there is nothing new here

Hello friends it's that time of year again where I you know see an article and then read it to you in just disbelief that someone wrote this down in this [Music] order is this what I care about right now today it's 11: p.m. this article came out today like is this what I'm doing with my life and I I will just say yes I think scientific communication is incredibly important I think the media misrepresenting science is one of the reasons we're here where we are today and I think it's important to call it out when

you see it and that's what this video is hi if this is your first video weird Choice I'm Angela some lore for this channel I am a professional theoretical physicist I I work in computation and I use Ai and machine learning tools every single day recently I've had to start writing software and that sucks for me but I understand when and how to use Ai and you know I even made a whole video about it I said AI doesn't exist actually these language learning models these machine learning tools are tools and unless you're an expert

you know who can recognize the utility of the results the Fidelity of the results when and where to actually use this tool it's kind of just garbage and you don't need it and in this this video I also made a bunch of predictions about how companies were just going to ram into our throats and it would just cause a lot of bad things to happen and I mean I have to say I nailed it I mean rarely does one make a video with predictions and then just they all just come true that all just happened

so so it is interesting to me that I still get comments on this video that came out a year ago predicted a bunch of things that actually happened and they'll be like this age liked milk look at this random Daily Mail article I about I saw about a new AI tool that cured cancer and I just please show me this super cool AI tool that came out in the last year that changed the world and actually worked and actually did anything useful please show it to me link it to me that'd be great like I

understand when it's used for scientific research do that every day what are you referring to is it that super cool AI tool that told Matthew mccon to find a restaurant that would not put him outside at a dining table during a hurricane is that the tool the booking app I used didn't have agent for so an AI agent didn't know to move my reservations inside or know what I like to eat this company is trying to trick its investors into thinking this tool is so useful that they paid for a Super Bowl commercial why does

it seem like the way they plan to monetize their language learning models is to make decisions that we already make every day like I I don't need you to AI Shop for pants for me I I just I know how to do that I'm not going to pay you to do that it's very uh right this store doesn't have agent force so an AI agent didn't tip off the stylus as to what I might actually wear yes oh I know the whole thing I just did was like super annoying and I should try to be

humble but like I released a bunch of predictions and they came true that's all that's all I'm saying anyway I think people are on to the bit now like it's been three full years of these companies with this billion dollar valuation zero product zero like useful utility to the software they're presenting us zero ideas for how to monetize this we we get it we get that it's not a thing that's useful at this moment like really the best you can do is search for a restaurant Google already does that I don't need an i it

boggles the mind but most people get it right it's only like the dumbest guy you know who still thinks Elon Musk is a genius is thinking like somehow AI is real and also it's going to specifically make them Rich so I think it's progress over the last year like for the people but you see the capitalist overlords they still need to make the AI thing make money even though they don't have an idea for what the product should be that people would actually pay for like Zuckerberg still has to release these embarrassing metaverse videos where

he's like look at all the progress we made like come to them no one wants that bro that's nothing mine left a note it said it's not you it's the wifi I found it while reaching for the milk who over his Wi-Fi that is bad I understand that you need to do that to make sure you have like the invisible imaginary billion dollars you have in your company but like nobody wants that it's embarrassing just do something else just do the goggles again or whatever you can't buy a computer or a cell phone or a

washer and dryer set without some garbage AI tool that nobody wants that claims it's going to save you time by like determining the spin cycle or making your grocery list or responding to your emails but you can't let it do that because there's no Fidelity in AI because like AI as we think of it does not exist so you spend your time checking what it's actually doing it saves you zero time and also by the way way while it's doing that it's stealing your data it's built on Stolen data it's destroying the environment by using

all the electricity in the world and like you're the product it's just it's just in there to take your information and we're just speedr running a climate disaster because these companies are shoving these tools in our face that nobody wants anyway here's the article um so the title is can Google's new research assistant AI gives scientists superpowers question mark and I mean you just what is this called The Law of headlines if it's asking a question then the answer is no so we we don't need to read further superpowers from AI bestowed on scientists by

the Google overlords hooray so uh let me read the tagline to you researchers who have been given I'm going to spill this on myself researchers who have been given access to Google's new AI co- scientist tool are enthusiastic about its potential but it isn't clear whether it can make truly novel discoveries we learned three things right Google has created an AI co- scientist tool and actually I'm in Google right now don't tell them I'm talking them two scientists are enthusiastic scientists are enthusiastic about this tool okay I'm sure we'll hear from those scientists who are

enthusiastic this sounds great enthusiastic is a very specific word that means like overly excited so I'm sure we'll see we'll see that from them in quotes and then finally it's not clear whether or not this thing can truly make novel discoveries and just before we go on yes it is clear because you and I know how language learning models work and we know that they they do not make novel things they produce results based on the data that is input that they think you're expecting ing that's what they do they don't make novel discoveries that's

