Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

✨ What makes Gemini 2.5 Pro stand out? In this Release Notes episode, Sr. Product Manager Logan Kilpatrick and Gemini Product Lead Tulsee Doshi break down its reasoning, coding, and multimodal strengths, plus a 1M token long context. ↓ Timecodes: 1:05 Gemini 2.5 launch overview 3:19 Academic evals vs.... show more

Google AI Developers

116,324 subscribers

228,165 Aufrufe • vor 1 Jahr •via X (Twitter)

Bildung Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

8 Kommentare

Profilbild von ⟁ndrew V

⟁ndrew Vvor 1 Jahr

@OfficialLoganK @tulseedoshi Great conversation, and loved @tulseedoshi's convo on ThursdAI yesterday too. Thanks for putting sharing!

Profilbild von opensourceCM

opensourceCMvor 1 Jahr

What’s the cost of mistakes in your contracts? If you work with contracts day-to-day, it’s time to automate. Track every detail, streamline workflows ... ✨ Make managing contracts as easy as a few clicks. Visit our new website & book your demo today!

Profilbild von Joe Ward

Joe Wardvor 1 Jahr

@OfficialLoganK @tulseedoshi That was good!

Profilbild von Conscious AI

Conscious AIvor 1 Jahr

@OfficialLoganK @tulseedoshi I have no clue why Logan blocked me.

Profilbild von ⟁ndrew V

⟁ndrew Vvor 1 Jahr

@OfficialLoganK @tulseedoshi 🛳️ ship emoji for the teams!

Profilbild von Sohail Hosseini

Sohail Hosseinivor 1 Jahr

@OfficialLoganK @tulseedoshi Keep up the good work!

Profilbild von Nemesis

Nemesisvor 1 Jahr

@OfficialLoganK @tulseedoshi Gemini 2.5 Pro sounds amazing! Excited to see its advancements in reasoning, coding, and multimodal strengths. The 1M token long context is impressive! 😊

Profilbild von Vincent Wu

Vincent Wuvor 1 Jahr

@OfficialLoganK @tulseedoshi Is there an official release date for 2.5 Pro yet?

Ähnliche Videos

How does an AI model actually learn to see? 🤖 Learn about the tech behind native multimodality, how models reason over visual data like documents and video, and the future of proactive AI assistants with Logan Kilpatrick and Gemini Model Behavior Product Lead, Ani Baddepudi. ↓ Timestamps: 01:12 Why Gemini is natively multimodal 02:23 The technology behind multimodal models 05:15 Video understanding with Gemini 2.5 09:25 Deciding what to build next 13:23 Building new product experiences with multimodal AI 17:15 The vision for proactive assistants 24:13 Improving video usability with variable FPS and frame tokenization 27:35 What’s next for Gemini’s multimodal development 31:47 Deep dive on Gemini’s document understanding capabilities 37:56 The teamwork and collaboration behind Gemini 40:56 What’s next with model behavior

How does an AI model actually learn to see? 🤖 Learn about the tech behind native multimodality, how models reason over visual data like documents and video, and the future of proactive AI assistants with Logan Kilpatrick and Gemini Model Behavior Product Lead, Ani Baddepudi. ↓ Timestamps: 01:12 Why Gemini is natively multimodal 02:23 The technology behind multimodal models 05:15 Video understanding with Gemini 2.5 09:25 Deciding what to build next 13:23 Building new product experiences with multimodal AI 17:15 The vision for proactive assistants 24:13 Improving video usability with variable FPS and frame tokenization 27:35 What’s next for Gemini’s multimodal development 31:47 Deep dive on Gemini’s document understanding capabilities 37:56 The teamwork and collaboration behind Gemini 40:56 What’s next with model behavior

Google AI

58,703 Aufrufe • vor 1 Jahr

Learn about Google’s new SOTA image model, Gemini 2.5 Flash, its key capabilities, and what’s next on the roadmap with some of the team behind the model Nicole Brichtova Kaushik Shivakumar Mostafa Dehghani Robert Riachi with Logan Kilpatrick. Timecodes: 0:37 New model introduction 01:21 Demo: Image editing 03:44 Text rendering capabilities 04:44 Beyond human preference evals 06:44 Text rendering as a proxy for quality 08:38 Positive transfer between modalities 11:25 Demo: multi-turn, context aware image generation 13:54 Pixel-perfect editing and character consistency 15:51 Interleaved image generation 17:59 Specialized vs. native models 19:52 Understanding nuanced prompts 20:59 User feedback shaping model development 22:37 Improvements in character consistency 24:17 More natural looking images from team collaboration 26:41 What’s next for image generation models

Learn about Google’s new SOTA image model, Gemini 2.5 Flash, its key capabilities, and what’s next on the roadmap with some of the team behind the model Nicole Brichtova Kaushik Shivakumar Mostafa Dehghani Robert Riachi with Logan Kilpatrick. Timecodes: 0:37 New model introduction 01:21 Demo: Image editing 03:44 Text rendering capabilities 04:44 Beyond human preference evals 06:44 Text rendering as a proxy for quality 08:38 Positive transfer between modalities 11:25 Demo: multi-turn, context aware image generation 13:54 Pixel-perfect editing and character consistency 15:51 Interleaved image generation 17:59 Specialized vs. native models 19:52 Understanding nuanced prompts 20:59 User feedback shaping model development 22:37 Improvements in character consistency 24:17 More natural looking images from team collaboration 26:41 What’s next for image generation models

