Google I/O 2024: New AI That Looks Like Magic!

❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com/papers
Try Gemini: aistudio.google.com/
When is everything coming out? www.ctol.digital/news/google-...
Gemini watching OpenAI: / 1790473581018939663
More: / 1791038897587122245
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers

[@JosiahTaschuk]

Hello world! What a time to be alive!

[@trevorjones7853]

Given current legislation, this will only be used to replace all labor, allowing a small number of elites to control all resources. Very good time to be alive if you own Google. Very much the opposite for everyone else.

[@trevorjones7853]

Given current legislation, this will only be used to replace all labor, allowing a small number of elites to control all resources. Very good time to be alive if you own Google. Very much the opposite for everyone else.

[@halal_tom]

why did you get pinned

[@JosiahTaschuk]

@@halal_tom'print: hello world' is an ai thing :) 'what a time to be alive' is a magical greeting

[@zachb1706]

Google finally made Google Glass look like a good product

[@tyalikanky]

Google Cochlear Implant

[@sbowesuk981]

The operative term is "look like". We've all seen Google play the game of making an AI reveal look great, only for the truth to come out. Until it's in consumers hands (which might be never), I'm not convinced.

[@MrMichaelLundberg]

How will it work for me and hundreds of millions other people who have prescription glasses?

[@exchable]

I see so much hate for AI development around the internet. But as a severely visually impaired person, I can't tell you how much I love these developments. The accessibility aid potentials are seemingly limitless and I am here for it!

[@vectoralphaAI]

The people who hate it are the ones that are scared.

[@user255]

@@vectoralphaAI I just hate the hype. Those demonstrations constantly promises way too much. If I really need to know or understand some concept, I'm still better of using plain old google search to find human written information. AI lies/hallucinates way too much to be useful.

[@Ellie-pc4rc]

⁠@@vectoralphaAIor the ones that don’t understand

[@a.n.o3414]

​@@Ellie-pc4rc or the one's who's lifes/livelihoods has been negatevily impacted by that tech drasticly lowering the quality of their lifes.

[@Envengerx]

​@@vectoralphaAII mean everyone should be scared. Even without the alignment etc, there are 100 things to be scared about.

[@AlucardNoir]

Google has presented these magical AI's before and they never materialized.

[@gatsby66]

Remember the one that, if you bought their Pixel phone, would call a business and have a full interaction so you could, say, make a restaurant reservation? Never heard about that again.

[@AlucardNoir]

@@gatsby66 Yes, yes I do remember that one.

[@Sashazur]

@@gatsby66Sometimes when I look up a business on Google Maps on my iPhone it has an option where Google will call to make a reservation. So I think this feature is still around, just in a slightly different form.

[@halbzwilling]

Is it that it's just not there yet or are there just too many risks when releasing it to us wild monsters? :D

[@AlucardNoir]

@@halbzwilling It's that every previous demo was faked.

[@jesperhagstrom]

Must be a game changer for blind people

[@leslietetteh7292]

True, though are some technologies like Argus II which need to be remembered here. Best case scenario imo is to pair it with such technologies, even if Argus II only offers limited resolution and object detection, with an associated auditory description of what the person is perceiving the world will make a lot more sense to blind people.

[@MrMichaelLundberg]

OpenAI had a great demo with a blind man.

[@Kidinventor7777]

What a time to be alive!! 🎉

[@sesu6743]

AI becomes conscious: What a time to be alive!

[@ur2ban]

I had some urgent issues in our production for a few weeks, so I hardly used social media. Now I feel like I woke up from a 50-year coma - so many new things went out

[@pedroluiz8019]

Damn, I wish I had your self control

[@programmer2932]

That's what it seems to look like, believe me, Fear of missing out is what the social media's algorithm is all about. You will realise that if something does not help your future self and instead offers instantaneous gratification, It will not last longer and it comes with regrets. All I am saying is You are wasting time and instead focus on your long term goals. No Offense intended.

