What is generative AI and how does it work? - The Turing Lectures with Mirella Lapata

How are technologies like ChatGPT created? And what does the future hold for AI language models?
This talk was filmed at the Royal Institution on 29th September 2023, in collaboration with The Alan Turing Institute.
Subscribe for regular science videos: bit.ly/RiSubscRibe
Donate to the Ri and help us bring you more scientific lectures: www.rigb.org/support-us/donat...
Watch the Q&A: • Q&A: What is generativ...
Generative AI refers to a type of artificial intelligence that involves creating new and original data or content. Unlike traditional AI models that rely on large datasets and algorithms to classify or predict outcomes, generative AI models are designed to learn the underlying patterns and structure of the data and generate novel outputs that mimic human creativity.
ChatGPT is perhaps the most well-known example, but the field is far larger and more varied than text generation. Other applications of generative AI include image and video synthesis, speech generation, music composition, and virtual reality.
In this lecture, Mirella Lapata will present an overview of this exciting-sometimes controversial-and rapidly evolving field.
Mirella Lapata is professor of natural language processing in the School of Informatics at the University of Edinburgh. Her research focuses on getting computers to understand, reason with, and generate natural language. She is the first recipient (2009) of the British Computer Society and Information Retrieval Specialist Group (BCS/IRSG) Karen Sparck Jones award and a Fellow of the Royal Society of Edinburgh, the ACL, and Academia Europaea.
00:00 Intro
2:38 Generative AI isn’t new - so what’s changed?
8:43 How did we get to ChatGPT?
12:38 How are Large Language Models created?
22:48 How good can a LLM become?
26:57 Unexpected effects of scaling up LLMs
28:05 How can ChatGPT meet the needs of humans?
32:30 Chat GPT demo
38:07 Are Language Models always right or fair?
40:21 The impact of LLMs on society
42:54 Is AI going to kill us all?
--
A very special thank you to our Patreon supporters who help make these videos happen, especially:
modsiw, Anton Ragin, Edward Unthank, Robert L Winer, Andy Carpenter, William Hudson
Don McLaughlin, efkinel lo, Martin Paull, Ben Wynne-Simmons, Ivo Danihelka, Kevin Winoto, Jonathan Killin, Stephan Giersche, William Billy Robillard, Jeffrey Schweitzer, Frances Dunne, jonas.app, Tim Karr, Alan Latteri, David Crowner, Matt Townsend, THOMAS N TAMADA, Andrew McGhee, Paul Brown, David Schick, Dave Ostler, Osian Gwyn Williams, David Lindo, Roger Baker, Rebecca Pan
--
The Ri is on Twitter: / ri_science
and Facebook: / royalinstitution
and TikTok: / ri_science
Listen to the Ri podcast: podcasters.spotify.com/pod/sh...
Our editorial policy: www.rigb.org/editing-ri-talks...
Subscribe for the latest science videos: bit.ly/RiNewsletter
Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.

[@AbhinavJain-rq8fl]

Mirella Lapata’s ability to simplify complex topics with vivid examples is truly commendable. Her presentation style is both insightful and accessible.

[@shanti0369]

Brilliant, outstanding presentation on GenAI! She has put it in such a simple way that anyone can understand! I forwarded it to everyone I know! I strongly believe that the greatness lies in simplifying things and not complicating them! It shows her command and down-to-earth attitude on the subject! She is amazingly passionate! Her last name is "Lapata," which in Hindi language (plus a few other native East Indian languages) means "Lost," but in fact, it is reverse in her case because she actually "finds" those lost! For those lost with AI , she found them an easy way! Kudos to her! Congratulations to her and Ri!

[@sudkam827]

Mirella Lapata's clarity on the subject. This is very easy even for a Non-Technical person to understand .

[@GeorgeZoto]

A fresh air of presenting generative AI from different perspectives, wonderful to watch 😀

[@bnartey1]

When we have complex technologies explained in such clarity and understanding, there is nothing to say but to salute the presenter. Excellent presentation. My 10year old could simply understand the technology behind generative AI / ChatGPT4 and all the inherent benefits and risk associated. Thanks Prof.

[@devhims]

Always a great pleasure to listen to these lectures from the royal institution.

[@anshikagupta4931]

so beautifully presented honetly just so relatable, refreshing and just the right blend of everything !

[@UFG18]

This presentation is very easy to follow and understand. Great!

[@rahulgoyal3179]

amazing lecture.. loved the way Merella explained each aspect of GEn-AI without going into much technical details..

[@delliscool4924]

Mirella Thank you , You are full of wisdom like Athena !

[@Hybrid_Netowrks]

Thank you so much for explaining such a complex subject in this brilliant easy way. Much appreciate this.

[@asyraftojiman6801]

A very very good explanation. The presenter is very fluent in presenting and speaking clear vocabs. ☺

[@user-zi5cm3xu1q]

She is so real and passionate about what she is talking ❤

[@ericahoward7272]

Really great job. Refreshing, real energy. Easy to understand. Thank you!