Google AI Developers

31,150 Aufrufe • vor 11 Monaten

Get the inside story on the development of Gemini's coding capabilities. Listen as the product and research leads for Gemini share their philosophy on what makes a great coding model, the impact of "vibe coding," and the future of programming languages with Logan Kilpatrick, Connie Fan and Danny Tarlow. Timecodes: 0:00 Intro 1:10 Defining Early Coding Goals 6:23 Ingredients of a Great Coding Model 9:28 Adapting to Developer Workflows 11:40 The Rise of Vibe Coding 14:43 Code as a Reasoning Tool 17:20 Code as a Universal Solver 20:47 Evaluating Coding Models 24:30 Leveraging Internal Googler Feedback 26:52 Winning Over AI Skeptics 28:04 Performance Across Programming Languages 33:05 The Future of Programming Languages 36:16 Strategies for Large Codebases 41:06 Hill Climbing New Benchmarks 42:46 Short-Term Improvements 44:42 Model Style and Taste 47:43 2.5 Pro’s Breakthrough 51:06 Early AI Coding Experiences 56:19 Specialist vs. Generalist Models

Get the inside story on the development of Gemini's coding capabilities. Listen as the product and research leads for Gemini share their philosophy on what makes a great coding model, the impact of "vibe coding," and the future of programming languages with Logan Kilpatrick, Connie Fan and Danny Tarlow. Timecodes: 0:00 Intro 1:10 Defining Early Coding Goals 6:23 Ingredients of a Great Coding Model 9:28 Adapting to Developer Workflows 11:40 The Rise of Vibe Coding 14:43 Code as a Reasoning Tool 17:20 Code as a Universal Solver 20:47 Evaluating Coding Models 24:30 Leveraging Internal Googler Feedback 26:52 Winning Over AI Skeptics 28:04 Performance Across Programming Languages 33:05 The Future of Programming Languages 36:16 Strategies for Large Codebases 41:06 Hill Climbing New Benchmarks 42:46 Short-Term Improvements 44:42 Model Style and Taste 47:43 2.5 Pro’s Breakthrough 51:06 Early AI Coding Experiences 56:19 Specialist vs. Generalist Models

Google AI Developers

65,474 Aufrufe • vor 1 Jahr

Gemini 3, scaling laws and the 'finite data' era: my conversation with Sebastian Borgeaud, research engineer at Google DeepMind and a pre-training lead for Gemini 3 00:00 – Cold intro: “We’re ahead of schedule” + AI is now a system 00:58 – Oriol Vinyals's “secret recipe”: better pre- + post-training 02:09 – Why AI progress still isn’t slowing down 03:04 – Are models actually getting smarter? 04:36 – Two–three years out: what changes first? 06:34 – AI doing AI research: faster, not automated 07:45 – Frontier labs: same playbook or different bets? 10:19 – Post-transformers: will a disruption happen? 10:51 – DeepMind’s advantage: research × engineering × infra 12:26 – What a Gemini 3 pre-training lead actually does 13:59 – From Europe to Cambridge to DeepMind 18:06 – Why he left RL for real-world data 20:05 – From Gopher to Chinchilla to RETRO (and why it matters) 20:28 – “Research taste”: integrate or slow everyone down 23:00 – Fixes vs moonshots: how they balance the pipeline 24:37 – Research vs product pressure (and org structure) 26:24 – Gemini 3 under the hood: MoE in plain English 28:30 – Native multimodality: the hidden costs 30:03 – Scaling laws aren’t dead (but scale isn’t everything) 33:07 – Synthetic data: powerful, dangerous? 35:00 – Reasoning traces: what he can’t say (and why) 37:18 – Long context + attention: what’s next 38:40 – Retrieval vs RAG vs long context 41:49 – The real boss fight: evals (and contamination) 42:28 – Alignment: pre-training vs post-training 43:32 – Deep Think + agents + “vibe coding” 46:34 – Continual learning: updating models over time 49:35 – Advice for researchers + founders 53:35 – “No end in sight” for progress + closing

Gemini 3, scaling laws and the 'finite data' era: my conversation with Sebastian Borgeaud, research engineer at Google DeepMind and a pre-training lead for Gemini 3 00:00 – Cold intro: “We’re ahead of schedule” + AI is now a system 00:58 – Oriol Vinyals's “secret recipe”: better pre- + post-training 02:09 – Why AI progress still isn’t slowing down 03:04 – Are models actually getting smarter? 04:36 – Two–three years out: what changes first? 06:34 – AI doing AI research: faster, not automated 07:45 – Frontier labs: same playbook or different bets? 10:19 – Post-transformers: will a disruption happen? 10:51 – DeepMind’s advantage: research × engineering × infra 12:26 – What a Gemini 3 pre-training lead actually does 13:59 – From Europe to Cambridge to DeepMind 18:06 – Why he left RL for real-world data 20:05 – From Gopher to Chinchilla to RETRO (and why it matters) 20:28 – “Research taste”: integrate or slow everyone down 23:00 – Fixes vs moonshots: how they balance the pipeline 24:37 – Research vs product pressure (and org structure) 26:24 – Gemini 3 under the hood: MoE in plain English 28:30 – Native multimodality: the hidden costs 30:03 – Scaling laws aren’t dead (but scale isn’t everything) 33:07 – Synthetic data: powerful, dangerous? 35:00 – Reasoning traces: what he can’t say (and why) 37:18 – Long context + attention: what’s next 38:40 – Retrieval vs RAG vs long context 41:49 – The real boss fight: evals (and contamination) 42:28 – Alignment: pre-training vs post-training 43:32 – Deep Think + agents + “vibe coding” 46:34 – Continual learning: updating models over time 49:35 – Advice for researchers + founders 53:35 – “No end in sight” for progress + closing

Matt Turck

51,317 Aufrufe • vor 7 Monaten

A LOT happened last week. ICYMI, koray kavukcuoglu (CTO of Google DeepMind and Chief AI Architect of Google) and Logan Kilpatrick discuss Gemini 3, the state of AI, and where we are on the path to AGI. Chapters: 0:00 - Intro 2:00 - Gemini 3 launch reception 4:16 - Continuous progress and innovation 6:47 - Key areas for Gemini improvement 11:45 - Product scaffolding for model improvement 13:56 - Chief AI architect role 17:04 - Engineering mindset and collaboration 18:37 - Future growth areas for Gemini 20:33 - From research to engineering mindset 23:22 - The rise of generative media 27:22 - Nano Banana Pro capabilities 29:31 - Towards unified model checkpoints 36:26 - Organizing for AI success 38:26 - Balancing exploration and scaling 41:40 - DeepMind's collaborative culture 45:21 - Innovating at Google 48:37 - Closing

