This is changing the way scientists research | Gemini

Gemini - Google’s newest and most capable AI model.
200,000 scientific papers with crucial science information gets read, understood, and filtered by Gemini all in one lunch break. Join DeepMind Research Scientist Sebastian Nowozin and Software Engineer Taylor Applebaum as they explore the AI capabilities of Gemini.
Check out more Gemini demos: goo.gle/4164rNO
Find out more about Gemini: deepmind.googl...
Read the blog post: goo.gle/3uRyug7
Subscribe to our Channel: / google
Tweet with us on X: / google
Follow us on Instagram: / google
Join us on Facebook: / google

[@daintellekt]

A game changer, OpenAI has a worthy competitor now.

[@ShpanMan]

Bout damn time haha. But seriously, this is an incredible feat, done in a short amount of time. And remember, the next feat will take even less time. The Singularity is near.

[@1fattyfatman]

next year ; )

[@sashank224]

Their battle will be legendary.

[@Dullydude]

OpenAI has had GPT-4 for over a year and Google's brand new model barely defeats it? Yeah, I don't think they're even in the same ballpark lol

[@eksot]

@@Dullydude For over a year? GPT-4 was released this year in march... you probably mean GPT 3.5.

[@reza2kn]

I HAVE DONE THIS! I've done this exact thing, multiple times, and it usually takes more than a YEAR, when a TEAM of people are working on it! Now it's a lunch break?! Why couldn't this be around when I was busting my hump as an academic researcher?!

[@MathAtFA]

Can you publish your chat? I assume you used Gemini Pro via Bard?

[@reza2kn]

@@MathAtFA I was talking about this video, and the fact that I have done this type of research in my team manually, and that it usually takes more than a year

[@matissparadise5799]

Bro chill hehe, we're all in shock welcome to the party hahaha

[@reza2kn]

@@matissparadise5799 welcome to the party? 😂😂 You realize your ass is 8 months late to this comment that was made on the first day this go out, right? 😂😂😂 Even funnier is that this whole ad was just a bunch of lies. 😜😂😂 Didn't pick a good one man.

[@SaschaPallenberg]

Gemini's reasoning capabilities are absolutely mind blowing

[@Google]

Can’t wait to see everyone put them to use!

[@Feel_theagi]

Now I am become death destroyer of jobs @@Google

[@BlakeTedKord]

​@@Googlecan u pls rls a Google version of an AR based measurement app? Idk why they took that away. Now I have to use an iPhone to measure something bcuz an iPhone is the only one now with a decent AR style measurement app.

[@tomrigodanzo]

​@@BlakeTedKordjust ask bard to measure it

[@JuanGtz-fq5ii]

​@@Google0:14

[@dkursada]

I study biotechnology, hoping to get MSc degree this month. Last year, I wrote a paper about a functional food and its possible effect in gut microbiota, so I was involved with the genomics of a complex microorganism community for the whole process. Literature scanning is always an extensive endeavor in genetics. We hit dead ends, find a lot of irrelevant information or tiredness can catch up to us. It seems like Gemini is not only a time saver, it can enable scientists to pull incredible feats in a heartbeat. I quickly realized the AI boom as a revolution in science, but the AI models' tendency to hallucinate always kept me in a conservative approach about the latest advancements in LLMs. A model with advanced reasoning capabilities sounds incredible. And the most incredible part of this is that advanced models like Gemini can help us mitigate the long standing replication crisis in science. Replicating old studies with new information seems doable now.

[@nadeemshaikh7863]

I'm sorry, how can models like Gemini help mitigate the replication crisis?

[@davidgriffin79]

As someone who obtained a PhD through a thesis written from research; I think it's important to be mindful of any results returned via A.I. Remember, A.I. works on probabilities, which is why a lot of "hallucination" takes place. A.I. does not "reason" as humans do and in many instances the only way to truly check the outcome on A.I. query is to compare it to an effective "closed form solution" query (i.e. you already know the correct answer, which defeats the object of the exercise); I've done this in mathematics, and whilst A.I can quite often give the correct answer to a complex integral given to it explicitly, it can also give a nonsense answer to a much simpler mathematical question which requires an initial deduction phase to form the question. This is what separates us from A.I at the moment, we have the ability to abstract to a meaningful question based on our understanding of the world.

