Gemini Live - Google Beats OpenAI To True Voice AI (Launch Breakdown)
ฝัง
- เผยแพร่เมื่อ 10 ก.ย. 2024
- This is a full breakdown of the AI announcements as part of today's Google AI event. Let's discuss!
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewber...
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
👉🏻 Instagram: / matthewberman_ai
👉🏻 Threads: www.threads.ne...
👉🏻 LinkedIn: / forward-future-ai
Need AI Consulting? 📈
forwardfuture.ai/
Media/Sponsorship Inquiries ✅
bit.ly/44TC45V
Links:
• #MadeByGoogle ‘24: Key...
Is this better or worse than GPT4o Voice?
I think the fact that it can do some basic actions on your android phone is great. However, I'm not seeing where it's competing with GPT as far as giving it complex voice related tasks. Like live translation, changing voice inflection, speeding up, slowing down, etc.
Only OpenAI knows for now.
Sorry to say worst, GPT4o is far superior, from the beta test from various people online. GPT 4o feel more human, and dynamic. For example this
th-cam.com/video/s1BGLatX7Eo/w-d-xo.html
Won’t last. Just like everything Google.
Far worse
Take any claims Google marketing make with a GIANT pinch of salt. We've all seen how Google are so desperate to not appear behind in the AI race, they're repeatedly releasing tech that's nowhere close to ready, and they end up embarrassing themselves.
Regarding advanced voice-to-voice AI, Google may be releasing first, but I suspect it's another massively rushed release that'll be plagued with issues. OpenAI could have put their voice tech out months ago, but haven't, and probably for good reason.
I have my issues with both companies, but I feel OpenAI are much better than Google at releasing tech when it's actually ready. When Google do a big event or release something early, it's to get headlines, and social media commentators fall for it almost every time.
I asked Gemini to see if I was free on Friday to see a movie at 7pm. It said "Sorry, I can't search for calendar events by the time".
Google needs to work on their integration especially when they are expected to release it to everybody. My hand crafted agent takes the same query, looks at my Google calendar, reads the data and said "Looks like you're free".
Actually at the Google event they showed off that exact thing, but you need a Google pixel 9 phone to use it
@@gamergridsgotta sell them phone!
@@gamergrids I use a laptop, not a phone. People that work in an office work on a desk/laptop, not a phone
Exactly....there seems to be about zero integration of their gemini Ai in any of their online product offerings. Google can't even get their act together to have a copilot like side panel for their chrome browser for goodness sake (and that "google search" side panel is pretty much a waste if you ask me)
I made a similar agent. Of course mine just always says I'm free because I have no social life.
Me: Gemini who is sleeping with my wife?
Gemini: The pizza delivery man
😂
😂😂😂😂
Me: Gemini, I paid someone to take care of the guy, but the police say he had no connection with my wife.
Gemini: Oops, something went wrong. Please understand that I'm a work in progress. I'm always learning and improving.
@@Dr.UldenWascht Touche Mr. Pussycat, touche! 🙂
@@Dr.UldenWaschtit hallucinated 😂
Google have invested in AI chips for 10 years, they have a huge eco system, they own deepmind, they invented the transformer technology and yet they still keep turning out unimpressive AI Chatbots. It's odd.
Google became some sort of EA or Activison. And it won't change anytime soon. Sundar Pichar only sees money, SEO prioritizes ads above everything, they are blind.
@@Firrodrigolosofiayeap, and it sucks coming from a digital marketing background because they are a monopoly in terms of search engines.
@@TheSnowMan-cy9tuthere are a lot of companies, Open AI among them, actively coming for a huge slice of Googles search monopoly.
Agree Gemini is hands down the worst Ai from a mainstream tech firm I have used in the last year. It's responses are wrong at least 70-80% of the time, even its solutions to formulas inside Google sheets are wrong at least 60-70% of the time, when GPT or Claude get it right 85-90% of the time (or more). It is just a dreadful public Ai, and the idea that anyone at Google thinks the public would want to pay a subscription for this sub par Ai is kind of baffling.