A LOT happened last week. ICYMI, koray kavukcuoglu (CTO of Google DeepMind and Chief AI Architect of Google) and Logan Kilpatrick discuss Gemini 3, the state of AI, and where we are on the path to AGI. Chapters: 0:00 - Intro 2:00 - Gemini 3 launch reception 4:16 - Continuous progress and innovation 6:47 - Key areas for Gemini improvement 11:45 - Product scaffolding for model improvement 13:56 - Chief AI architect role 17:04 - Engineering mindset and collaboration 18:37 - Future growth areas for Gemini 20:33 - From research to engineering mindset 23:22 - The rise of generative media 27:22 - Nano Banana Pro capabilities 29:31 - Towards unified model checkpoints 36:26 - Organizing for AI success 38:26 - Balancing exploration and scaling 41:40 - DeepMind's collaborative culture 45:21 - Innovating at Google 48:37 - Closing

Google AI

181,225 Aufrufe • vor 8 Monaten

Exclusive: Google just released two upgraded Gemini 1.5 models, 1.5-Pro-002 and 1.5-Flash-002 — achieving new, state-of-the-art performance across math. I sat down with Logan Kilpatrick (Logan Kilpatrick) to discuss the models, AI agents, AGI, and more. Timestamps: 00:00 Intro 01:01 Google rolls out two new Gemini models 2:18 What makes the new models so unique 3:40 Math improvements 8:20 Examples of Gemini 1.5 in the real world 10:19 Future problems AI could solve 12:54 Advantages of Gemini 1.5 16:50 Where beginners can get started 19:01 Advice to thrive in the new age of AI 22:02 Turning notes into podcasts 26:30 AI agents: Definitions and what they can do 31:22 What’s the final form for agents? 33:26 Proactive AI agent systems 36:00 Context windows in the agent era 41:01 AGI’s definition in Logan’s eyes 42:05 The current bottlenecks towards AGI

Exclusive: Google just released two upgraded Gemini 1.5 models, 1.5-Pro-002 and 1.5-Flash-002 — achieving new, state-of-the-art performance across math. I sat down with Logan Kilpatrick (Logan Kilpatrick) to discuss the models, AI agents, AGI, and more. Timestamps: 00:00 Intro 01:01 Google rolls out two new Gemini models 2:18 What makes the new models so unique 3:40 Math improvements 8:20 Examples of Gemini 1.5 in the real world 10:19 Future problems AI could solve 12:54 Advantages of Gemini 1.5 16:50 Where beginners can get started 19:01 Advice to thrive in the new age of AI 22:02 Turning notes into podcasts 26:30 AI agents: Definitions and what they can do 31:22 What’s the final form for agents? 33:26 Proactive AI agent systems 36:00 Context windows in the agent era 41:01 AGI’s definition in Logan’s eyes 42:05 The current bottlenecks towards AGI

Rowan Cheung

269,890 Aufrufe • vor 1 Jahr

New Episode: Carolina Parada leads the robotics team at Google DeepMind. Carolina believes in a future with a broad, rich ecosystem of diverse robot types, where AI is smart enough to embody any robot. We discuss her journey, advancements in Gemini Robotics 1.5, cross-embodiment transfer, RL versus imitation, humanoids, world models, scaling laws, societal concerns, and more. 1:17 Introduction 1:38 Career journey and inspirations 3:50 Reaction to the 2022 ChatGPT launch 4:53 Google DeepMind's robotics mission 9:13 Key upgrades in Gemini Robotics 1.5 14:06 Agentic system's web usage 15:33 Robotics data-gathering methods 16:57 VLA vs. reasoning model training differences 18:22 Convergence of action and reasoning models 19:35 Cross-embodiment transfer 22:34 Learning directly from humans 24:20 Generalization challenges 27:01 Imitation versus reinforcement learning 28:51 Use of world models in robotics 30:31 Applications and testing of Gemini Robotics 1.5 33:04 Do humanoids deserve special focus? 35:06 Home humanoids timeline & limiting factors 37:19 Scaling laws in robotics versus self-driving 38:46 Value of learning classical robotics approaches 40:38 Ethical issues and societal impact

New Episode: Carolina Parada leads the robotics team at Google DeepMind. Carolina believes in a future with a broad, rich ecosystem of diverse robot types, where AI is smart enough to embody any robot. We discuss her journey, advancements in Gemini Robotics 1.5, cross-embodiment transfer, RL versus imitation, humanoids, world models, scaling laws, societal concerns, and more. 1:17 Introduction 1:38 Career journey and inspirations 3:50 Reaction to the 2022 ChatGPT launch 4:53 Google DeepMind's robotics mission 9:13 Key upgrades in Gemini Robotics 1.5 14:06 Agentic system's web usage 15:33 Robotics data-gathering methods 16:57 VLA vs. reasoning model training differences 18:22 Convergence of action and reasoning models 19:35 Cross-embodiment transfer 22:34 Learning directly from humans 24:20 Generalization challenges 27:01 Imitation versus reinforcement learning 28:51 Use of world models in robotics 30:31 Applications and testing of Gemini Robotics 1.5 33:04 Do humanoids deserve special focus? 35:06 Home humanoids timeline & limiting factors 37:19 Scaling laws in robotics versus self-driving 38:46 Value of learning classical robotics approaches 40:38 Ethical issues and societal impact