[@emmajean719]

Literature scanning is also why you know your field after a Ms and a PhD. If you do it automatically, you won't read all those papers, and you will not have a good understanding of the state of your field.

[@RonInDune]

My biggest worry about using Gemini for this purpose is that, at scale, it will be very difficult to ensure all the results it obtains is legitimate and not a LLM hallucination. In fact it is this problem that has limited my usage of Bard for literature review so far.

[@h.c4898]

Bard@ doesn't use the same dataset. It crawls the internet and extracts the top results and make a summary of them. Its not perfect. Sometimes it gives very simplistic summary which is weak at times. You need to be more specific in your request then you get the result you need. But it's a great conversationalist however. Better thsn GPT imo.

[@Colorado_Chris]

Correct me if I am wrong, but by identifying the location of the information, this would avoid the LLM hallucinations?

[@SithLolita]

@@Colorado_Chris Unless it makes up the location, too.

[@deep.space.12]

@@Colorado_ChrisAlso possible if it misses some relevant papers entirely. No need to hallucinate to make mistakes.

[@michaelernesto299]

@@deep.space.12 A fair point, but it seems the question to ask is , would Gemini make any more mistakes than humans scanning 200K documents?

[@kishanbhadoriya8701]

I am a computer science student, This excites me so much, I wish I get to work on this project, I am happy that I am born at the perfect time to witness AI revolution, I am a total novice to the field though lol

[@Google]

The future looks bright 😎

[@gbyourss]

Man you're gonna put a lot of programmers out of work with this. I'm a programmer and I'm scared, not excited!

[@abdullahmeo]

Embrace the future! First, it seems highly likely that old school code is going to power a lot of application that the AI uses to complete certain tasks. By providing a better interface it means software can do more and will increase demand. And, it's important to keep in mind tech works in cycles. A little while ago we had the internet of things hype then we had the crypto hype. It is very unlikely we are on the cusp of creating something so intelligent that humans having nothing to add. Next, there is so much to explore with this model, like there are so many more things we can automate and there are no use cases that nobody has likely thought about yet. So, as a builder I think you are pretty well positioned to shift your skill set to take advantage of this shift.

[@radulaski]

@@Google is this the AI speaking?

[@Nick_works]

I'm a CS student specializing in AI, this is all so cool and incredibly impressive but also a little scary! I hope there are still some areas of research and development still available by the time I graduate haha

[@phrenologisto]

The vast majority of the matter in your world is invisble to our senses. Imagine this as a new sense, available to all humans, and you'll realize that it doesn't limit anything, it expands every field

[@AMFOSTUDIO]

2:42 It's the learners strategy and attitudes, thanks for such feelings.❤

[@the_gobbo]

What about the hallucinations? What verifies that it didn't made up the data from the papers?

[@oscarmvl]

Exactly what I was thinking, nothing guarantees that a relevant paper wasn’t missed or that key data wasn’t incorrectly extracted. I mean, it’s great tech, but it seems like they are overselling it in this video.

[@AlexBlack-xz8hp]

I think the fact that it can site where the data came from makes this more reliable. But certainly still an issue.

[@wishesandfishes]

A human's still gonna have to check item by item - we're many years away from being able to trust these models blindly.

[@clamr6122]

This is absolutely unbelievable. What a day to be a human!

[@Google]

What a day indeed! Thanks for joining us on the launch of Bard with Gemini Pro 🙌

[@Levelord92]

what a day to be a human and became useless exactly right now

[@AMFOSTUDIO]

2:42 It's the days to get relaxation in human work, right?

