) physical data." 👀 "The model uses neural operators instead of conventional Transformer architectures, and it's designed to simulate phenomena such as weather, fluid dynamics, materials, geology and semiconductor behavior." "AU says it has trained models with up to 1 trillion parameters, with scaling experiments reaching 35 trillion parameters. A 5-trillion-point inference can reportedly generate around 22 TB of output." Now, AI is moving from predicting what comes next in a sentence to predicting what happens next in the physical world. Acceleration is everywhere! 👀

SciTech Era

38,310 views • 28 days ago

Jim Fan says NVIDIA trained humanoid robots to walk and move like humans -- zero-shot transfer from simulation to the real world. 10 years of learning, compressed into 2 hours of simulation. Turns out you don’t need a giant model to master motion. “1.5 million parameters, not billion, to capture the subconscious processing of the human body.” Big embodiment, small brain.

Jim Fan says NVIDIA trained humanoid robots to walk and move like humans -- zero-shot transfer from simulation to the real world. 10 years of learning, compressed into 2 hours of simulation. Turns out you don’t need a giant model to master motion. “1.5 million parameters, not billion, to capture the subconscious processing of the human body.” Big embodiment, small brain.

vitrupo

714,883 views • 1 year ago

NEWS: Billionaire investor Ron Baron has placed a $1 billion order for SpaceX IPO shares and says he is not selling in his lifetime. Baron, the founder and CEO of Baron Capital, laid out his case on a client call. His firm has put about $2 billion into SpaceX across 27 employee tender rounds since 2017, a position now worth roughly $15 billion after compounding at 54% per year. He believes SpaceX, coming public at under $2 trillion, will reach $10 trillion to $30 trillion within 10 to 15 years and become the largest, most profitable company on the planet. People inside the company tell him he is lowballing it. On Starlink alone, Baron projects 300 million subscribers and about $1 trillion in annual revenue within a decade, making it worth around $14 trillion on its own.

NEWS: Billionaire investor Ron Baron has placed a $1 billion order for SpaceX IPO shares and says he is not selling in his lifetime. Baron, the founder and CEO of Baron Capital, laid out his case on a client call. His firm has put about $2 billion into SpaceX across 27 employee tender rounds since 2017, a position now worth roughly $15 billion after compounding at 54% per year. He believes SpaceX, coming public at under $2 trillion, will reach $10 trillion to $30 trillion within 10 to 15 years and become the largest, most profitable company on the planet. People inside the company tell him he is lowballing it. On Starlink alone, Baron projects 300 million subscribers and about $1 trillion in annual revenue within a decade, making it worth around $14 trillion on its own.

Muskonomy

1,292,759 views • 3 months ago

Rockstar spent 7 years, 2,000+ developers, and $500 million to build Red Dead Redemption 2. Kimi K3 rebuilt the same scene in 10 hours. One prompt. 4 million tokens. It's open source. 2.8 trillion parameters. A 1 million token context window. It beats Claude Opus 4.8 and GPT-5.6. It sits on par with Fable 5. Fable 5 results, at Sonnet pricing. The wall between you and a AAA game used to be a studio. Now it's a prompt.

Rockstar spent 7 years, 2,000+ developers, and $500 million to build Red Dead Redemption 2. Kimi K3 rebuilt the same scene in 10 hours. One prompt. 4 million tokens. It's open source. 2.8 trillion parameters. A 1 million token context window. It beats Claude Opus 4.8 and GPT-5.6. It sits on par with Fable 5. Fable 5 results, at Sonnet pricing. The wall between you and a AAA game used to be a studio. Now it's a prompt.

Dep

99,354 views • 2 months ago

2.8 trillion parameters. Number one on frontend code arena. The largest open-weight model ever. Sputnik moment? It’s totally mind boggling that Elon Musk can spend $16 billion on Colossus 2 for a 20 trillion parameter model, and there's a 1% cost distilled version that’s coming out of China within weeks. Alex Karp said last week that you can’t throw all your proprietary weights over the wall to Anthropic. But he didn’t give a roadmap. Here we are, a week later, and it’s suddenly a free-for-all. Intelligence wants to be free.

2.8 trillion parameters. Number one on frontend code arena. The largest open-weight model ever. Sputnik moment? It’s totally mind boggling that Elon Musk can spend $16 billion on Colossus 2 for a 20 trillion parameter model, and there's a 1% cost distilled version that’s coming out of China within weeks. Alex Karp said last week that you can’t throw all your proprietary weights over the wall to Anthropic. But he didn’t give a roadmap. Here we are, a week later, and it’s suddenly a free-for-all. Intelligence wants to be free.

Dave Blundin

11,513 views • 2 months ago

Adebayo Ogunlesi founded GIP and manages about $100 Billion in Assets with Larry Fink, the co-founder CEO of BlackRock, a company that manages $10 Trillion in Assets. It is believed that Larry is the most powerful person in world finance!