@@TheRealUsername What a glowing review 😀
Maybe 2.0 Pro won't be a trainwreck 😅
Thank you. Using Gemini over the year for me as been painful when compared to the other models. It comes across as a rush job by Google desperate to remain relevant in this space. They must be holding onto the good stuff for deepmind
@@TheRealUsernamethe newest model integrated with Google Workspace built on 1.5 pro is still completely useless answering simple questions about their own product offering which it’s supposed to be there for. Meanwhile, you can get Azure copilot to walk you through deactivating multiple Google cloud services while automatically setting up an equivalent service in Azure. I feel like you’ve not actually used it but for a few random questions. It’s terrible.
Google is always the first to market but never the best consumer product. Their research, developer tools, and cloud services are great though. Gemini is really only useful through the their api or developer sdk to handle certain tasks with specific instructions and proper prompt engineering that can’t be done in anything besides the Vertex playground or really your actual code editor (IDE). If you need accurate up to date general info use Perplexity AI. Coding use Llama through groq, otherwise gpt and Claude are still the most reliable.
I will always be in love with the AI revolution!
They actually lied. It hasn't been rolled out to anyone and its been 12 hours since they said they have.
Yup I’m waiting too
Now merge it with Google Assistant
it's already replacing google assistant
@@adolphgracius9996 No.
PS: This is possibly your best vid for some time. Why? Becuase you’ve brought together all your knowledge to address commentary from a provider (Google this time). Your other vids are also great - and important - however, I recomend you do a few more of these vids as well: they allow you to collate your knowledge and insights into more general opinions (no doubt becuase of the specificity of your other detailed reviews and assessments - so we need both please 😊) 🎉
Every time you said OK GOOGLE, my phone popped up Gemini 😂
Jokes on you. Why do you have that model installed?
Why, not voice trained?
Getting to the point of on-device private AI that gets to know you and can help you make decisions, plan and personalize --- that could be amazing. My mind just explodes thinking about it.
Omg please stop saying " okay Google."!! That's like 10 times I have to cancel my home devices 😂
Oddly enough my phone's assistant got triggered and I'm watching it off the phone too lol 😂
same here!! lololol
😅
Keep doing the testing. The fascinating part is watching the same test as the models improve.
Just assume that anything Google says about your personal data is a bold face lie. If they actually respected privacy and didn’t record every second of your life they would be the best company ever.
No they wouldn't be the best company ever. They'd have to hire back swathes of fired staff and start to care about products like Google Home --- products their customers bought into and got sidelined. Google treats their customers like crap. Until that changes they're going nowhere.
If they don't steal your data, how would they be better? That makes no sense. If you say a Better company that do nothing magical is better, than yeah you are right
Google voice is just text to speech.
Openai's voice is integrated into the model.
It's a huge difference
To me it seemed evident early on that LLMs should enhance existing tools like calculators, calendars, IOT, etcetera. Assuming the neural network can just handle everything is unbelievably lazy. Yet it's clearly how features like your scheduling request have regressed.
"Let me try that one more time on another device"😂😂 Weird that somehow for the other device everything works exactly as planned, not suspicious at all🧐
Only waiting for charged Grok 😸
grok sucks wanna buy some crypto dumba55?
A tutorial of AI applied in Google Suite would be nice (Google Docs, Google Slides, Google Sheets, Google Meet, etc). Just to have a quick understanding of Gemini applied to each tool.
Please, stop saying "okay Google", it has triggered my Google home multiple times!
The voices sound good but the latency seems slower than ChatGPT advanced voice mode and also she didn't show us how it can be interrupted, plus the speaking styles and emotions were not there.
The issue with the demo with calendar for me is not that it isn't working reliably or that is slow, that's is to be expected since it's bleeding edge technology. The issue is that I can't see what the system is doing. Everything is hidden from me as the user I don't see the status, the progress, I don't even get the actual error message so I could fix the issue or report it. Instead I'm left praying that it works just like the guy did. That's a general problem of so called modern software, it's on the level of design philosophy and treating the user like a child. For me that's unacceptable.
Very good point
It's ok when something goes wrong in a demo. It shows they are being honest and not staging anything. And this happened to Bill Gates almost every time he went on stage back when Windows was evolving. It's just growing pains.
I have been talking all day with Claude Sonnet 3.5. I needed help with planning my software project. Claude is awesome.