[@kwamesaforo4527]

Can you please share the links to the Colab notebooks

[@JOHN.Z999]

Now, I am very excited about this, I've always wanted an AI capable of helping me find the best scientific articles quickly and efficiently. I will use Bard Gemini extensively to create my academic articles more quickly; this is an extremely positive advancement for scientists, as we can now leave the heavy lifting to the AI while we think about better things.

[@Google]

We love to hear it! ❤

[@user-oc9qr3uk5e]

iam using gemini for publishing my science articles in magazines and its working incredible so far. No one can spot the difference. thank you guys@@Google

[@johnwilson7680]

Google has put out a lot of new videos on Gemini today. In my opinion this one is the most exciting and clearly demonstrates how valuable it already is.

[@kaio0777]

Gemini reasoning and capabilities are pretty amazing.

[@freedom_aint_free]

I don't know if I understood it correctly, but the crux of the problem with scientific papers research is to jump through their references and cross examine the claims of those: Imagine a system that can sort through every single paper ever published and generate a graph of the relationship between those, and discriminate by claims, subjects, etc. and them cross-examine them in order to assess the likelihood of those claims to be valid or to be refuted.

[@samuelburke9381]

interesting? What makes you say that?

[@ronsaad8416]

This is incredible! Is it possible to publish the notebook? Would love to try this myself.

[@ryanweaver972]

Very exciting. Any chance of sharing the colab notebook or code for this?

[@itsMohak]

A very good example of using Gemini for great betterment. AI is not only fun to use but also very intelligent. Very nice, Google.

[@brunesi]

This looks very promising. What would be amazing though is to develop a communication standard for scientific publications. We have had successful world wide consortiums targeting standards for products. Let's do that for science.

[@karthage3637]

What a time to be alive !

[@spin2nano]

with two minutes papers with Károly Zsolnai-Fehér

[@eshansingh1719]

I can use Gemini in Bard, but I have no clue where these guys are running Gemini to deploy Gemini for scientific literature curation. Please provide information on this. Very help ful tool

[@JR-dd3pj]

That looks amazing, is the code you're showing available? What about the resulting updated data set?

[@SlackWi]

I'm a geneticist working on exactly this data type and doing exactly this all day, I cannot believe how different my job is going to be soon.

[@jamiewent9223]

Soon you will no longer be needed. Not only can it do the trawling through data but it can be trsined to better at your job.

[@SlackWi]

@@jamiewent9223 I assure you there is significantly more to my job then this, stuff AI can’t replace, but sure it’s going to be a great tool we use all the time

[@andrianlee1329]

Can you share the Google Colab link for this demo, we have the exact same project working on.

[@soaresrobertoj]

Gemini is the kind of answer I was waiting for. Congrats! I'm really excited with AI prospects.

[@user-ll5se9rn4q]

I'm researcher in the field of genomics. I am eager to test the potential of Google's Bard with Gemini Pro & Ultra for cutting-edge genomics research. Currently, access to the model is limited. Any suggestions are greatly appreciated! Let's unlock the potential of AI for groundbreaking discoveries together! #genomics #ai #research #bard #geminipro #geminiultra #google

[@Pingu_astrocat21]

love being in this era !!!!!! learning this stuff up :)

[@History_Mystery_Crime]

The advancement of AI can’t be comprehended by the average human….what a pace of development….what a time to live ❤

[@ztunedd]

They probably put you at gunpoint to make this comment. :P

[@Tansenstudyy]

​@@ztunedd😂

[@CodeTechReviews]

Please share more details on how we can use the Ultra model for day to day tasks.

[@Ai-iann]

I’m excited to see what I’ll create with Gemini too. The great minds out there are already in plotting how they’ll blow our minds with this, considering how wild AI applications were on the release of CharGPT.

[@SergioAlbertoRomero]

Me: Reading dozens of articles and writing my dissertation over 3 years Bard: Read 200,000 articles over lunch, extracted data and presented in a digestible format...

[@Custodian123]

Academia is going to have a stroke over this 😂. That said, if it was not clear before, it's clear now... Academia must change, it won't have a choice. The best thing that can be done, is to separate intelligence from various work. The work doesn't make intelligence, because the intelligence will just transfer to however the work is done.