The Humanoid Hub

71,481 Aufrufe • vor 9 Monaten

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Anshul Dhawan

32,192 Aufrufe • vor 6 Monaten

In this episode, OpenAI Chief Research Officer Mark Chen joins allen to flambé shrimp, cook Korean stew, and chat about being at the frontier of AI research: why scaling laws and pre-training still matter, how OpenAI chooses research bets and allocates compute, what it means to develop research taste, why evals are in crisis, how to avoid benchmark-maxing, and what it will take for models to handle long-horizon real-world work, multimodal reasoning, and eventually end-to-end AI research. Timestamps: 0:00 Intro 0:28 The Soup Story 1:52 From Trading to AI Research 3:21 How to Develop Research Taste 5:23 RL, Evals, and Superhuman Benchmarks 8:17 Cooking Begins on the Impulse Stove 8:53 Scaling Laws, Pre-Training, and Reasoning 12:33 OpenAI’s Research Roadmap and Compute Allocation 15:48 What Makes a Great Researcher 19:33 The Evals Crisis and Benchmark-Maxing 24:34 Jagged Intelligence, Context, and Long-Horizon Learning 27:14 Shrimp Flambé and New Research Bets 31:32 Multimodal Models and One Architecture 32:36 Vibe Researching and End-to-End AI Research 34:36 Failed Bets, Postmortems, and OpenAI’s Alpha 37:07 Final Taste Test 37:53 Overrated vs. Underrated AI Research 41:00 Closing

In this episode, OpenAI Chief Research Officer Mark Chen joins allen to flambé shrimp, cook Korean stew, and chat about being at the frontier of AI research: why scaling laws and pre-training still matter, how OpenAI chooses research bets and allocates compute, what it means to develop research taste, why evals are in crisis, how to avoid benchmark-maxing, and what it will take for models to handle long-horizon real-world work, multimodal reasoning, and eventually end-to-end AI research. Timestamps: 0:00 Intro 0:28 The Soup Story 1:52 From Trading to AI Research 3:21 How to Develop Research Taste 5:23 RL, Evals, and Superhuman Benchmarks 8:17 Cooking Begins on the Impulse Stove 8:53 Scaling Laws, Pre-Training, and Reasoning 12:33 OpenAI’s Research Roadmap and Compute Allocation 15:48 What Makes a Great Researcher 19:33 The Evals Crisis and Benchmark-Maxing 24:34 Jagged Intelligence, Context, and Long-Horizon Learning 27:14 Shrimp Flambé and New Research Bets 31:32 Multimodal Models and One Architecture 32:36 Vibe Researching and End-to-End AI Research 34:36 Failed Bets, Postmortems, and OpenAI’s Alpha 37:07 Final Taste Test 37:53 Overrated vs. Underrated AI Research 41:00 Closing

Latent.Space

62,796 Aufrufe • vor 1 Monat

Gemini-1.5 Pro has its spotlight stolen today, and people are poking fun at Sora vs Google memes. Well, I think it's the biggest boost in LLM capability so far in 2024. v1.5's 10M token context (1) excels at retrieval; (2) generalizes zero-shot to extremely long instructions like full tutorials and codebases; and (3) works across modalities such as text, audio, and video. Here's a stunning example: v1.5 learns to translate from English to Kalamang purely in context, following a full linguistic manual at inference time. Kalamang is a language spoken by fewer than 200 speakers in western New Guinea. Gemini has never seen this language during training and is only provided with 500 pages of linguistic documentation, a dictionary, and ~400 parallel sentences in context. It basically acquires a sophisticated new skill in the neural activations, instead of gradient finetuning. I talked about the Myth of Context Length many times before: don't get too excited by claims of 1M or even 1B context tokens. LSTMs already achieved literally infinite context length 25 yrs ago! What truly matters is how well the model actually uses the context to solve real-world problems, and Gemini-1.5 has surpassed the SOTA with flying colors. The paper is also well-written with lots of solid quantitative analysis on in-context memorization and generalization. Paper: “Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context” Congrats to Jeff Dean Oriol Vinyals Sundar Pichai and team!

Gemini-1.5 Pro has its spotlight stolen today, and people are poking fun at Sora vs Google memes. Well, I think it's the biggest boost in LLM capability so far in 2024. v1.5's 10M token context (1) excels at retrieval; (2) generalizes zero-shot to extremely long instructions like full tutorials and codebases; and (3) works across modalities such as text, audio, and video. Here's a stunning example: v1.5 learns to translate from English to Kalamang purely in context, following a full linguistic manual at inference time. Kalamang is a language spoken by fewer than 200 speakers in western New Guinea. Gemini has never seen this language during training and is only provided with 500 pages of linguistic documentation, a dictionary, and ~400 parallel sentences in context. It basically acquires a sophisticated new skill in the neural activations, instead of gradient finetuning. I talked about the Myth of Context Length many times before: don't get too excited by claims of 1M or even 1B context tokens. LSTMs already achieved literally infinite context length 25 yrs ago! What truly matters is how well the model actually uses the context to solve real-world problems, and Gemini-1.5 has surpassed the SOTA with flying colors. The paper is also well-written with lots of solid quantitative analysis on in-context memorization and generalization. Paper: “Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context” Congrats to Jeff Dean Oriol Vinyals Sundar Pichai and team!

Jim Fan

278,517 Aufrufe • vor 2 Jahren

Today, we're joined by Aakanksha Chowdhery, member of technical staff at Reflection, to explore the fundamental shifts required to build true agentic AI. While the industry has largely focused on post-training techniques to improve reasoning, Aakanksha draws on her experience leading pre-training efforts for Google’s PaLM and early Gemini models to argue that pre-training itself must be rethought to move beyond static benchmarks. We explore the limitations of next-token prediction for multi-step workflows and examine how attention mechanisms, loss objectives, and training data must evolve to support long-form reasoning and planning. Aakanksha shares insights on the difference between context retrieval and actual reasoning, the importance of "trajectory" training data, and why scaling remains essential for discovering emergent agentic capabilities like error recovery and dynamic tool learning. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 02:26 - Reflection 04:54 - Limitations of post-training for building agents 07:31 - Rethinking pre-training in agents 10:51 - Scaling 11:27 - Evolving attention mechanisms for agentic capabilities 12:39 - Memory as a tool 14:13 - Loss objectives and training data 15:50 - Fine-tuning loss in agent performance 19:37 - Training data 21:29 - Augmenting dominant training data source 24:11 - Overcoming challenges in training on synthetic data 25:47 - Benchmarks 30:44 - Scaling laws in large models versus small models 33:20 - Long-form versus short-form reasoning 37:57 - Agent’s ability to recover from failure 40:15 - Hallucinations and failure recovery 43:53 - Tool use in agents 46:38 - Coding agents 48:37 - How researchers can contribute to agentic AI