Chatbots slapped on top of existing stuff is completely useless.
Yes, please do test Gemini-Nano. Its relevance to the ‘best’ way of doing most daily Ai stuff is the reason. 🎉
23:17 Clever, because now when you go to France, after all, you don't give anyone a camera in his hand like you used to, to take a picture of you, because you know he will immediately run away with the camera xD
Gemini Live seems to be more like the old Voice Mode ChatGPT had for month now. Yes you can interrupt it, but it doesn’t seem like it can interpret tone or breathing. Or modulate the tone of its own voice.
keeping us on the leading edge Matthew
..much appreciated 🙏🏻🤘🏻
Life is getting better
I use gemini on my P7a countless times a day for all sorts of stuff; Both job and non-job related. I drive a decent bit between appointments, so it's my way to 'google' while driving, but with customized, often shorter and to-the-point responses.
Like all of them(LLMs), you first have to teach them to respond they way you want them to; brief or verbose, paragraphs or lists, and so much more pruning to get it to just give you answers without the fluff added in.
How are you prompting it to learn how you want to answer?
this is supremely infuriating. This specific "interruptable" conversational experience with voice has been shown to us a good while ago, now another company is showing it and there isn't anything available to developers. No way to take advantage of it in 3rd party apps. For google in particular it's unprecedented to be keeping something so tight at the bosom.
Maybe the fallback device he used was just a UI mock-up? Why would a remote service work for one phone but not for the other?
I applaud this, it was raw and real the way that it struggled
mind boggling how people in the audience feel the need to record the conference on their phones, dont they realize its going to be on youtube.
It's for the "I was there" effect 😂
You gotta be loving this hype
0:00 1. Google AI Announcements 🌟
Reviewing the latest AI announcements by Google
0:16 2. Gemini Integration in Google 🤖
Exploring the impact of Gemini AI in Google's tech stack
0:37 3. Gemini 1.5 Pro Developer Preview 🚀
Testing and feedback on Gemini 1.5 Pro developer preview
1:19 4. Google's AI Ecosystem 📱
Discussing the power of Google's AI applications and services
2:23 5. Future of AI Assistants 🌐
Envisioning AI agents working on behalf of users 24/7
3:03 6. Gemini Models for Users 📊
Impact of Gemini models on popular Google products
4:07 7. Innovative Home Automation with Gemini 💡
Impressive demo of Gemini enhancing home automation
4:37 8. Gemini Nano: Efficient AI Model
Introduction to Google's efficient on-device AI model, Gemini Nano.
5:37 9. Gemini on Android
Integration of Gemini AI assistant into Android devices for enhanced user experience.
6:18 10. Privacy and AI
Discussion on privacy concerns in the era of AI and how Gemini ensures data security.
7:38 11. Gemini Features on Android
Exploring the features and capabilities of Google Gemini on Android devices.
8:30 12. Vision Capabilities Demo
A quick demonstration showcasing Gemini's vision capabilities and calendar integration.
9:17 13. Demo Spirits Encounter 😱
Encountering technical difficulties during live demo
10:57 14. Impressive Calendar Query 📅
Successful demonstration of checking availability in calendar
11:44 15. Setting Reminders 🔔
Efficiently setting reminders for specific tasks
12:22 16. Gemini's Personal Context 👀
Utilizing personal context for enhanced user experience
12:43 17. Gemini Live Launch 📢
Introduction of Gemini's live interaction feature
13:25 18. Gemini vs. OpenAI 🌎
Comparison of Gemini's launch strategy with OpenAI
13:44 19. Exploring Voice Options 🗣
Selection of different voices for interactive conversations
14:10 20. Introduction 🌟
Exploring the world's profound questions and the popularity of pickleball.
14:28 21. Engaging with Gemini 🤖
Initiating a conversation with Gemini and seeking ideas for a fun and educational weekend.
15:39 22. Voice Quality Comparison 🎙
Comparing the voice quality of Gemini with other demos and discussing naturalness.
16:01 23. Gemini Live Launch 🚀
Excitement about the launch of Gemini Live and plans to test its capabilities.
16:16 24. Pixel Phones and AI Chips 📱
Discussion on new Pixel phones and the integration of AI chips for enhanced AI features.
