for the first time ever

@zauey

115.5K followers · usually 31.5K views a post

Views

162.4K

Likes

15.1K

Comments

655

2.6K

SoundTikTok sound · original sound - zauey

Why it worked

The video effectively uses striking visuals and a clear narrative to explain a complex scientific breakthrough, highlighting both the potential benefits and the alarming implications of AI in designing novel viruses.

Summary

Scientists have used an AI model called Evo to design novel viruses that have never existed in nature. These AI-designed viruses were successfully manufactured, infected bacterial cells, and some even outperformed natural viruses in spread and multiplication, showing potential for phage therapy against drug-resistant infections. However, the ability to design such pathogens with AI raises concerns about potential misuse due to a lack of regulatory oversight.

Structure

1AI designs novel viruses
2Evo model explained
3Bacteriophages and their potential
4AI-designed viruses outperform natural ones
5Concerns about AI-designed pathogens

Product placement

Evo (AI model that designs viruses and predicts DNA letters, acts, changes video content) Phi X 174 (virus that infects E. coli, appears, changes video content) Science (publication, appears, does not change video content) Wikipedia (source, appears, does not change video content) ChatGPT (AI model, mentioned as comparison, does not change video content) New York Times (publication, appears, does not change video content)

On-screen text

for the first time ever AI designs viruses not found in nature AI designs viruses not found in nature create new kinds of viruses that have never existed and they work the paper was published in Science today around this model called Evo and it's basically like a ChatGPT but instead of predicting the next word in a sentence it predicts the next letter of DNA they fed it about 9 trillion nucleotides of genetic code from entire tree of life bacteria plants animals and microbes and then they fine tune it for one specific virus called Phi X 174 which is actually a bacteriophage meaning it's a virus that infects bacteria this one deliberately only infects E. coli and can't infect humans they prompted it to produce its own original versions of that virus's genome 700,000 candidate genomes scientists then picked 300 of the most promising designs decided to physically manufacture them by dropping them into bacteria and what they found was 16 of them came to life these ended up becoming real viruses that infected and were able to kill the cells the crazy part is these AI designed viruses didn't just work a few of them actually outperformed the real Phi X 174 and spread and multiply faster than the natural virus so the model wasn't just imitating nature it actually produced new designs that beat the original on at least one measure the good news good news... what it could potentially unlock for the field of antibiotics good news... bacteriophages are these viruses that kill bacteria and doctors have always wanted to use them to treat infections but bacteria as we know evolves resistance and when that happens viruses often stop working so now you potentially have a way that you can design them faster than bacteria can evolve and this is important especially when drug resistant infections directly kill more than a million people a year now for the less exciting news not so good news... this team deliberately never trained Evo on anything that can potentially infect humans however the ability to write brand new genomes with AI now exists but there are currently very little rules oversight systems or even detection tools built to manage all this big risk is that there's a future version of this that is trained on something that isn't bacteria only that could very possibly design a human pathogen that we've never seen before

Original caption

For the first time, an AI model designed viruses that had never existed in nature (and they worked) 🧬 Scientists used an AI model called Evo, trained on 9 trillion nucleotides across the tree of life, to design viruses that never existed in nature. 🦠 Of ~700,000 generated genomes, they built 300 and 16 came to life, infecting and killing bacterial cells. ⚡ A few AI-designed versions actually spread faster than the natural Phi X 174 💊 The upside is phage therapy against drug-resistant infections, which kill over a million people a year. ⚠️ But the ability to write new genomes now exists, and the rules, oversight, and detection tools don't. #biotech #Science #ai #aisafety #biology (Views expressed are my own and don't represent my employer. This is for informational and educational purposes only.)

More from @zauey

More breakout videos from this creator.

Transcript

for the first time ever scientists have used AI to create new kinds of viruses that have never existed and they work the paper was published in Science today around this model called Evo and it's basically like a ChatGPT but instead of predicting the next word in a sentence it predicts the next letter of DNA they fed it about 9 trillion nucleotides of genetic code from the entire tree of life bacteria plants animals and microbes and then they fine tune it for one specific virus called Phi X 174 which is actually a bacteriophage meaning it's a virus that infects bacteria this one deliberately only infects E. coli and can't infect humans they prompted it to produce its own original versions of that virus's genome and it generated about 700,000 candidate genomes scientists then picked 300 of the most promising designs decided to physically manufacture them by dropping them into bacteria and what they found was 16 of them came to life these ended up becoming real viruses that infected and were able to kill the cells the crazy part is these AI designed viruses didn't just work a few of them actually outperformed the real Phi X 174 and spread and multiply faster than the natural virus so the model wasn't just imitating nature it actually produced new designs that beat the original on at least one measure the good news good news what it could potentially unlock for the field of antibiotics good news bacteriophages are these viruses that kill bacteria and doctors have always wanted to use them to treat infections but bacteria as we know evolves resistance and when that happens viruses often stop working so now you potentially have a way that you can design them faster than bacteria can evolve and this is important especially when drug resistant infections directly kill more than a million people a year now for the less exciting news not so good news... this team deliberately never trained Evo on anything that can potentially infect humans however the ability to write brand new genomes with AI now exists but there are currently very little rules oversight systems or even detection tools built to manage all this big risk is that there's a future version of this that is trained on something that isn't bacteria only that could very possibly design a human pathogen that we've never seen before