not in their wheelhouse so we know right away that like no no the AI language learning model is not going to be making novel scientific discoveries yes it is clear whether or not a language learning model can do that can make a novel Discovery clearly it cannot and the people at Google know this I'm sure the person that wrote this article also knows this but I feel like for some reason they don't want you and I to know this you know because their company would lose value if you knew that like actually no AI isn't

going to be doing science for us because it's it's a computer and that's not how science works but you know then they might lose money so this article is by Michael leage and it was printed febru printed on the internet February 19th 2025 so let's get into it Google has unveiled an experimental artificial intelligence system that uses Advanced reasoning to help scientists synthesize vast amounts of literature generate novel hypotheses and suggest detailed research plans according to its press release the idea with the AI Co scientist is to give scientists superpowers says Alan cariculam that is

an amazing name and I hope I said it correctly at Google right so first of all I'm sure that Google definitely purchased all of these scientific works like they now hold the copyright to them so that the authors whose Blood Sweat and Tears went into those scientific articles can like get some monetization from Google trying to like make the job of scientist go away I'm sure that that's all above board and they didn't steal a single research paper and just shove the entire internet into their llm I'm sure because that that would be evil and

we know Google isn't evil they say so when they handbook and honestly I thought the headline was hyperbole like you know how sometimes a science writer will do interviews and like call and write it up and then send it back to the scientists and be like is this what you were trying to say does this make it clear and then an editor will just be like so the God particle is real as the headline and it's not anyone's decision other than the editor I thought this superpowers thing was like a last minute decision by the

editor but the Google guy in caric kosim literally says give scientist superpowers it's not a radioactive spider it's a language learning model like you feeded a bunch of data in the form of scientific papers it develops an algorithm to respond to input based on how all those papers are written so it spits something back at you that it has formulated from the text of all those papers it's it's not a superpower it's it's an advanced search engine but but you know this guy's an mdphd so maybe he knows what he's talking about maybe he literally

thinks it's a superpower I mean he did say that very specifically it gives scientists superpowers I mean he's a medical doctor I'm not a medical doctor I'm sure he knows what he's talking about but also he works for Google so the teams that tried it are enthusiastic about its potential and these examples suggest that the AI co- scientist will be helpful for synthesizing findings however it is debatable whether the examples support the claim that an AI can generate novel hypotheses and it just must be devastating to grow up with the intention of becoming a writer

and then having to do this like this was the by line you already wrote this exact same sentence we haven't seen anyone be enthusiastic about it yet except for the guy who knows worked at Google like we get it we've heard there are enthusiastic people I'm I'm waiting to hear for them like this is just word filler I I hate this for us it's almost like these articles are written because they know no one is actually going to read them they're only going to read the headline which says AI gives scientists superpowers even though that's

not what the text of the article says I think we haven't gotten to it yet I also just really wish they would elaborate on this debate that they say is being had over whether or not the llm can do novel hypotheses because we know it can't because we know how l M's work the scientists working in this field know that they know that an llm cannot make novel hypotheses does Michael Le page know that why doesn't he elucidate us the audience surely the entire audience of this article if they read past the headline doesn't understand

this yet you're not posting anything about the debate what's the debate who's debating that no actually some people actually do think that this llm is going to solve all of science who's saying that I just don't understand for instance Google says one team used the system to find new ways of potentially treating liver fibrosis however the drugs proposed by the AI have previously been studied for this purpose here's the quote the drugs identified are well established to be antibotic I didn't say that right antifibrotic says Steven O'Reilly at UK biotech company there is nothing new

here so Stephen O'Reilly a doctor he works in biotech he has a bunch of Publications does not work for Google and he says there is nothing new here because an llm would be a great tool to help you write like a review paper something to read everything that has been written and summarize like the most cited important results and then of course an expert would have to read the output go back and check for the Fidelity and make sure everything made sense of course but you could use this tool as like a library or perhaps

just a simple Google search but to suggest that it's going to come up with new ideas there's nothing new here that's not how this would work are we supposed to include step O'Reilly in the enthusiastic scientist group that are stated in the title because uh he doesn't seem very enthusiastic he said there's nothing new here team member Gary pelts at the Stanford University School of Medicine in California says two out of the three drugs selected by the AI co-senior the two he personally selected did despite there being more evidence to support his choices so this

group looked for drugs to do a thing he came up with his own idea to like test like man versus machine the Google AI and he's like whoa surprisingly two of the drugs that the Google came up with seemed like they could be promising but like Google didn't make Google the AI didn't make drugs right the AI read a bunch of articles about different drugs and was like maybe I try this I guess so that's what happened and um Gary pelts he works at Stanford he's a PhD MD he has a bunch of citation he's