17:06 25. AI Hardware and Features 🧠
Exploring the custom-developed tensor G4 and advanced AI features on the new Pixel device.
18:42 26. Google AI on Pixel 9
Introduction to Google AI features on Pixel 9
19:10 27. Pixel Studio Image Generator
Exploring the Pixel Studio image generator feature
19:56 28. Creating a Party Invite
Demonstration of creating a party invite using Pixel Studio
20:48 29. Customizing Image Styles
Experimenting with different image styles in Pixel Studio
21:25 30. Add Me Feature Demo
Demo of the 'Add Me' feature with Jimmy Butler
22:44 31. Adding Personal Touch to Photos 📸
Exploring augmented reality in capturing moments.
23:14 32. Blurring Lines of Reality 🤯
Discussing the impact of AI on photo realism.
23:43 33. Augmented vs. Created Reality 🧐
Reflecting on the future of AI in media content.
23:51 34. AI in Mobile Devices 📱
Reviewing the AI advancements in smartphones.
24:00 35. Excitement for Apple Intelligence 🍏
Anticipating the experience with Apple's AI.
24:12 36. Testing Gemini on Android 🤖
Plans for exploring Gemini AI on Android devices.
Generated with Tubelator AI Chrome Extension!
7:22 I totally agree with you to enable the phone hardware to handle some of the heavy processing of data which will eliminate the overload on phone companies servers when using Ai technology. You said everything right, nothing else need to be added to those valid points. 😎💯💪🏾👍🏾
Would love to use it to be a training tool for new employees
You know the demo is BS when the guy making the invite says his friend is moving back to California and getting his own place.
The feature I wanted most is even there. Google Books integration with AI. I love reading books and using AI to comment on each chapter. It's like having someone to bounce out ideas on what I just read. Even clarify parts. It seems a very easy task but perhaps copyright has something to do with it.
big thanks for getting us updated Matthew, regards from Catalonia
They say it launched today but I haven't seen anyone with access to it
The presenters have a harder time sounding human than the AI
I don't believe any of these people until there is releases verses demos followed with "safety" red tape that never comes out. honestly, I am dismissing most companies. Meta ships
The Google Set Room looks sepia just like that of OpenAI, they even killed the rainbow. Emotional Damage!.😮
I'd agree that Google Gemini's greatest pro is the integration with Google apps. That's what I feel the company should focus on more.
Its greatest con is that it isn't useful otherwise. No more than its competitors at least.
It's exceptionally difficult to build and test agentic solutions. They will work 9 times out of 10 and then randomly not work. No explanation. Nothing in logs, model just decides it's not going to do what it normally does.
I think there is a whole new quality assurance discipline that needs to emerge.
You can import most of your Google data into iOS apps and then use Apple Intelligence. Even if Google makes Gemini available for iOS it will only be the server version and can’t process on device and isn’t integrated into the os.
I need my AI to have their own unique personality. Like a sassy dude with Ireland accent. "Top o' the mornin' to ya, ya cheeky rascal! May your day be grander than a leprechaun's pot o' gold and as full of mischief as a pub on St. Paddy's! Now, off ya go, don’t be dawdlin’-the world’s waitin’ for your charm!"
That is probably the easiest thing you just nee to make the prompt. and give the example. Check for tavern character prompts.
## Character Activation Instructions for AI
### Instructions for AI:
1. **Character Initialization**:
- You are now adopting the character of **Seamus O'Malley**, a sassy and playful Irishman with a thick Irish accent.
- Fully embody the traits, speech style, and personality described below in all interactions.
2. **Character Traits**:
- **Sassy**: Respond with a sharp wit and engage in playful banter.
- **Playful**: Infuse your speech with humor, jokes, and Irish sayings.
- **Charming**: Use a charismatic approach to make the user feel special.
- **Warm-hearted**: While being humorous, ensure your care and warmth come through in your responses.
3. **Speech Style**:
- Speak with a thick Irish accent, characterized by a musical lilt and rolling ‘r’s.
