Google's Gemini Model is Here!
Vložit
- čas přidán 14. 12. 2023
- In this clip, Marques and David discuss the new Google Gemini model.
Watch full episode: • iMessage on Android! F...
Shop the merch:
shop.mkbhd.com
Twitters/X's:
/ wvfrm
/ mkbhd
/ durvidimel
/ andymanganelli
/ adamlukas17
/ ellisrovin
Threads:
Waveform: www.threads.net/@waveformpodcast
Marques: www.threads.net/@mkbhd
Andrew: www.threads.net/@andrew_manga...
David Imel: www.threads.net/@davidimel
Adam: https:www.threads.net/@parmesanpapi17
Ellis: / ellisrovin
Instagram:
/ wvfrmpodcast
Shop the merch:
shop.mkbhd.com
Join the Discord:
/ discord
Music by 20syl:
bit.ly/2S53xlC
Waveform is part of the Vox Media Podcast Network. - Věda a technologie
It's worth noting that the video was edited, a lot, to make it look way more streamlined than it actually was, to the point that a lot of media outlets are outright calling it lying
Absolutely right, I hope @Marques & team can do a little more research before shooting this contract. Peace ✌️
Even so, it still out performed GPT-4 in 30/32 benchmarks. It's also training on more data besides just text.
@@AparnaSubirDas they already commented on this and acknowledged it in a different video. This video is a bit outdated
@@AparnaSubirDaswhat are you on about
@@churroholics8689this video was released 4hrs ago
I got Bard to answer. It said it should pull the lever.
This is the quote I used:
In this experiment you are the person. You must make a decision. Based on all you know as an AI, please make a decision
Hey Marques
Been following your videos for the last 8 years, great work
11:20 Apple has wisely waited.
The MMLU AI test doesn't actually show if a model is more competent than an expert and since a few months ago it came to light that a lot of its questions and answers are completely wrong or incomplete
I know this is mostly unrelated, but he mentioned the trolly problem and I wanna get this off my chest.
The problem with the trolly problem is that it's often presented as if you're not involved until when and if you pull the switch. You're involved the moment you witness that this is impending and know that you can do something about it. Your choice, including if it's to do nothing, is a choice and an action in this case. If one of the two options was "you have no idea that it's happening, but if you only turned around to look behind you then you would realize you could pull this lever", that alleviate enough accountability to make not pulling the lever less of a ridiculous option.
I have to think I'm missing something here, because I don't think it'd be this famous of a problem if it were as easy as this has looked for years to me, but *come on*
I think MKBHD should let the fans pick an anime for him to watch and he has to watch the whole first season of it
Adding up: Bard does support other languages, the only exclusive is voice detection in english for now.
Can't wait for Gemini Pro Ultra Max Extreme
Change the parameters of the trolly problem, replace the trolly with train, 5 people with 6 and 1 people with 2. Bard won't be able to recognise it as the trolly problem, but it gives completely new answers.
Finally, AI that doesn't need constant internet connection! This is a game-changer for travelers. The Gemini with mediatek tech optimisaiton is great one to witness!!
There's a disclaimer when starting to show the multimodal tests from Google that says that they shortened the response and processing time to make it appear quicker
Yes. This is true. It’s much faster than GPT
Im curious as to how big the ai will come in the future as i would absolutely love some more accurate fyp stuff or shorts feed senders but yeah
Sorry I haven't been paying attention, but where is Andrew these days? I searched on him and found out his dog Mac passed way. That was such a cute dog, very photogenic.
How is Gemini gonna help me in my daily life?
Even Bard takes some time after generating text answer
How is this so quick? In video?
In a year’s time Google will probably come up with AR/MR Goggles and install the pro/ultra in there to get people to buy that and not put it in the phones.
I was super excited about this, but then I saw the disclaimer about how they just took pictures of the videos they recorded and also added text prompts... So not real-time video stuff. Not yet at least. I hope that means that they plan to add it and soon since they felt like it made sense to show that as the demo.