like whoa it proposed treatments that I didn't even think of the AI co- scientist L LED Gary pels into a new idea to test oh my God oh my goodness but notice that little line at the end where it says Google gave him money to run this test okay guy who was paid by Google to say this the thing was good says the thing is good is not an enthusiastic endorsement but also that paragraph starts with the phrase while the potential use of the treatment isn't new so this these drugs have already been used and

tested to treat this illness and the AI knows that because it's fed hundreds of thousands of studies surely above board surely the scientists who work on that were paid for their labor before it was just shoved into Google's AI I'm sure but obviously a computer search program like an llm is going to be better at finding results than a human person with limited memory capacity has not read all the Articles has not had access to all that information the llm will pull out more things than a human this test has been done for the past

15 years of of course of course of course an llm can do that these drugs have already been used to treat this illness just not by this person so yeah this is a helpful tool like I said for something like a review paper to like make sure you're getting all your options but this is not making new science this is just reading old results and presenting them to you and the title of this article calls this a superpower it's essentially a library it's fast finding information it's a helpful tool it is not doing new new

science and all I'm saying is that we've heard from two separate scientists in this article and I've yet to see an enthusiastic endorsement of this tool that's going to change the world that's actually a superpower except for the guy who like actually directly works for Google called it a superpower so so let us keep reading in another paper Jose pinatas at Imperial College London and his colleagues describe how the co-cio thesis matching an unpublished Discovery he and his team study mobile genetic elements bits of DNA that can move between BAC by various means blah blah

blah unpublished work so the way science works is you have an idea and you study a thing and while you're studying that thing hundreds of more questions come up right so while you're studying the one thing that you're going to write the paper up about you're writing all this other stuff down to do later right so this is what happens his group studied a thing they they saw a bunch of interesting stuff they wrote up one paper they published it and then they wrote up and have yet to publish the next results so what they're

suggesting is that the Google AI Co scientists read all the papers read their paper and then discovered the unpublished result that Jose pinatas and team have written up but not yet put on the internet to publish okay wow that sounds exciting AI tools because that would mean that the AI had found something new a thing that shouldn't be possible so what's in this llm what's happening here and I mean maybe this is this is the enthusiastic endorsement uh we were shocked says piñatas I sent an email to Google saying you have access to my computer

is that right because otherwise I can't believe what I'm reading here so like maybe this is it um let's keep reading what is clear is that it was fed everything it needed to find the answer rather than coming up with an entirely new idea everything was already published but in different bits the system was able to put everything together you see the team had actually published those results in 2023 and the language learning model did have access to it so it just did what language learning models do and pulled all the relevant information together and

use an algorithm to to create sentences so I mean Professor Jose pinatas is like really impressed I guess I guess that's an enthusiastic endorsement but like I just you gave it all that information it is functioning exactly how you expect an llm and exactly how you would use an llm I mean it just did a literature review like where is the superpower you knew it could do that you just forgot it had the paper why is this news why is this in the news it's literally a functioning llm which is great can be used as

a tool you have to check the results yourself because there's no Fidelity AI doesn't exist but like it's it's not new here but also pinata's group had already done this research they're suggesting that the AI tool didn't know about this paper and then came up with an idea but like they had already done that research they had already gotten so far into that project that they had written it up so even if the llm had suggested that idea remember it didn't because it had access to that paper that wasn't fully published yet why would you

pay Google $30,000 a year for that like what exactly is that offering you that traditional just scientist coming up with a bunch of an ideas doesn't why specific would you pay Google money to come up with hypotheses for you a thing that just there's no short supply of interesting questions to ask in science why would anyone ever pay Google a million dollars a year for this and I know some people are going to be like but they're not that expensive chat GPT is only $1,200 a month but like the thing is that's the business model

right Google very specifically is very good at this business model they'll just email me University and be like hey guys what if we gave all your research teams unlimited storage wouldn't that be crazy like we support research we would love to do this for you and of course all the universities take them up on that offer and then four years later when there's 700 million files stored on Google's server Google is just like oh you know what actually $10 million a year and do you think the university is going to build the infrastructure to like

store all that data on site and hide H the staff it would take to do that no they're just going to give Google $10 million a year the same thing happened with Gmail like hey guys these are free email service give all your students emails it's going to be great and then four years later when they have everything set to Gmail everything is already set all the documents are already stored Google's just like okay you have to pay for that now and I'm not saying they shouldn't do that of course Google should make money of

course storing data takes money hosting web servers to host email addresses Tak money of course I'm just saying you should never trust that Google's going to give you something for free because they never ever will by the way as a scientist I don't think I want to be beholden to a company who does things like change the name of the Gulf of Mexico like that makes sense you can see how that might be a problem in scientific research you can see how quickly that becomes evil right a thing that Google would never do cuz like

it's literally in their hand book to not be evil but like just just imagine it just take this thought experiment with me like what if for some reason Google was like actually we're going to get rid of Dei and we're going to wait papers with only male authors as like 50% more important or we're going to wait papers with only American sounding man names 50% higher you can see how that would happen right so why why would you be foolish enough to wrap your entire research career around this it doesn't make sense I mean besides