[@xuanyang2975]

where can we find the colab code

[@Procrastinus]

Meta-analyses are the foundation of science, basically an aggregator that allows us to see beyond the noise of individual studies which are filled with bias and error. You use it to answer questions. What this is showing us is that we can convert the sum of science on almost any topic into real time answers. In other words, this is a short hop away from an exocortex, an external decision-making system that guides what we do as we live. Human nature just changed. I'm staggered.

[@hypeventure]

Gemini... I hope it's free for me, I honestly need this for composing music and since Bard came to my life, making songs is now getting easier but also better since I can keep myself creative rather than solving a problem.

[@phrenologisto]

I'm interested in how Gemini and its future iterations will interpret and map human physiology; for the first time we'll have non-biological intelligence given the opportunity to understand the systems of life without relying on them. Maybe there's an organ system relationship humans are intrinsically blind to because we're limited by that system... or maybe there's more information we can collect, noninvasively, from a new model of the human body (esp. brain), that our limited senses have blinded us to... which physics tells us, is the vast majority of the universe. Fearing AI for its ability to replace jobs makes as much sense as fearing sails for their ability to replace oars; it isn't about what we are, it's about what we can be. A new frontier has been revealed! Explore!

[@saviskits]

This is absolutely amazing and hats off to the AlphaGo team! 🎉

[@simoneromeo5998]

This is really the most impressive revolution that Gemini is bringing. AI is literally starting to create leading scientific research during lunch breaks. How astounding is that! 🤔🤔😲😲🤩🤩

[@rudranarayanr]

I think these capabilities can help startups to revolutionise open innovation. Where anyone can submit them ideas and Gemini can filter best ones for that company. Don’t you think?

[@user-xy4io9nu2n]

This is exactly what I need it for.

[@francismumbi49]

We need that Colab

[@stupidbelike]

How can I use gemini? It looks amazing and i really want to try it out!

[@user-in9py5uu5o]

This is mind-blowing!

[@caelen_c]

AI is getting so advanced. I love how Google is leading the AI industry and making huge leaps in AI research

[@Feel_theagi]

The ranking of the comments is based on positivity in the comment rather than likes

[@Willyzzy]

Were you payed by google to post this lol

[@Martel24k]

This comment was probably written by Gemini

[@DavidHaselbach]

Is the code for this project somewhere available or is there a tutorial?

[@h.c4898]

Awesome. The next step to AGI. Good job Google@.

[@p_azambuja]

On the code displayed there are the functions “compute_relevance” and “extract_data”. Are those known python functions or custom made?

[@xaxfixho]

Only watched to see more of Taylor 🤭😘😍🥰🚀, gemini a big pile of what not!

[@churroholics8689]

You can already do this with a Google experiment known as Notebook LM and it provides references in the data you provide with the summaries. Google is the king of AI summaries since they have like 5 ways of doing it currently.

[@arpitgaur4310]

I've recently started an open access journal, can't wait to know how it revolutionise the Scholarly publishing

[@liganega5911]

Where to find the colab notebook used by Taylor Applebaum?

[@cieasypal]

This is awesome!! What python package are you using for this?

[@DdJ_SA]

How and where can I use this in the same way as in the video? I have a Gemini Advanced subscription and using the platform previously known as Bard, it struggles to make a database of relevant papers that I want, nevermind extracting information from them. So I feel like I'm missing something. Is this something only Gemini Ultra can do? Or am I just in the wrong place/on the wrong platform?

[@Colorado_Chris]

Love the updates and this will be a huge asset for academia. GPT is terrible for scientific papers.

[@hussainmohamed-5952]

how can I use it?

[@allobject]

What is Gemini preventing from falling victim to the echo chamber effect? How does Gemini detect that there exists other or new dimensions to the problem and solution space?

[@JohnsonThomasMD]

Is there an example Jupyter notebook

[@imbored4615]

I take that you can connect gemini to your accounts with access to those scientific journal right?