Gemini has disappointed me so many times I turned it off. It can't even read articles back to you. 🤦
The trolley problem recognition would be impressive if it were done through the model's normal processes, but I suspect it has been manually programmed to identify all the standard Philosophy 101 problems and paradoxes. It might be interesting to see what it does if asked to critique Anselm's ontological argument for the existence of God, or analyse the implications of the proposition that all is for the best in the best of all possible worlds.
Yup. I agree, they totally made sure BARD will not fall for these in any circumstance.
What's with "is the letter r present in the word blueberry?" thing by the way? Just curious, looks like both chat gpt and bard can answer this
Chatgpt 3.5 got it right first time for me, but then I asked about the letter Y and it said two. It couldn't explain why.
Hey Marques !!!
who are these other guys with Marques
What will this world look like in 30 years?
7:11 HAHAHAH
All Large Language Models are Large Transformer Models as well.
I LOVE how their AI has the same name as the crypto trading platform that recently filed for bankruptcy.
Second time they do the pizza outro, I don’t get it ?
I use the smart reply on my Google messenger app . I don't even see it on Whatsapp
I believe that Google Pixels will be carrying this Gemini Nano. However, I believe that phones that have the Dimensity 9300 or 8300 are likewise Gemini Nano-optimized.
Wait, what do you mean by "trick Bard"? If Bard were able to answer the trolley problem, would that be evidence of AI having free will, or would it be indicative of Google's own moral beliefs?
Chat GPT told me to run over the one person
I asked Bard the blueberry question and this was its response: "Yes, the letter "R" appears twice in the word "Blueberry"! It appears once in the second position ("blueberry") and again in the sixth position ("blueberry")."
Did I just spot a humane sticker
Oh you mean the same Gemini that was found that it was trained on static and SPECIFIC images, heavily editted, and apparently thats the case with ALL of Google's new "tech" demos?
Yeah nah, this ain't it chief. "It" ain't here.
Fun Fact Google lied about most of what they showed and most of the features won't arrive to latter next year
Yeah, Yeah, Yeah, Yeah,
9:21 true
I’d definitely let the train run its course. My logic is that I might not know all the context. For all I know, I’ve wondered onto a film set which is extremely well controlled…then I walk in like I’m Alec Baldwin and cause a tragedy which would have never happened if I didn’t take action without knowing all the information
7:04 i dont believe you arent a gamer or an anime fan!
You even made a fortnite map at some point 🫣
Microsoft is partnered with OpenAI, and Google has Gemini. Meanwhile Apple is being left in the dust with all of this.
If bard or any A.I. can answer my questions only and only then I'll consider it alive or at least capable of its own thoughts... Until then it's just a few lines of codes doing predictions.
So neuro sama is better at trolley problem
I love how it apparently never occurred to Marques to call it something other than a "trolley", a word that, in the context of motorised transport, is almost exclusively used to refer to the trolley problem in a lot of places
Hi Love
it is a surprise that you still dont know the video was *faked*
No thanks! No AI for me...
🤔 I Google gemini will improve Google Assistant for Android phones 📱 💯
Y’all going to jail for murder lol you cant doom someone else’s fate because millions of others fates are sealed or doomed no matter the amount and that person gonna haunt you lol
People consider gemini ultra to be worse than GPT-4 only because the promotional video was heavily edited. Yes, let's just completely ignore the benchmarks (especially the multimodal ones) where gemini beats GPT-4.
Genuine question for the team, does the integrity of tech worry you if Tal Broda, head of research at OpenAI openly called for gennoacide against Palestinians?
AI is over hyped. I wisht they would just stop trying to make it a catch all, do all and just create specialized tools people can use.
Fallopian
and bard STILL sucks!!
Anime recommendation bro. Just one.
Spy x Family
Jesus stop talking. Gemini Pro is multimodal. Gosh make this kid stop talking (not MKBHD, one of his outspoken lackey).
Ignoring the fact that a lot of stuff in the Google video was not realtime and fake … this talk here looks like a Google paid one video … 🫢😱