Adebayo Ogunlesi founded GIP and manages about $100 Billion in Assets with Larry Fink, the co-founder CEO of BlackRock, a company that manages $10 Trillion in Assets. It is believed that Larry is the most powerful person in world finance!

Toyyib Adewale Adelodun

1,665,394 views • 2 years ago

AMERICA IS BROKE * China holds $790 Billion of US Bonds * China has just started providing info on US military assets in the gulf * The US needs to sell $3 trillion in new bonds this year * BRICS is challenging US world dominance with Petrodollars * The #USIranWar is costing us $1 billion a day * In just 10 days it has cost us $97 billion

AMERICA IS BROKE * China holds $790 Billion of US Bonds * China has just started providing info on US military assets in the gulf * The US needs to sell $3 trillion in new bonds this year * BRICS is challenging US world dominance with Petrodollars * The #USIranWar is costing us $1 billion a day * In just 10 days it has cost us $97 billion

Tomthunkit™

75,630 views • 6 months ago

New LLMs that control UIs! ByteDance Research releases UI-TARS, fine-tuned GUI agent that integrates reasoning, and action capabilities into a single vision-language model. Think of computer use but open. 👀 TL;DR; 3️⃣ Available in 3 sizes: 2B, 7B, and 72B parameters 🧠 Trained Qwen2-VL models with SFT & DPO 🥇 72B version achieves 82.8% on VisualWebBench (beating GPT-4 and Claude) 🏆 Achieves state-of-the-art results on 10+ GUI agent benchmarks 💡 Reasons before taking an action 🧑🏻‍💻 Can Click, Long Press, type, scroll, open app, navigate back/home, wait 🤗 Released under Apache 2.0 on Hugging Face

New LLMs that control UIs! ByteDance Research releases UI-TARS, fine-tuned GUI agent that integrates reasoning, and action capabilities into a single vision-language model. Think of computer use but open. 👀 TL;DR; 3️⃣ Available in 3 sizes: 2B, 7B, and 72B parameters 🧠 Trained Qwen2-VL models with SFT & DPO 🥇 72B version achieves 82.8% on VisualWebBench (beating GPT-4 and Claude) 🏆 Achieves state-of-the-art results on 10+ GUI agent benchmarks 💡 Reasons before taking an action 🧑🏻‍💻 Can Click, Long Press, type, scroll, open app, navigate back/home, wait 🤗 Released under Apache 2.0 on Hugging Face

Philipp Schmid

48,170 views • 1 year ago

Current robotics models really underscore how wildly efficient the human brain is. Robotics models are relatively small (low billions of parameters), are only storing a second of context, and doing 100ms inference steps. By contrast the human brain has 100 trillion synapses (roughly analogous to parameters), processes 24 frames per second, while considering hours of previous context. So how do we expect to get many order-of-magnitude improvements in robotics across all these dimensions (which are trading off against each other in terms inference compute) at the same time? And how do we explain what's going on in the brain? Is it more analogous to a software advantage or a hardware advantage? Full episode with Sergey Levine out now!

Current robotics models really underscore how wildly efficient the human brain is. Robotics models are relatively small (low billions of parameters), are only storing a second of context, and doing 100ms inference steps. By contrast the human brain has 100 trillion synapses (roughly analogous to parameters), processes 24 frames per second, while considering hours of previous context. So how do we expect to get many order-of-magnitude improvements in robotics across all these dimensions (which are trading off against each other in terms inference compute) at the same time? And how do we explain what's going on in the brain? Is it more analogous to a software advantage or a hardware advantage? Full episode with Sergey Levine out now!

Dwarkesh Patel

87,247 views • 1 year ago

Ron Baron put $1.7 billion into SpaceX while it was private - it turned into $15 billion he just placed a billion dollar order at the IPO and said "10 trillion, 20 trillion, 30 trillion and I could be very low" he started with $100 million in 1992, made his clients $61 billion in profit when he was a kid he drove an ice cream truck, now the president of the NYSE comes to him for advice "I pushed all the chips to the center of the table to be tied to the most successful man on the planet, clearly risky, but that's my game" "never bet against the guy with superpowers who would never give up" bookmark and watch it today ↓

Ron Baron put $1.7 billion into SpaceX while it was private - it turned into $15 billion he just placed a billion dollar order at the IPO and said "10 trillion, 20 trillion, 30 trillion and I could be very low" he started with $100 million in 1992, made his clients $61 billion in profit when he was a kid he drove an ice cream truck, now the president of the NYSE comes to him for advice "I pushed all the chips to the center of the table to be tied to the most successful man on the planet, clearly risky, but that's my game" "never bet against the guy with superpowers who would never give up" bookmark and watch it today ↓

Сarm1ne

2,987,878 views • 3 months ago