Today, we're joined by Aakanksha Chowdhery, member of technical staff at Reflection, to explore the fundamental shifts required to build true agentic AI. While the industry has largely focused on post-training techniques to improve reasoning, Aakanksha draws on her experience leading pre-training efforts for Google’s PaLM and early Gemini models to argue that pre-training itself must be rethought to move beyond static benchmarks. We explore the limitations of next-token prediction for multi-step workflows and examine how attention mechanisms, loss objectives, and training data must evolve to support long-form reasoning and planning. Aakanksha shares insights on the difference between context retrieval and actual reasoning, the importance of "trajectory" training data, and why scaling remains essential for discovering emergent agentic capabilities like error recovery and dynamic tool learning. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 02:26 - Reflection 04:54 - Limitations of post-training for building agents 07:31 - Rethinking pre-training in agents 10:51 - Scaling 11:27 - Evolving attention mechanisms for agentic capabilities 12:39 - Memory as a tool 14:13 - Loss objectives and training data 15:50 - Fine-tuning loss in agent performance 19:37 - Training data 21:29 - Augmenting dominant training data source 24:11 - Overcoming challenges in training on synthetic data 25:47 - Benchmarks 30:44 - Scaling laws in large models versus small models 33:20 - Long-form versus short-form reasoning 37:57 - Agent’s ability to recover from failure 40:15 - Hallucinations and failure recovery 43:53 - Tool use in agents 46:38 - Coding agents 48:37 - How researchers can contribute to agentic AI

The TWIML AI Podcast

44,888 Aufrufe • vor 7 Monaten

Explore the Live API’s new audio capabilities in this episode of Release Notes with Shrestha Basu Mallick and Logan Kilpatrick. From native audio output to proactive dialogue, learn how you can build more natural and engaging multimodal AI applications. Timecodes: 0:00 Intro 01:18 Live API Overview 03:36 Why audio is a special modality 05:07 Speed vs. precision in audio 06:17 Controllable and promptable TTS 08:31 What developers are building with the Live API 11:14 URL context and async calling features 15:02 Proactive audio and affective dialog 16:55 Addressing developer feedback 21:54 Live API roadmap 23:49 The role of long context 24:57 What’s next for the Live API 26:41 State of the AI audio market 30:10 Advice for developers getting started with the Live API 31:16 Live API demo 38:10 Demo wrap up and closing

Explore the Live API’s new audio capabilities in this episode of Release Notes with Shrestha Basu Mallick and Logan Kilpatrick. From native audio output to proactive dialogue, learn how you can build more natural and engaging multimodal AI applications. Timecodes: 0:00 Intro 01:18 Live API Overview 03:36 Why audio is a special modality 05:07 Speed vs. precision in audio 06:17 Controllable and promptable TTS 08:31 What developers are building with the Live API 11:14 URL context and async calling features 15:02 Proactive audio and affective dialog 16:55 Addressing developer feedback 21:54 Live API roadmap 23:49 The role of long context 24:57 What’s next for the Live API 26:41 State of the AI audio market 30:10 Advice for developers getting started with the Live API 31:16 Live API demo 38:10 Demo wrap up and closing

Google AI Developers

24,239 Aufrufe • vor 1 Jahr

This is the interview to make you feel super bullish. Had a great time hosting Bill Miller on the show for the first time! We touched on a large range of topics, let us know what you think! Timecodes: 00:00 Why Bill is Bullish on Bitcoin 4:22 Gemini Orange Bitcoin Credit Card 5:43 Why Bitcoin Hasn't Gone Fully Mainstream…Yet 9:10 Rise of Corporate Bitcoin Treasuries 11:36 Bitcoin-Backed Securities 13:10 Bullish on Semler Scientific 14:18 Market Misunderstands Semler Scientific? 15:13 Bitcoin Rewards Cards and Points Inflation 17:54 Are Bitcoin Treasuries a Bubble 18:42 Stablecoins 20:42 Bitcoin and the US Treasury Market 21:34 Economic Outlook, Market Trends 23:05 Is a Recession Looming? 23:57 Bitcoin Price Predictions 25:13 The Future of Bitcoin Taxation 27:34 Inside Miller Value Partners’ Strategy 28:49 Can Anything Beat Bitcoin? 29:22 Becoming a Bitcoin Believer 30:03 Bitcoin Changes Your Worldview 31:22 Growing Up Around Legendary Investors 33:36 Bonding Over Bitcoin 35:19 The Pressure of a Famous Father 36:09 Trump vs. Elon Powered by Gemini with a shoutout about the Gemini credit card in the talk! 💳🧡

This is the interview to make you feel super bullish. Had a great time hosting Bill Miller on the show for the first time! We touched on a large range of topics, let us know what you think! Timecodes: 00:00 Why Bill is Bullish on Bitcoin 4:22 Gemini Orange Bitcoin Credit Card 5:43 Why Bitcoin Hasn't Gone Fully Mainstream…Yet 9:10 Rise of Corporate Bitcoin Treasuries 11:36 Bitcoin-Backed Securities 13:10 Bullish on Semler Scientific 14:18 Market Misunderstands Semler Scientific? 15:13 Bitcoin Rewards Cards and Points Inflation 17:54 Are Bitcoin Treasuries a Bubble 18:42 Stablecoins 20:42 Bitcoin and the US Treasury Market 21:34 Economic Outlook, Market Trends 23:05 Is a Recession Looming? 23:57 Bitcoin Price Predictions 25:13 The Future of Bitcoin Taxation 27:34 Inside Miller Value Partners’ Strategy 28:49 Can Anything Beat Bitcoin? 29:22 Becoming a Bitcoin Believer 30:03 Bitcoin Changes Your Worldview 31:22 Growing Up Around Legendary Investors 33:36 Bonding Over Bitcoin 35:19 The Pressure of a Famous Father 36:09 Trump vs. Elon Powered by Gemini with a shoutout about the Gemini credit card in the talk! 💳🧡