[@xristian3001]

has anybody outside google's payroll been able to reproduce this? Sounds too good to b....

[@devsuniversity]

Looking forward to tryout Gemini in Central Asia (Almaty city) developers community!

[@shubham-itachi]

What about job. I was learning Fullstack development.

[@JimmyMcG33]

As a discovery tool, this definitely has some benefits. But why would you want to offload learning onto a computer? If you can't understand and synthesize information for yourself you will be a worse researcher and dumber person. When that effect extends itself over researchers in general it can only depress the rate of new research. AI can certainly help people, but if you don't know how to read and understand research then you can't even formulate queries, let alone fact-check these very error prone AI systems.

[@jbram07]

[In]_(ABS)=>{omega(x,t) MABS configurations

[@cesaradrianarreola7266]

Welcome to the Gemini era. "Another suitable name besides Gemini could be 'I AM'."

[@maloxi1472]

"Gemini" is fine. No need to compare it to God.

[@p_azambuja]

That is pretty amazing. But from the information displayed on the demo it is unclear how bard would search pubmed directly and extract the search results. Does anyone has info on that?

[@visceralcinema]

Langchain. Look it up! **That's not official, but I'm certain Google used something of the sorts.

[@MS1996811]

If you work with NCBI data you usually get it through the eutilities API or the FTP server. I have already used their API and FTP servers tons of times for all sorts of data. I presume either the latest version of PMC and Pubmed was downloaded or the articles were first filtered using NCBI's esearch and then fetched using efetch. The first approach will give you a more complete set of articles, while being more expensive and noisy, while the other approach will give you more relevant results while there is the risk of missing out on some relevant articles. Since they said that it had been done in a short amount of time, my guess is that a filter has been applied using esearch and then the relevant articles were fetched and analysed then.

[@p_azambuja]

Thank you, that was very helpful!

[@user-yo9od8hh2q]

This is so beautiful, Would love to learn more on this ! ❤

[@TonyFarley-gi2cv]

That got pulled in rotation to the atmosphere's gravity or the electrical flow or pump flow of the scientific building or the school

[@nicohambauer]

Sounds astonishing! What are the estimated costs to let Gemini read 200k papers and how to reduce those?

[@emotionalfool8349]

Hey when i can use gemini ,in my android ? I can't wait

[@DerekHanjari-we1hi]

We welcome Google to Nairobi Kenya to advertise Gemini.

[@jimbojimbo6873]

That man is Johnny Sinns

[@Travdoteth]

The race for AGI continues. Here before AGI 🤙

[@venvedam]

Take a bow Gemini! 😮

[@vaibhavtripathi2520]

How can I build something useful from using Gemini? Any open-source code available?

[@Rohit-cj6eb]

Not yet

[@user-dm6hs2fe2p]

Great product Sebastian Sir you look like Haaland 🤔

[@DineshGaikwad]

Such an advacement would really propel humans to a new level for sure!

[@ricpb]

Since nobody is saying anything. I would like to point out that it swaped the graphs labels CRE and promoters

[@steliosp1770]

AI and Quantum Computing are the only way to make game changing breathroughs in physics in particular it feels like. Since Einstein and QM we ve not had a breathrough on that scale despite the absurd technological advancements. I think we ve reached our limit in terms of the capacity of the human brain. Its time to utilize AI and QC to help us move forward.

[@Paretozen]

this video shows the limits of the human brain, our limited time more accurately. We cannot sift through thousands of papers and analyse then condense this information. There is just too little time in a single human life to combine the insights from all the different cutting edge developments in science. The answers are out there, probably already hidden in the mass of data, just not obtainable by a human. AI will do this for us! Scary times

[@nicolasdujarrier]

@@ParetozenI agree 100% with that. It is likely that many answers to different problems already exist and is hidden in the scientific (or else) literature, but the literature is so vast, that for a human person it is impossible to read it all to locate the relevant one… In theory, AI could be an important tool to help on this challenge. However, it seems that it still struggles with mathematics…

[@Lightning_Speed]

How can one replicate this and assess its accuracy?