Alan caric selinam you know mdphd and Gary pelt's mdphd and Dr Jose Pates at least we have Dr Stephen O'Reilly in our Corner speaking truth to power saying the obvious the truthful statement of there is nothing new here right the team tried other AI systems already on the market none of them came up with the answer he says in fact some didn't manage it even when fed the paper describing the answer the system suggest things that you never even thought about except you did pinatas you did cuz you already wrote that paper like what are

you talking about dude who hasn't received any funding from Google I think it will be gamechanging Jose come on at least make them buy you dinner before here's another line from the article in 2023 however the company announced that around 40 new materials had been synthesized with the help of gnome AI yet according to a 2024 analysis by Robert palgrave at the University of London not one of the synthesized materials was actually new this is nothing new and yet the headline calls into a superpower so Dr Dr palgrave coming in clutch with none of this

is new but also you and I know this because llms do not make new things that is not what they do to suggest that they do do that or to suggest that there is a debate about whether or not they can make new things is a not true statement they rearrange things they're very good at that they search things they're very good at that they are tools to be used by experts who can check the results they can rearrange data in new ways that maybe a human wouldn't think about but they do not make new

things that is not how this works in no way should this be called a co- scientist any more than you would call like the love Library a co- scientist like what is Google's role expected to be in this like are you allowed to sue Google when you get fired and all your papers get retracted because you used their co- scientists and they gave you a bunch of fake papers that they hallucinated but like it's fine we're going to ignore that that's a problem so I I don't I don't think this is a co- scientist at

all I think it's disingenuine to call it that it's frustrating anyway despite his findings pal grave thinks AI can help scientists in enal I think AI has a huge amount to contribute to science if it is implemented in collaboration with experts in their respective Fields he says which like yeah of course I'm a scientist hi I use AI every day I understand Ai and machine learning and language learning models are fantastic tools and of course we use them on very specific scientific implementations right what Dr palgrave said is not that Google's AI co- scientist is

a great tool to he said AI is a great tools when when used with experts is this is this the endorsement claimed in the by line because he said nothing about Google he said AI is a tool I just I just think that the media could could be a little more honest about what they're saying in their articles Michael Le page AI superpowers is the top not a single scientist gets a good result that wasn't already a result and suddenly there's a debate about whether AI llms can come up with new novel scientific research there's

no debate in this article there's no enthusiastic support of this tool there's the people who are paid by Google who are like yeah it's pretty good and the one guy who forgot he gave it a paper and then was surprised when it gave his paper back why is this the headline because the thing is people only read the head line and so people are going to look at this article Michael the page and they are going to say I heard that scientists don't even do science anymore they just use AI I heard that AI can

do all the science I heard that AI is just coming up with new theories all the time so like we don't even need scientists that's what's going to happen like this as written is not true right this title is not true you're just doing free advertising for Google and boy am I going to be so like red in the face if I found out this website is just owned by Google like I'm just actually reading advertising for [Music] Google because none of the scientists gave a ringing endorsement only that scientists use AI tools as long

as they're experts and can understand the output which is what everyone said a year ago on this video so like this is nothing new that's the quote of the video this is nothing new so why is it being reported on like this and the answer is because you want people to click on it you know they're not going to read it and you know they're just going to take away the little nugget of like I heard scientists don't even do anything anymore and I feel like that's bad man I feel like that's bad [Applause] [Music]

[Music] and yeah [Music] [Music] [Music] here's the tweet and my response on Blue Sky not Twitter everybody get off Twitter we're done with it if you need help coming up with a hypothesis you shouldn't be a scientist every scientist I know has a 100 projects ideas and lack time to get to them no one is oh I typo see that's why that's why Twitter is fun no one is sitting around twiddling thumbs without an interesting question to ask we are in the dumbest timeline and I just like read the author's presentation of the article need

some help with your research Google's AI Co scientists will come up with hypotheses and research plans for you imagine a myologist just like sitting in their office freaking the out because they're like oh God I answered all the questions about mushrooms I don't know what to do with all this funding that's not how science works this is like a child's view of science do you think scientists literally go sit in their office and they're like I'm going to come up with a new hypothesis today that's not how it works