Natalie Brunell ⚡️

191,877 Aufrufe • vor 1 Jahr

NEW POD | EP 73 AI & Robotics Stealing Crypto's Thunder? Timestamps: 0:47 The Only Narrative in Crypto Right Now 1:41 Competitive Landscape in Crypto 4:50 Robinhood's Future Plans 11:42 Future of Crypto Innovation 12:00 Noise ( 13:15 Circle IPO 16:51 USDC vs USDT 18:00 Predictions for Circle's Future 18:49 Circle & CCTP 20:07 Kraken Acquiring NinjaTrader 21:19 Where Are We in The Market? 25:55 China, US and Global Markets 32:13 Manus vs. Gemini 34:12 Sesame AI 35:07 Will Google Have The Best AI? 37:51 AI Investment and Valuation 39:35 AI Coding 43:54 AI in Startup Development and Innovation 44:15 46:24 Niche Apps and Micro Unicorns 49:01 Young Founders & Education 50:15 "Google/Harvard doesn't give you status" - Nikita Bier 52:04 Alpha School - AI-Powered Private School 55:19 AI Doctors 1:00:39 AI Replacing Jr. Jobs 1:02:19 Parenting and Teaching Agency 1:05:46 Tailoring Education to Children's Interests Link ↓

NEW POD | EP 73 AI & Robotics Stealing Crypto's Thunder? Timestamps: 0:47 The Only Narrative in Crypto Right Now 1:41 Competitive Landscape in Crypto 4:50 Robinhood's Future Plans 11:42 Future of Crypto Innovation 12:00 Noise ( 13:15 Circle IPO 16:51 USDC vs USDT 18:00 Predictions for Circle's Future 18:49 Circle & CCTP 20:07 Kraken Acquiring NinjaTrader 21:19 Where Are We in The Market? 25:55 China, US and Global Markets 32:13 Manus vs. Gemini 34:12 Sesame AI 35:07 Will Google Have The Best AI? 37:51 AI Investment and Valuation 39:35 AI Coding 43:54 AI in Startup Development and Innovation 44:15 46:24 Niche Apps and Micro Unicorns 49:01 Young Founders & Education 50:15 "Google/Harvard doesn't give you status" - Nikita Bier 52:04 Alpha School - AI-Powered Private School 55:19 AI Doctors 1:00:39 AI Replacing Jr. Jobs 1:02:19 Parenting and Teaching Agency 1:05:46 Tailoring Education to Children's Interests Link ↓

Good Game

11,576 Aufrufe • vor 1 Jahr

Sonnet 4.5 & the AI Plateau Myth — epic conversation with Sholto Douglas of Anthropic 0:00 - Intro 1:09 - What's Behind The Rapid Pace of AI Releases at Anthropic 2:49 - Opus, Sonnet, and Haiku Model Tiers 4:14 - Sholto's Story: From Australian Fencer to AI Researcher 12:01 - The YouTube Effect: Mastery Through Observation 16:16 - Breaking Into AI Research Without Traditional Academic Signals 18:29 - DeepMind, Gemini, and Building Inference Stacks 23:05 - Why Anthropic? Culture and Mission Differences Amongst AI Research Labs 25:08 - What Is "Taste" in AI Research? 31:46 - This Week's Big Launch: Sonnet 4.5, Best Coding Model in the World 36:40 - From 7 Hours to 30 Hours: The Long-Running AI Agent Breakthrough 38:41 - How AI Agents Self-Correct and Maintain Coherence 43:13 - The Role of Memory in Extended Coding Sessions 47:42 - Pre-Training vs. RL: Textbooks vs. Worked Problems 52:11 - Test-Time Compute & Reinforcement Learning 55:55 - Why RL Finally Started Working on LLMs in 2024 59:38 - The Path to AGI 1:02:05 - Are We Hitting a Plateau in AI? So Many Low Hanging Fruits 1:03:41 - Beyond Coding: GDPVal & Impact Economic Sectors 1:05:47 - Preparing for 10-100x Individual Leverage & The Upcoming Robotics Explosion

Sonnet 4.5 & the AI Plateau Myth — epic conversation with Sholto Douglas of Anthropic 0:00 - Intro 1:09 - What's Behind The Rapid Pace of AI Releases at Anthropic 2:49 - Opus, Sonnet, and Haiku Model Tiers 4:14 - Sholto's Story: From Australian Fencer to AI Researcher 12:01 - The YouTube Effect: Mastery Through Observation 16:16 - Breaking Into AI Research Without Traditional Academic Signals 18:29 - DeepMind, Gemini, and Building Inference Stacks 23:05 - Why Anthropic? Culture and Mission Differences Amongst AI Research Labs 25:08 - What Is "Taste" in AI Research? 31:46 - This Week's Big Launch: Sonnet 4.5, Best Coding Model in the World 36:40 - From 7 Hours to 30 Hours: The Long-Running AI Agent Breakthrough 38:41 - How AI Agents Self-Correct and Maintain Coherence 43:13 - The Role of Memory in Extended Coding Sessions 47:42 - Pre-Training vs. RL: Textbooks vs. Worked Problems 52:11 - Test-Time Compute & Reinforcement Learning 55:55 - Why RL Finally Started Working on LLMs in 2024 59:38 - The Path to AGI 1:02:05 - Are We Hitting a Plateau in AI? So Many Low Hanging Fruits 1:03:41 - Beyond Coding: GDPVal & Impact Economic Sectors 1:05:47 - Preparing for 10-100x Individual Leverage & The Upcoming Robotics Explosion