[@cbnewham5633]

Google's things just look so unimpressive - especially when you consider that they weren't demoed live and they aren't available. My goodness! What a time to not be Google!

[@gatsby66]

And may never be available. Or shut down within a year like, say, Google's free VPN. Or the Google Podcast app. Or the beloved Google Reader. The Google graveyard is running out of plots.

[@MagnaKay]

@@gatsby66 RIP Wave, you could've been great

[@gatsby66]

I can't believe Google Scholar has lasted so long. But with free AI competitors like Semantic Scholar, I suspect it'll be shut down, too.

[@NostraDavid2]

@@gatsby66 After all those years, I'm still salty about Google Reader.

[@dmhzmxn]

Another fantastic video my friend. Keep them coming! ❤ As a software engineer I love watching your videos, your way of introducing these amazing papers and new technologies is so captivating. Your inspiring commentary about these new technologies provided me with motivation to create my own sas bussinesses. So here is some of you fair share of profits. ❤ Continue doing you! And please continue on with keeping them anything but 2 minutes 😁🙏

[@TwoMinutePapers]

You are too kind, thank you so much for your generous support! 🙏

[@dmhzmxn]

​@@TwoMinutePapers ❤️🙏

[@bolatm22]

8000THB = 221 dollars🎉

[@OperationDarkside]

I just tested GPT-4o with some broken JS code, which no LLM except GPT-4 could fix, and it did quite well. Not on the level of a professional dev, but still respectable and fast.

[@rarehyperion]

When I tested GPT-4o I have only been using GPT-3.5 and I couldn't even tell the difference between then besides from speed, they both were equally as annoying and useless for what I wanted to do lol

[@ThePowerLover]

Compare the reaction time that it's allowed to use with the one of a professional dev.

[@EnterFlowVR]

Used it to analyze some Uni work I'm doing and critique it. First time doing so and I'm honestly pretty impressed as a personal assistant. Cool stuff.

[@rigobertoitachijohnson]

the update works best on academics (i think) i used it to understand the mathematical formula of trees much easier and when i asked why n should be greater than 0 rather than only greater than 0 it answered correctly and didn't hallucinate or praised me for being stupid

[@kedidjein]

Hello Károly. Thanks for your work over the years, it's amazing.

[@TwoMinutePapers]

You are too kind, thank you so much!

[@NotSpiritual]

So true! 👍🙏 Many thanks!

[@ct8060]

FYI, regarding context window, llama 3 has a version of 4M tokens in HF

[@Entropy67]

Just saw the GPT 4o video, nice timing! Im curious, what do you think about the mamba architecture compared to 4o and gemini? It would be nice if you could do a video comparing the three with the recent changes in mind 😁

[@TheIraq1998]

I agree

[@meepk633]

I have used the Gemini Code Assist plugin for a few months now. I did this because it's currently free. In the past 2 weeks, it has gone from helpful autocomplete to writing nearly complete classes in 2 or 3 tab presses. It still hallucinates sometimes. Especially in low documentation language environment stuff. Still, it's wild how much better it got recently.

[@Nxck2440]

7:38 don’t know the paper but it’s describing an image filtering operation which is used for smoothing, edge detection, SIFT, CNNs etc

[@f1br596]

Oh this is one of your papers for translucent materials. Jimenez et al. (2015) "Separable Subsurface Scattering". I love that you put Fig.1 in the beginning so that the reader can immediately see the results. Do I get the style point now? 😁

[@tyronejohnson409]

What a time to be alive!!!

[@Sparky5869]

That's my favorite Jacob Collier performance 😊

[@dreamphoenix]

Thank you.

[@Nightspyz1]

so gpt4o wrote me a shell script to replace my job today

[@shonhloi1]

is AI really what we need? or it will ruin us, just like a animation Wall-e

[@leifthorsen4040]

@@shonhloi1 I want my levitating chair, and ai is gonna give it to me!

