วีดีโอ

Is there any indication that there will be multimodal devices?

OK...This is probably the most interesting TVD you've published. I'm trying to rebuild my Mac (Mini) environment and I'd like to work on some of this. My MM isn't working (apparently the disk is corrupt from not using the machine for years. I'm going to get an M4 when I can get one discounted a bit. At that point I'd like to chat with you John to try to figure it out.

Well done!!!

Could you please guide to make a multi agent framework with voice integration, like a very low latency voice agent with langgraph integration

Great description and ambitious project. Super useful combination of everyday Google technology, and cutting edge integration with Gemini. Looking forward to seeing it in action!

Mark and Allen - What is your opinion HuggingFace? I believe HuggingFace has the Model registry, keeping the Code on Docker Image separate from the Model.

Congrats Allen and Matk 👏

Congratulations on 200th episode! And great talk - casual, fun and inclusive!

Grrrrrr

Great to see you both at I/O!

P r o m o S M 😭

Hi would you be able to create a video guide on how to install your plugin please?

Being a developer these days is like trying to build a sandcastle on a tsunami prone beach!

getting a rabbit this month, but how can I practice with a LAM in the meantime? like where can I practice with LAMs?

I like y'all

Important topic/conversation. Thanks for the discussion

Have you guys ever used pyannote audio before? I'm trying to figure out what's best to get the "essence" of someones voice from their voice embedding. Do I extract embeddings from longer audio clips of the person's voice? Or do I extract embeddings from 30ms chunks of voice audio and then obtain an average embedding from the embeddings I extracted?

waste of my time.

super interesting ways to test this new model. thanks! can't wait to get my hands on it

hey, congrats on preview :) Just curious, did you get access via Google AI Studio or Vertex?

Also, it should not only limit to English, i,e, the phonemes! This can apply to how we can learn multiple languages!

This is so WILD!!! 😯😳🤯

Hi...can we talk about how to use LLM to apply fuzzy logic for capacitor management in pv grid??

21:29 they say 21 minutes and 29 seconds into the 44 minute video, 22:12; try content + '...' + contents23:07 pairing with top_p and top_k can influence negative randomness; use coroutine without mallocate;25:26 would be helpful which seems to be an anomaly in these ecosystems.

How about (maybe temporarily) increasing the time for response?

Hey, I know these guys

Happy Thanksgiving🎉

You said, this seemed to be a rushed release. Maybe the sudden firing of Altman is related to this.

Hi, I recently found your channel and like the style. On what other channel I might contact you?

Great discussion Noble and Allen. I especially liked the part starting at 22:54 😀 Spot on about the relational DB analogy and yes we need "SQL for LLMs" so to speak. And yes, next year more talks on QA and how to take LLM powered applications to production.

good information. do you have any direct information on configure Google Matching Engine? Hard to find good tutorials on local set up on my machine.

I have an observation that might have made this discussion clearer at the start: the model is just data for the transformer, much like tables are/contain the data for a SQL server. The question that Allen was asking was: what uses the model, which you finally answered. I have two related questions: 1) How does one tell the transformer library to use my local GPU? Is that even possible from a python interpreter framework? 2) What it would take to hook up langchain to your locally running transformer instance, so you could still code using the langchain paradigm/framework which people are more familiar with?

Global Intents !!

Is this like Air? The video that we have been seeing on LinkedIn with customer service? 🤔

Fascinating..

⚡⚡⚡⚡⚡⚡⚡⚡🔥🔥! Crack

Awesome episode, I just discovered LangChain after many failed attempts at creating the raw code myself in my Nextjs webapp to force GPT api to search the internet. Now I don't have to reinvent the wheel, can just implemet LangChain by deploying it on Huggingface. Thanks for the insights

This is gold. Thank you.

😁🎅

😍

I think App Actions will be a bigger failure than Conversational Actions apparently was, although Conversational Actions actually extended Assistant's functionality opening up enormous possibilities whereas App Actions just launches apps. Shortcuts into an app?!?! Seriously?!?! That is their big idea? I don't speak to my phone in public. I think the primary location where voice apps are used is at home. If you're at home you likely have cheap smart speakers everywhere, and your phone is on a charger or the counter. I think leaving smart speakers/displays behind is a huge mistake, and creating separate Assistant functionality based on user surface is a huge mistake. If I need to remember what I can say to Google Assistant based on what I am speaking to, I'm not going to use it.

😈 ρяσмσѕм

Very interesting...

I'm watching and...

Ah! I worked on a similar project during my first year in Conv. AI. It was a CRUD application, basically, having a Google Spreadsheet as db. You could create a knowledge base by voice. I haven't seen many use cases of this kind so far. That would be suitable for assurances or banks. Great demo, btw! I'm not sure about the prompt 'prompt me': I would have the assistant say directly 'what's your weight today?' or something similar. The conversation would be shorter and anyone could understand what the app is about. But I do understand that this a personal tool.

So in the callback discussion I'm seeing a video starting where a troll comes out and nails the badge to the wall.

Started with a TI 99 4/A playing Munchman and Parsec. Learned to code Basic there. Then got a Commodore 128 and learned to code Extended Basic. Eventually graduated to a PC with Windows 3.0. (The C-128 got me through college.)

👀 p̶r̶o̶m̶o̶s̶m̶

Voiceflow's api tester is pretty straightforward...Postman must be easier. now..or I know more now

Design a sample so that if directions are followed...it works