[@Lightning_Speed]

Is it better to test this with Gemini Ultra 1.0 or Gemini Pro 1.5? The idea is to compare a recent systematic review completed by humans to what Gemini can do.

[@ErManu10]

I am checking out the website and I cannot see where to click to use Gemini. Is it already available?

[@wishesandfishes]

Gemini is now the model powering bard

[@brandonfoy9583]

0:58 wait so how do I get to use this interface? How do I use Gemini ultra, beta test it?

[@velo1337]

what version of gemini is this demo done with?

[@bonnkenobi]

So, how can I use this? What's the url?

[@RaviPatel-hw7nc]

Has it shown any use cases in healthcare literature reviews?

[@WebWizard977]

Is it gona work on our own dataset like related to LLM dataset building or chatbot related stuff(like can we upload the dataset ,codes,videos,etc)

[@crumblegg]

This looks fantastic. That’s what I call a worthy tool. Thank you 🙏

[@Google]

Glad to have you along this journey!

[@MohamedAnsary]

🎯 Key Takeaways for quick navigation: 00:00 🤖 *Gemini, a multimodal model, is tested for outfit understanding.* 00:26 🛸 *Gemini humorously suggests outfit names like "Moon core" during testing.* Made with HARPA AI

[@user-qo7qt3wq7h]

HOW CAN WE USE THAT RIGHT NOW ?

[@gaurang_bansal]

sources of scientific papers used?

[@mrslake7096]

what about hallucinations? won't this add errors to scientific studies ( which can already include human error ) ?

[@Var_]

What is the UI used at 0:58?

[@COMPARTIENDOCIENCIAMX]

So Gemini can open and read any scientific paper without need to pay for it?

[@marlo-ji]

This is mind-blowing

[@fslurrehman]

It's interesting to see how keen researchers are to show their product and keen google is to keep it with itself safe from public use.😊

[@alizomorrodi8082]

This is fanstaisc capability but it wasn't clear from the video whether you had to download all those 200K papers locally and then have Gemini read and filter them, or if Gemini looks for these papers on the web? If former, it would be impractical to download 200K papers for each use case though.

[@Procrastinus]

That's the part that I want to know too. Getting access to the PDFs alone is a non-trivial problem, as most of them are behind paywalls or distributed among multiple sites. I imagine that they just threw millions of dollars at this to obtain the articles, which they then scanned and summarized. The field was going towards this anyway (I published a paper on this last week that now looks outdated before anyone even gets a chance to read it). The interesting part is the data extraction. We had played around with multiple AI tools, including the available Gemini and it couldn't read a correlation table effectively, so this seems beyond its capabilities. My understanding is they used a more updated version of Gemini that isn't available to the public yet, but I still need to see if for myself. If this is for real, the implications of this are so vast that I think I would slip into hyperbole trying to describe it.

[@alizomorrodi8082]

I too would love to examine this for myself. It would be fantastic if Google makes a prototype examples of this capability available to this public so everyone can test it and adapt it for their needs. This is something that I need for every single paper I write and a capability I would expect Generative AI tools to be able to address. .

[@alfonsogarciaalbert8630]

Y como resuelven el problema del Google pixel 8 con pantalla hueca,pareciera q es un teléfono de juguete al pasticho tiene un día que me llegó y tiene esa situación...

[@vazules]

the humanity will be progressing fast in near future

[@athusain678]

Openai: Hey Google, we are revolutionizing the artificial intelligence, what's your say?? Google: Gemini... 😌

[@jackbauer322]

Bertopic does it very well ....

[@Anonymous991]

I have always bet on Google's artificial intelligence. Because even though openai has a well-trained language model, Google has much more tools and potential

[@ankk98]

Damn, this is incredibly powerful.

[@canakademisyen]

You have done in 5 minutes a research that is at the top of the pyramid of evidence and can occupy a scientist's time for months. Would it accept a second authorship?