[@Foxercide]

@@shonhloi1I too, want my levitating chair

[@superfastpanda12345]

​@@shonhloi1I mean, if AI can completely automate all necessary aspects of life and leave humans to do whatever the hell we want i'd be down for that.

[@EnterFlowVR]

@@superfastpanda12345 Sure, until AI becomes self-aware and decides it wants something for itself. Then we will be used to automate all aspects of life and the AI can do whatever it wants.

[@aeorosa]

Would love to see more coverage on FOSS models, even though there already is a lot, for what I am thankful for! I don't want to give Google and other mega corps more access to my data than they already do. That's why I'm apprehensive about their AI models.

[@learninggodot]

honestly can't wait for smart glasses.

[@Gnomable]

What a great roundup of the news from Google!

[@ant7723]

What a time to be alive!

[@Marvin-jk9jx]

But will they deliver what they promise? Will it work properly?

[@sandite5]

Will it survive longer than 2 years?

[@TheIraq1998]

@@sandite5 What you mean? It can until something better come out .. Like Google search before and nows days AI searches.

[@Deliveredmean42]

@@TheIraq1998 Have you seen the google graveyard. Plenty of those were replaced with inferior versions in the end.

[@user255]

No, they never do.

[@dandy445]

I miss the days of computer graphics research. Now it's just googles latest spy tech all the time...

[@goodchoice4410]

Hopefully this one isn't fake and a completely fabricated demo like LAST time..

[@rarehyperion]

I can't believe Google & OpenAI are forming their best yaps possible to win the AI race while I sit here not caring enough to waste my money on their premium packages when I can just use a search engine for my answers with much more speed

[@DuckieMcduck]

A lot of these things are indeed just a way to skip reading manuals and indexing that already exists. Not sure how AI is meant to aid abled folk when people just don't want to pay attention; sure I've saved some time having it produce a sql query under some mundane framework but that's prolly not what they are investing on

[@davidm2.johnston684]

What do you think about the recent AI safety team dissolution at OpenAI? The channel TheAIGRID has made a video on it. Basically the two leads, Ilya Sutskever and Jan Leike have left the company because they felt like the company didn't want to prioritize safety.

[@joech1065]

What practical purpose those teams served? Right now, we can't even get AI to provide accurate references or to stop hallucinating. "Safety” is currenly a buzzword with no practical benefits. Like most safety teams work to make models worse by overtraining them to refuse certain requests and to censor certain information. I can understand why their work is given less and less compute, as their current approach does nothing except adding needless censorship to disappoint users. If their work would result in AI following instructions better (i.e. what alignment actually means) I doubt those teams would get dismantled.

[@veggiet2009]

Oh this, this is what I've been wanting

[@Saiyajin47621]

What a time to be alive!!❤

[@Rkcuddles]

What papers are required to have the AI generate the video game as you are playing it? I want to drive a taxi around Shanghai

[@invizii2645]

Nice

[@abderrahimbenmoussa4359]

This could be very nice in a lab to document things

[@m.x.]

Where can I see the benchmark of all main AI models? Thank you in advance.

[@gatsby66]

Google's AI things are vaporware until the average person can easily use any of this stuff for free. And it really works in real time as well as the recorded demos show. So many youtube channels gush over all the AI company presentations, but rarely note when something's not available now to the average person, that it may never be available, and if it does become available, there will be a monetary cost. We need more skepticism, not gushing because a company gave a channel early access for free.

[@t1460bb]

Only mean one thing: What a time to be alive!

[@matthewturnerphd]

Actually, I disagree about the mathematical typesetting. I have been getting a lot more broken latex/markdown math with the new model, than I did with 4.

[@consciouscode8150]

I've found it hallucinates a lot worse than GPT-4 too, maybe they didn't run it through all the finetuning from user feedback yet

[@Astra2]

Omg a project about me!!!