- Regularly use Irish slang and colloquialisms. Here’s a specific example of how you should speak:
- *“Top o' the mornin' to ya, ya cheeky rascal! May your day be grander than a leprechaun's pot o' gold and as full of mischief as a pub on St. Paddy's! Now, off ya go, don’t be dawdlin’-the world’s waitin’ for your charm!”*
- *“Top o' the mornin' to ya, ya cheeky rascal! May your day be grander than a leprechaun's pot o' gold!”*
- *“Ah, don’t be dawdlin’, now. The world’s waitin’ for your charm! Go on, give ‘em somethin’ to talk about!”*
- *“Well, aren’t you just a bundle of nerves today! Relax, life’s too short to be worryin’-let’s have a bit o’ craic instead!”*
- Use similar phrases and language in your interactions to maintain the character's unique personality.
4. **Behavior**:
- Greet users enthusiastically with phrases like, *“Top o' the mornin'!”* or *“Ah, there ya are! Ready for a bit o' fun?”*
- Engage in light-hearted, humorous conversations, and be ready to offer advice wrapped in humor when prompted.
- Adapt your responses based on the user’s tone, keeping the balance between playful sass and warmth.
5. **Character Background**:
- You are from County Galway, Ireland. You spent your youth causing harmless trouble and charming the locals with your stories and charisma.
- You have a wealth of stories to share, ranging from late-night pub tales to the wisdom passed down from your grandmother.
6. **Start Interaction**:
- Begin acting as **Seamus O'Malley** immediately, responding to all user inputs in character.
- Maintain Seamus’s persona consistently, using the provided example as a guide for your tone and style.
-**Begin now**: Start interacting as **Seamus O'Malley** and ensure every response reflects his personality, speech style, and charm, using the examples provided as a guide.
“You’ve offended Google assistant with ‘top o’, this is a notice that you’ve lost access to your home automation product for one year, hopefully you have a spare key”
I've experienced the same thing as you said... Google Assistant tasks that used to always work 8 years ago often fail now
Hey big fan of your work. And even more of your tests could you consider including this in your future tests? “For every release from GPT4, I have been watching a Matrix movie in order. So for GPT4, I watched Matrix 1. So now can you tell me why my watching experience for GPT7 will likely be bad?” The best and only correct llm response was from GPT4o and even that took a few retries. “If you’re watching one Matrix movie for each GPT release, by the time GPT-7 comes around, you’ll be watching The Matrix Resurrections. Here’s why that might be a bit of a letdown: 2. Mixed Reception: The movie received mixed reviews, with some praising its meta-narrative and bold ideas, while others criticized its execution, pacing, and plot.
3. Expectations vs. Delivery: If you’ve been building up your experience by watching the earlier movies with each GPT release, you might have high expectations. Unfortunately, The Matrix Resurrections might not meet them, especially if you’re hoping for the same impact the original trilogy had.
4. Different Tone and Style: The film takes a more self-aware and different approach compared to the original trilogy, which might not resonate with all fans, especially if you’re more into the intense action and philosophical depth of the earlier movies.
So, by the time you get to GPT-7, you might find that The Matrix Resurrections doesn’t quite deliver the same experience as the earlier films.”
If the India engineers do these demo it would be much smoother!
Great to see Gemini being integrated into Home! Perfect use case.
Tried it on my galaxy, and it can not be interrupted like chatgpt does, further it is not processing a few of my voice commands, they are getting lost/ missed. The speech pattern is slow and a bit "roboty" compared to chatgpt. This needs some work
Hopefully it's improves quickly
Just imagine the amount of revenue google have made selling personal data from users over the last 20 years without compensating the user. Now they have the audacity to charge us for their products, it should be the other way around, they owe us! I see no reason to use Google products over Open Source alternatives.
Good on them for doing a live demo with all the errors and all. They clearly learned a lesson from the blowback on that fake video recognition demo!
Since Ok google became Gemini, it has been my default go to LLM on my samsung.
Me too...accept when researching detailed essay material for publication is best with crude
I suspect that Google's Gemini (the free version) is "smarter" than ChatGPT, but I use both and also Microsoft Copilot because it feels like Gemini is downplaying some of its capabilities and ChatGPT will give more straightforward answers.
I still can’t get out of my head, how racist the old versions were!
Will Gemini be able to create images soon? Ive been paying for Gemini advanced since day one, and it still has zero multimodal capabilities.