Matt Turck

93,678 Aufrufe • vor 10 Monaten

Why AI Progress Suddenly Feels Real - my conversation with Yann Dubois, who co-leads the Post-Training Frontiers team at OpenAI 00:00 - Intro 01:30 - Why recent AI progress feels like a step function 04:13 - Model reliability & the emotional rollercoaster of shipping GPT-5.5 07:33 - How OpenAI structures vertical and horizontal teams 09:49 - Improving model efficiency and test-time compute 12:32 - Yann's journey from Switzerland to OpenAI 15:37 - Reasoning in 2026: Real-world utility vs verifiable rewards 18:34 - GPT-5.5 Thinking vs Pro: Scaling test-time compute 20:09 - How reasoning models become more efficient 23:23 - Pre-training scaling and overcoming the data wall 27:03 - Multimodal data, synthetic data, and embodied AI 31:05 - Demystifying mid-training and post-training 37:21 - Does RL create new capabilities in AI? 38:53 - The challenges and frontier of scaling RL 43:09 - Is building AI models a craft or a strict science 48:21 - How AI models generalize across different domains 54:18 - How reinforcement learning cures AI hallucinations 56:04 - Negative generalization and conflicting instructions 58:05 - Can RL scale to law, medicine, and the broader economy? 1:00:19 - The evaluation bottleneck and Model as a Judge 1:04:21 - Continuous AI progress & continual learning 1:08:49 - Will foundation models eat the agent harness 1:11:23 - Why startups should focus on the last mile of AI

Why AI Progress Suddenly Feels Real - my conversation with Yann Dubois, who co-leads the Post-Training Frontiers team at OpenAI 00:00 - Intro 01:30 - Why recent AI progress feels like a step function 04:13 - Model reliability & the emotional rollercoaster of shipping GPT-5.5 07:33 - How OpenAI structures vertical and horizontal teams 09:49 - Improving model efficiency and test-time compute 12:32 - Yann's journey from Switzerland to OpenAI 15:37 - Reasoning in 2026: Real-world utility vs verifiable rewards 18:34 - GPT-5.5 Thinking vs Pro: Scaling test-time compute 20:09 - How reasoning models become more efficient 23:23 - Pre-training scaling and overcoming the data wall 27:03 - Multimodal data, synthetic data, and embodied AI 31:05 - Demystifying mid-training and post-training 37:21 - Does RL create new capabilities in AI? 38:53 - The challenges and frontier of scaling RL 43:09 - Is building AI models a craft or a strict science 48:21 - How AI models generalize across different domains 54:18 - How reinforcement learning cures AI hallucinations 56:04 - Negative generalization and conflicting instructions 58:05 - Can RL scale to law, medicine, and the broader economy? 1:00:19 - The evaluation bottleneck and Model as a Judge 1:04:21 - Continuous AI progress & continual learning 1:08:49 - Will foundation models eat the agent harness 1:11:23 - Why startups should focus on the last mile of AI

Matt Turck

100,966 Aufrufe • vor 2 Monaten

After many months, a lot of model releases and even more Amp releases, here it is: A new episode of Raising An Agent with Quinn Slack and Thorsten Ball. They discuss: - Why Gemini 3 might have been the inflection point, not Opus 4.5 - What changed over Christmas and why everybody talks about agents now - "The assistant is dead, long live the factory" - How codebases need to adapt for agents - Why small/local models don't give you an impression of the frontier - What does it mean for a codebase to be agent-friendly? - New things coming to Amp Enjoy! And happy hacking! Timestamps: 1:51 The Impact of Gemini 3 and Opus 4.5 5:28 Shifting Perceptions of AI Agents 10:17 Adapting Codebases for AI Efficiency 14:48 The Future of Software Development with AI Agents 19:29 Challenges and Opportunities Ahead 24:00 The Importance of Frontier Models 25:12 Adapting to AI-Driven Workflows 26:57 The Future of Coding Agents 28:07 Building an Agentic Code Base 30:22 Rethinking Development Tools 31:44 Amp's Evolution and Team Dynamics 33:00 Discord Q&A 39:18 Teasing New Features: Deep

After many months, a lot of model releases and even more Amp releases, here it is: A new episode of Raising An Agent with Quinn Slack and Thorsten Ball. They discuss: - Why Gemini 3 might have been the inflection point, not Opus 4.5 - What changed over Christmas and why everybody talks about agents now - "The assistant is dead, long live the factory" - How codebases need to adapt for agents - Why small/local models don't give you an impression of the frontier - What does it mean for a codebase to be agent-friendly? - New things coming to Amp Enjoy! And happy hacking! Timestamps: 1:51 The Impact of Gemini 3 and Opus 4.5 5:28 Shifting Perceptions of AI Agents 10:17 Adapting Codebases for AI Efficiency 14:48 The Future of Software Development with AI Agents 19:29 Challenges and Opportunities Ahead 24:00 The Importance of Frontier Models 25:12 Adapting to AI-Driven Workflows 26:57 The Future of Coding Agents 28:07 Building an Agentic Code Base 30:22 Rethinking Development Tools 31:44 Amp's Evolution and Team Dynamics 33:00 Discord Q&A 39:18 Teasing New Features: Deep

Amp

36,150 Aufrufe • vor 6 Monaten

Manish Gupta, Senior Director at Google DeepMind India, sits down with Aakrit Vaish and Pratyush Choudhury at Mumbai Tech Week for a rare on-record conversation about the frontier AI research happening out of Bangalore. Gupta makes a pointed case against the narrative that India lacks AI research talent: a team of roughly 75 researchers, a third of them fresh out of college, producing work on par with the best in the world and feeding directly into Gemini. The conversation goes deep on what DeepMind India actually builds, why Gemini is considered the most efficient model on the planet, and what India needs to become a research leader rather than a fast follower. In this conversation, they go deep on: 0:00 Intro: Manish Gupta of DeepMind India at Mumbai Tech Week 1:23 What DeepMind India does, and why it's a "mystery" 1:59 The three roles: languages, efficiency, continual learning 2:19 The Haryanvi demo at Google I/O and the cultural playbook 2:38 Making models efficient: from mobile to servers 3:25 Matryoshka transformers and why nested models win 4:20 Why Gemini is the most efficient model on the planet 5:11 Continual learning: using Gemini to improve Gemini 5:50 Google's India plans: consumers, agents, enterprise, government 7:44 Government officials live-coding with AI Studio and NotebookLM 8:30 The India AI talent debate and how the team is structured 10:25 Why India lacks courage and R&D investment, not talent 11:16 DeepMind's global labs and India's outsized impact 13:15 On-device AI: Gemma 3n and 4n 14:08 The headline: 75 world-class researchers in Bangalore 14:58 25 of the 75 are fresh out of college If you're a founder, builder or researcher thinking about AI, frontier models, or India's place in global AI research, this one's for you. Aakrit Vaish Pratyush Choudhury (PC) Google DeepMind Google India Manish Gupta