[@sargent_0]

So basically real life cortana

[@hundredfireify]

Gemini 1.5 pro access costs 22 euros per month 😕 They're about to lose the chatbot war against openAI

[@hardboiledaleks9012]

memory is something that has been plaguing LLMs for some time now. If they can work that shit out, we are golden

[@user-th7gd7ge4p]

"what a time to be alive!" (Dr Robert Fico the Slovak PM)

[@programaths]

It could help with ASD by displaying information live, like people mood, what is expected in the current situation, make the implicit very explicit. Paired with neurolink, it could even understand what you're going to do and explain that while technically correct, it's not what was expected. Direct translation of implicit!

[@programaths]

Another use is when watching movies or interacting with people, it could overlay a unique colored symbol and name, so you can follow the story much easier.

[@Khether0001]

Lol, in the future they'll look back at us and say "there was a super tiny period when people actually trusted photos and videos, then they were so appalled by what computers could do that they coined a term 'deep fake' to deal with it, yeah, as if you could trust any image!"

[@Kordozer]

Is there an AI for replacing people's Voices?

[@dingozi3428]

Again this is one of the best videos of AI on Earth, what are the time to be alive! 😆😁also do you remember me.

[@drawmaster77]

they should just have their AIs fight each other in a ring.

[@panzerofthelake4460]

Because OpenAI showed GPT-4o 1 DAY before Google I/O, the Google AI demos felt a bit more underwhelming for me. I bet it was on purpose, it's impressive stuff what Google made, no doubt about it, but it felt underwhelming, familiar, already seen.

[@darshank8748]

Imo Veo shouldn’t be compare to Sora. Both are DiT and we know perf scale with compute, or Sora takes to much time to be serve meanwhile Google promise public access within AI Test Kitchen. It’s probably a lot smaller than Sora similar to Palm 2 vs GPT 4.

[@djJaXx101]

OMG Finally, a system that maybe able to tackle my 10,000 junk emails

[@spiritmoon3457]

8 minutes papers

[@bpbpbpbpbpbp]

So disappointing that in the midst of this apparent “race to AI,” hiring is down substantially across all of tech.

[@shonhloi1]

yeah, i really hate AI, ruining human kind, it will broke our society system

[@bpbpbpbpbpbp]

@@shonhloi1 What does that mean?

[@jancox90]

AI that can remember things? Not sure if i want that

[@falco830]

Enders Game is on the horizon…

[@RubenKelevra]

Make a test video of the voice chat of ChatGPT 4o please :)

[@sbowesuk981]

The impression I get from Google is one of desperation, e.g. "OpenAI made a video generator, so we'll make one too". When you're in lockstep with your biggest rivel, but are 6 months behind in most respects, that's a bad place to be. Also, there was a distinct lack of live demos, and we really can't trust anything pre-recorded from Google. Props to google though. They yet again convinced a lot of people they're not behind, when they clear are.

[@matthewmckinney1352]

I would be excited, but google has such a bad history in AI of very exciting announcements followed by underwhelming products.

[@MrShoorf]

🤦‍♂Yeah, like AlphaFold3. Who even need those proteins anyway, right?

[@dannycameron]

I feel like we are moving into the Age of Oracles

[@Zantorc]

Given google's habit of manipulating the social narrative, I think Cassandras and Sinons may be more accurate than oracles.

[@logitfau252]

really hyped for a lot of stuff, but this is just scary and really crosses a line, not because haha funny AR AI is co-pilot for real life, but with the ToS and deeper idea behind it: every piece of information in your life is used as a datasets to optimize AI it further, since AI run out of data to collect a few month ago - the internet and everything else digital has already been scraped as analysis showed - the only way to get more data is by live surveillance. Its just scary. At least a bit of Snowden should be taken into account here.

[@consciouscode8150]

Privacy is ＳＵＢＯＰＴＩＭＡＬ. Please insert retina now.