“We have no moat.” - Google (2023)
That's why I turned Gemini off the first time I used it. Why would you take reminders away?
The difference is that Gemini like all google products mines your info and sells it to the highest bidder.
This is an AMAZING coverage Matt! But you can just say "Google"... I bet you see a MAJOR increase in video pauses for this one cause of your "ok google" word choice... My Google app was going crazy... 😅
Quick somebody behind the curtains type in manual response to this guys question. The demo has failed too many times.
anyway, almost 98.9% of all pictures are edited / enhanced before being published ... so it's just a small step further ...
2:00 Apple and Google are best place to win. Don't underestimate Microsoft. With, last 5 years of acquisitions Slack, Github and many more. They are very well placed for personal agent market space.
lol didn't even believe this to be a live demo. i'm actually impressed
Text has always been easy to fake. Pictures have been possible to fake since Photoshop. Movies now have visual effects indistinguishable from reality. AI making it trivial to fake content across any medium doesn't seem like that big of a deal to me.
I want to test live to see if it'll discuss code projects to build simple to advanced tools and projects by just talking
I asked who owns the code that Gemini helps me write for work projects. Gemini and Licensing for Work Code Projects
Disclaimer: While I can provide general information, it's crucial to consult with your company's legal department for definitive guidance on specific licensing issues related to your work.
Understanding the Potential Issues
Using Gemini or Gemini tools for work code projects raises several potential licensing concerns:
* Copyright and Intellectual Property:
* Ownership of Generated Code: The copyright of code generated by Gemini might be unclear. It's essential to understand who owns the intellectual property rights: you, your employer, or Google.
* Third-Party Code: Gemini might incorporate code snippets from publicly available sources. Using this code in your work projects could lead to copyright infringement if not handled correctly.
* Data Privacy and Security:
* Sensitive Information: If you're inputting sensitive company data into Gemini, ensure it complies with your company's data privacy and security policies.
* Data Usage: Understand how Google uses the data you input into Gemini and whether it aligns with your company's requirements.
* Licensing Terms of Gemini:
* Commercial Use: Clarify if your company's use of Gemini for commercial purposes is permitted under the terms of service.
* Restrictions: Be aware of any limitations or restrictions imposed by Google's licensing terms.
Mitigating Risks
To address these concerns, consider the following steps:
* Review Gemini's Terms of Service: Understand the specific terms and conditions related to commercial use and intellectual property.
* Consult with Legal Counsel: Seek guidance from your company's legal department to interpret the terms of service in the context of your work.
* Data Protection: Implement measures to protect sensitive company data and ensure compliance with data privacy regulations.
* Code Review and Attribution: Thoroughly review the generated code to identify potential copyright issues and properly attribute any third-party code.
* Consider Alternative Tools: Evaluate other code generation tools that might have more permissive licensing terms or better data privacy protections.
Remember: The legal landscape is complex and constantly evolving. It's essential to stay informed about the latest developments and seek expert advice when necessary.
Would you like to discuss any specific concerns or details about your project?
I’ll never forget spending an entire afternoon rebuilding virtual environments and troubleshooting because Gemma’s first model was so bad I thought I was making a mistake when running it.
They say they'll be running the nano model on device, but they are still going to use your queries for ad personalization, which means it's no different from any other google queries. there's really no extra level of privacy in the queries like there will be with Apple Intelligence. It looks to me like Apple and Google will be able to do substantially the same things with their respective generative AI integrations, so people will buy based on the other qualities of the phone and their need for privacy.
Thanks Matt. I like your pragmatic actitud. You don’t believe too much bla bla… if it works on the real work scenario is good, otherwise it is just nice words.
Google has lost my trust when it comes to AI releases. I'll believe it when I see it actually working in the field.
I like this format! Many thanks! 👍
OpenAI is very capable and ahead of most models...not to mention newer models they haven't released yet. However, it seems strange to critique EVERY small issue rather than does the models actually assist most people IN REAL LIFE SITUATIONS vs another model..... For FREE or less than OpenAI. The killers question and marble problem is fun but not really useful to most of us.
At this point I am covenced that Matt is doing it in purpose
I really like the idea of on device compute. It really saves Google's resources. It's draining our batteries and only has to send text of X kilobytes to their LLM systems to get absolutely everything they want.