Manish Gupta, Senior Director at Google DeepMind India, sits down with Aakrit Vaish and Pratyush Choudhury at Mumbai Tech Week for a rare on-record conversation about the frontier AI research happening out of Bangalore. Gupta makes a pointed case against the narrative that India lacks AI research talent: a team of roughly 75 researchers, a third of them fresh out of college, producing work on par with the best in the world and feeding directly into Gemini. The conversation goes deep on what DeepMind India actually builds, why Gemini is considered the most efficient model on the planet, and what India needs to become a research leader rather than a fast follower. In this conversation, they go deep on: 0:00 Intro: Manish Gupta of DeepMind India at Mumbai Tech Week 1:23 What DeepMind India does, and why it's a "mystery" 1:59 The three roles: languages, efficiency, continual learning 2:19 The Haryanvi demo at Google I/O and the cultural playbook 2:38 Making models efficient: from mobile to servers 3:25 Matryoshka transformers and why nested models win 4:20 Why Gemini is the most efficient model on the planet 5:11 Continual learning: using Gemini to improve Gemini 5:50 Google's India plans: consumers, agents, enterprise, government 7:44 Government officials live-coding with AI Studio and NotebookLM 8:30 The India AI talent debate and how the team is structured 10:25 Why India lacks courage and R&D investment, not talent 11:16 DeepMind's global labs and India's outsized impact 13:15 On-device AI: Gemma 3n and 4n 14:08 The headline: 75 world-class researchers in Bangalore 14:58 25 of the 75 are fresh out of college If you're a founder, builder or researcher thinking about AI, frontier models, or India's place in global AI research, this one's for you. Aakrit Vaish Pratyush Choudhury (PC) Google DeepMind Google India Manish Gupta

Activate

14,136 Aufrufe • vor 2 Monaten

Just so you know Logan Kilpatrick and Josh Woodward — Gemini 2.5 & 3 Pro saved my life. I’m really not ready to talk about my heart condition or more so— the heart condition I THOUGHT I had, but this MUST be shared! G3mini really saved my fucking life! Gemini has: 🎯 helped me ask the right questions… 🎯 helped me understand what’s going on in my body before I even go to a doctor… 🎯 ask for extra blood panels, testing things that doctors seem to let slide… 🎯 Even more so - calm me down realizing that my heart is much stronger than I realize… 🎯 Reminds me of things I forget about when I ask it about other medication effects. 🎯 Recommended that I NOT take a specific supplement because of the side effect risk associated with another medicine I am taking. — (Reishi Mushrooms 🍄) I then told my doctor and she said Gemini was correct to tell me not to start taking those mushrooms until I am off the medication I am taking. It could have severe side effects. Again, I am not ready to share all the information but I am beyond grateful to have a PhD level doctor in my pocket at all time. Here is a snippet of the video that I uploaded to Gemini along with 4 other videos of the tech doing an Echocardiogram and Gemini shared a result that I have yet to hear from the doctor. I have one more appointment on January 5th to see if the main doctor is going to tell me what Gemini has shared. As I type this, I am scheduling a 2nd and 3rd opinion with secondary doctors in separate network affiliations to be even more safe. Among receiving the results and confirmation from G3mini from these test, I realize that Gemini 3.0 Pro lifted 15 elephants off my chest & legit saved my life! Here below 👇🏻 is a tiny snippet of my heart valve opening and closing pumping blood through it 🩸 We are magical creatures! 🪄 So complex, yet insanely fragile!

Just so you know Logan Kilpatrick and Josh Woodward — Gemini 2.5 & 3 Pro saved my life. I’m really not ready to talk about my heart condition or more so— the heart condition I THOUGHT I had, but this MUST be shared! G3mini really saved my fucking life! Gemini has: 🎯 helped me ask the right questions… 🎯 helped me understand what’s going on in my body before I even go to a doctor… 🎯 ask for extra blood panels, testing things that doctors seem to let slide… 🎯 Even more so - calm me down realizing that my heart is much stronger than I realize… 🎯 Reminds me of things I forget about when I ask it about other medication effects. 🎯 Recommended that I NOT take a specific supplement because of the side effect risk associated with another medicine I am taking. — (Reishi Mushrooms 🍄) I then told my doctor and she said Gemini was correct to tell me not to start taking those mushrooms until I am off the medication I am taking. It could have severe side effects. Again, I am not ready to share all the information but I am beyond grateful to have a PhD level doctor in my pocket at all time. Here is a snippet of the video that I uploaded to Gemini along with 4 other videos of the tech doing an Echocardiogram and Gemini shared a result that I have yet to hear from the doctor. I have one more appointment on January 5th to see if the main doctor is going to tell me what Gemini has shared. As I type this, I am scheduling a 2nd and 3rd opinion with secondary doctors in separate network affiliations to be even more safe. Among receiving the results and confirmation from G3mini from these test, I realize that Gemini 3.0 Pro lifted 15 elephants off my chest & legit saved my life! Here below 👇🏻 is a tiny snippet of my heart valve opening and closing pumping blood through it 🩸 We are magical creatures! 🪄 So complex, yet insanely fragile!

ChrisUniverse 🗽

15,003 Aufrufe • vor 7 Monaten