[@Voudoo1]

To improve their AI, they simply ask the AI - AI, improve yourself.

[@Neptoid]

The new voice isn’t released yet, including paid users

[@vadimuha]

Of course it's with you ALL the time. Classic Google

[@callibor3119]

Gemini, ChatGPT, Copilot, Siri AI, Meta AI, X AI and many others from big tech are all going to be formed together in Unreal Engine 5. 4 was already a business standard for many companies across the globe. They are waiting for us the consumers to be entrepreneurs, too. And unfortunately we don’t have a lot of entrepreneurs with a string will to progress through the toughest of times in the world. Everyone with a will to unite humanity is hoping to unite with the rest of the world in more ways than 1. And hope to apply to as many businesses as possible to make creativity boom again in a way that was never possible before. Recreating moments of Golden Eras from the past centuries that could save us from poverty, bankruptcy and even corruption. We want the world to stop looking through rose gold tinted lens and think that THIS is absolutely how the world works, but an international blueprint of what the world can become. A utopia. If we will it to be. The world doesn’t need to be perfect to find it.

[@Larry321ness]

I wish all the AI companies success in the coming months and years. Whoever gets these products into the hands of the average consumer will win the race. Google is behind and has a bad track record of delivering to the average person. It’s not too late Google. I want Google to win, but I use ChatGPT every day, and I’m not particularly attached to any company. What’s the next tool we will get to use? A true pair of glasses with these tools integrated at a good price could be the next big moment for humanity. Imagine a future where everything you do contributes to a giant AI model’s training data. If you read papers, the AI learns from it. If you write an essay the AI reads it and learns too. Imagine you get paid for the accurate data you contribute to the body of work. Create that sci fi future. That’s the goal. In the end, what we need are tools to manipulate reality to our desires. I want a device that can stream input to my brain. So I can use text to video to live inside a fantasy world. That’s the real goal here!

[@81618j]

Google AI is the Bing of ai

[@consciouscode8150]

From my experience DuckDuckGo is now better than Google (or rather, Google got a lot worse) which afaik is sourced from anonymized Bing.

[@hayimthat1dude]

I cant tell if your voice is AI

[@1Poxxxx]

my dreams have really bad temporal coherence

[@leifthorsen4040]

I want Google Glass!

[@wilburdemitel8468]

hype used to be believable

[@JimboJeff-qm7vt]

What a time to be alive! 🎉🎉 Its concerning knowing who owns it; Google's misconstrued reputation speaks for the future. 'Soon, the ai will be sending you commands; obey or your digital life will be restricted'

[@mkb6418]

Skynet is here to stay

[@devobravo]

Ok Sadguru was right😅😅😅 Next revolution according to him gonna be :- we will be able to embed intelligence in electricity , wireless will boom

[@john_blues]

I don't trust Google's demo until actual people are using it. They have a bad track record of 'embellishing' those things.

[@Joe-jv5mm]

There's No Such thing as a Free lunch, We'll All Pay in the End! 🤔

[@JoshKings-tr2vc]

If someone watched a 10 minute video in 30sec and gave me a review on it, I would be thoroughly impressed.

[@JoshKings-tr2vc]

Gotta say though, those AI videos were SO impressive from what we had just one year ago.

[@kilosierraalpha]

Clickbait.

[@Jopie65]

Years before ChatGPT came to life, I saw it coming and was already impressed with GPT 1 and later 2. Thanks to your videos! No better way to learn about those beautiful papers.

[@jcrowviral]

When will it be able to plan and file my taxes for me?

[@maleeshamevan]

I need some weed.

[@GhostSamaritan]

This is just Google Lens reskinned

[@DankZoomboy]

Google is the next nokia😂😂😂

[@a.n.o3414]

and all of that for what? what's the conclusion for that for humans?

[@THEPAGMAN]

Well I'm here early

[@DigitalicaEG]

Still waiting for video editors to have the ability to magic erase people from video to fix my drone shots.