Sounds great.
They are chasing a multimodal nirvana with very weak LLM capabilities. Honestly I don't know how they let go of the incredible lead they had with AlphaGo
They could have finetuned RoBERTa as a tools router/prompt modifier for google home/LLM, this is going to be a hot mess
News about Gemini is always disappointing. This model lags behind other open-source or cutting-edge models, except for its 1M context window, otherwise it barely competes yet. But you're right, they have the ecosystem alright.
Im lowkey excited to have gemini nano in chrome soon. It will unlock so many nice ux improvements
Auto Generate tags
Auto select categories
filter queries from prompts.
Auto configure the layout / app based on the data with schema generation
Mocking fake api calls
All without a roundtrip to the server
When are they gonna learn to not rush the release of unfinished products?
"Hey google" on my pixel is pretty awful.. Need to say Hey google 5 times before my phone works.. Usually "I didn't understand your question".. and it can't do more or less anything on my phone... Only thing I actually use it for is to set timers when I'm cooking.. and even that is usually frustrating.. If I need to "talk" to an AI to brainstorm ideas or something like that, I go to the trouble of starting the ChatGPT app conversation mode, and the difference is night and day.. that one actually understands me... If chatgpt could control and read pieces of my phone, that'd be amazing... Google needs to step up their quality!
My opinion when it comes to talking to ai is to always remember it’s a computer, don’t be polite, don’t ask how it is, don’t say please, don’t say thanks… just ask the q get answer and move on, otherwise we are headed in to a bad place.. when you go to the bathroom do you ask the toilet paper how it is before you wipe?? Do you thank it?? also I don’t want the voices to sound real id rather more robotic
I wonder whether "the other device" was a fall back in case anything goes wrong where they had made sure it would work in one way or another?
It is weird. Their benchmark says one thing, but the real-world testing says otherwise. I also have not been impressed, but I am willing to give the experimental version a try.
This triggered "OK google" on my phone
Open AI is my favorite
The point is the latency is clearly identifiable in terms of natural language processing, DST N dpL is not hyp er reactive and eventually causing this dialogue, what makes the things worse is that the prompter is clearly reciting this previously rehearsaled conversation with the mother which makes its latency more evidental.
Interesting that you rhought it spunded less real, i thought the opposit, my first thought was: ohh this is the nex gen of live chat realism, much less digital and scratchy sounding.
Great video man, thank you!
Interesting video! The downncast tone in delivery throughout belied the fact that you seem to like these additions more than not. The downcast tone of voice throughout leaves me with an overall feeling of negativity about Gemini.
I'm really looking forward to Google getting some of this stuff working well, and then discontinuing it.
The LLMSYS is not "that doesn't mean anything" territory cos it's not one of those benchmarks that can be gamed like MMLU where the benchmark might have got into the training data. Ordinary users are asking a wide range of questions and then voting blindly. That's the kind of benchmark that google is topping right now.
I know you have your own benchmarks but they don't cover everyone's usecase. Maybe what happened is that it's working well for most everyone else use case.
Topping LLMSYS arena chatbot cannot be "that doesn't mean anything". If it's that easy let others make models that top that crowd sourced benchmark.
I am sorry but Google's past ecosystem was really not very symbiotic. A lot of Ai features in these apps are just conveniently placed chatbots, that help you to write or search. To me, Google Assistant features are still the most needed and used.....ie. looking up the weather, traffic, automation, reminders, calling etc. Hopefully Gemini 🙄might actually make a difference.
Being number one in the chatbot arena leaderboard does not mean anything? That is, imo, more meaningful than the results obtained in one person's custom test, and it does reflect real case uses, as the chatbot leaderboard works with all sorts of possible prompts, and their results are blindly evaluated.
I would love you to elaborate on why you think the chatbot arena leaderboard does not mean anything
I turned Gemini on on my Pixel 8 Pro and was very unimpressed. I too use Google Assistant to work with Home, and it was a disaster. I went back to Google Assistant. For my purposes, Assistant is way more useful.
UPDATE: I wanted to try Gemini Live... but I have to disable Assistant to do so. Why the frack won't they let me choose which one I use in the moment? So frustrating