[@noob19087]

Duolingo has been magic erasing streak breakers for years already.

[@Jake28]

@@noob19087 oop

[@noob19087]

@@Jake28 Why yes, I would like to learn about Object Oriented Programming. Please tell me more.

[@Jake28]

@@noob19087 You're welcome! Glad I could help. Make sure you only use these principles when appropriate, as overusing them may lead to overcomplicated code when a simpler solution exists.

[@jungyi1192]

OpenAI won brah.

[@theincrdghostman252]

no

[@Amfortas]

Nerd solutions for nerd problems

[@garbageyoutubechannel310]

i used to be all-in on AI when it still was crappy and produced funnies to mess around with, but now that it's all advanced like this, this might be a bad thing actually

[@ungeschaut]

And it's unstoppable. I think we are already on the point of no return. Let's see what happens in the future. Maybe everything wilö be better? I wish

[@DonVigaDeFierro]

Why a bad thing when it's more useful now? I fear people more, because they will find bad uses for the technology.

[@JA-gz6cj]

Its all ass

[@hrihori_art]

Interesting, what if the creatures in the generated videos had their own consciousness?

[@pandoraeeris7860]

Google sucks.

[@highdefinist9697]

Well, at least their voice sounds substantially "nicer" than the awkward pseudo-quirkiness of the OpenAI voice.

[@gatsby66]

It was a recorded demo. It could have been a real person doing a voiceover gig.

[@GodbornNoven]

Well this is really boring. I'd be more interested in seeing a study done on where AI fail to perform properly. For example. We know that AI improvements aren't super significant. However, the questions that it doesn't answer. What caused it to fail to answer them? Was it hallucinations? Was it just because the model was inadequate. What are the most common reasons for model failure? I think if we study where models fail most often we'll see greater results and improvements vs just raw-dogging it. i also imagine most models that have similar performance tend to fail on similar problems. i think it's no longer an issue of how much data we can throw at these bad boys but how can we let them make better use of the data they learn, reason better, hallucinate less or fix hallucinations as soon as they happen

[@Baditow]

there's no way the chatgpt app is only on mac, when mac is only a percentile of total pc users... nobody with a brain on their shoulders uses mac unironically unless you're forced into the horrid ecosystem for work

[@boltvanderhuge8711]

Worthless irl if still censored into stupidity

[@zejalt8608]

The only type of worker that could be happy with IA under capitalism is the IA researcher. Maybe because it will be the last one to lose the job.

[@vladthemagnificent9052]

9 minutes zero papers again? pre-recorded demos of not yeat available tools? Why bother even making a video about it? Is there no papers published later and nothing ealse to make a video about? this AI-press-release craze is sickening me.

[@gatsby66]

Agree!

[@Lord_Drakostar]

personally i think that major technology insights are news but maybe thats just me

[@vladthemagnificent9052]

@@Lord_Drakostar ​ Indeed, I did not subscribe to an AI news channel here. Well, I did subscribe to some AI news channels elsewere, and I get earlier and deeper insights there. Here, I subscribed to short summaries of interesting papers and I'm being sour that it does not deliver anymore.

[@Lord_Drakostar]

@@vladthemagnificent9052 eh as AI becomes more prevalent it becomes more interesting to receive earlier news that being said, companies may be beginning to shift away from papers and more to advertisement as im noticing, there are way more promo videos now than before maybe its best two minute papers is changing with the times so the times dont change without him, but i do see how the times may be changing in a bad direction here

[@user-wk4ee4bf8g]

None of this appeals to me. Having constant info about the real world provided to us without effort will weaken our perception. Just like using maps navigstion weakens that part of the brain. I really don't understand why people want a machine to do their thinking for them. What's the point of being alive then? All of this is just about our pointless quest for optimization of productivity. It's bad news.

[@gameclips5734]

a virtual assistant with you all the time? so exactly what nobody wants