[@karthikc326]

Brilliant, the best talk on Gen AI for a layman and techie alike. Kudos Ms Lapata

[@vittacharan5354]

This is great and presenter is very clear about her explanation , I would suggest to everyone must watch. Thanks !!

[@franknugent2801]

Very clear delivery of a complex topic thanks

[@angelatakano6072]

Excellent and clear lecture for the general public! It’s meant to be a divulgation lecture

[@Leebrucehe]

I just sat before the screen to learn about AI and saw this video. I knew nothing about AI or Chat GPT, only heard about them. But I was able to understand whatever she said within my own limits. Thank you for this video MS Mirella

[@venkatgopal3820]

You are a gifted communicator. Nice simplification.

[@lllima432]

she is the best trainer on earth on AI and ML ... superb video

[@bardandrebendiksen3987]

Thank you for a very accessible, eloquent, and enjoyable video! :)

[@HammadNajam]

Thank you for the lively input on this conversation about AI and Generative Models. This was Beautiful.

[@vaneufrasio]

Great lecture. Smart choice of words and examples to explain hard to understand concepts. Well done Mirella.

[@sushilmishra1582]

A complex topic explained on simpler terms. outstanding

[@ajithboralugoda8906]

The simplistic explanation shows her depth of understanding the subject matter!! Loved it, Kudos to the Prof.!!

[@steliosp1770]

incredible lecture. thank you so much TRI and Mirella!

[@Gravelsis]

Brilliant explanation, thank you so much. I finally understand how it works.

[@taphaba4340]

Excellent lecture performed by Mirella ! So much things learned

[@charliebrown445]

What a wonderful and knowledgable speech!

[@anandg2966]

good video and clear explanation for everyone to understand. well done Ms. Mirella Lapata

[@IsEqualToNavneet]

Excellent... Many thanks for all your efforts and sharing this🎉

[@ChildSpecialist]

Lovely AI basics tutorial from a wonderful astute and intelligently likeable speaker :)

[@Sdreaasmit]

🤭

[@venkataponnaganti]

Helpful. Demystifies certain things. Great talk. Thanks.

[@TheAISpectrum1]

Mirella Lapata's explanation of generative AI was really insightful! Her ability to demystify complex concepts makes it enjoyable for enthusiasts like me. As a big fan of hers, I appreciate the clarity she brings to the discussion. It's through educators like Mirella that we can all better understand the fascinating world of generative AI and its transformative potential. Quite a interactive and interesting session!

[@omg5501]

>comment brought to you by generative AI

[@bagginsbilbrough]

awesome - thank you so much.

[@RabiYadava]

Fantastic Lecture to present something considered complex in very simple language.

[@davidadeogun1630]

It was a great presentation and explained most clearly.

[@getbash]

Really insightful lecture! For anyone interested, here are some key takeaways: 1. Generative AI can produce content across various formats, including texts, images, and code, by learning from vast datasets. 2. Google Translate and Siri are simpler forms of generative AI already integrated into everyday technology. 3. ChatGPT operates on the principle of language modeling, predicting continuations of given text based on learned data. 4. Despite their capabilities, generative AI models often require further tuning to align with human needs and expectations. 5. The societal impact includes both potential benefits in automating tasks and risks such as job loss and the spread of misinformation. 6. The technology can also be misused to create convincing fake content, posing challenges when it comes to information integrity. What was your main takeaway from the lecture?

[@chandankumarmishra336]

Lovely and memorable presentation!

[@ablejose6171]

Very good presentation

[@iwantpeace6535]

Thank you so much for this interesting material., early Greek are real builders behinde our modern civilization..Αγαπώ την Ελλάδα , Σε ευχαριστώ πολύ💙🤍💙

[@AlokKumar-ym8bl]

Excellent explanation of AI Technology..🎉 great lecture..thank you for describing

[@citiarun2p]

Great presentation, Thank you!

[@biorage6670]

When you see the energy, wit and passion of this wonderful lady you understand that AI has still a looong way to go!

[@ashfaquehossain8592]

Superb lecture.Very insightful

[@KamalAPerera]

Lovely presentation...🌹❤️

[@NoNTr1v1aL]

Absolutely brilliant lecture!

[@trbt177]

This presentation is for someone that has been living under a rock for the last two years

[@omnibrain8]

Very easy to understand. Thank you.

[@sakshikothari7689]

Insightful. Loved the whole video 🙂. Great speaker.

[@roshancoolkarni]

Excellent introduction. Thank you Mirella and RI.

[@shekharputturputtur938]

Great presentation...Thank you

[@tresajessygeorge210]

THANK YOU...!!!

[@kemibrianolimba682]

Really Great Presentation…Thanks for sharing

[@ashamilan1661]

Informative, and very interesting Thank you

[@sarazio3484]

Thanks for Sharing

[@gagnassss]

Great lecture! Thank you.

[@adrikadatta1894]

Excellent excellent Lecture

[@akshayaramv]

Thank you so much :)

[@riverbender9898]

Wonderful presentation. Thank you.

[@leduran]

Gorgeous professor ❤❤

[@anitabhatia1900]

Amazing lecture! It was so easy to understand ❤🎉

[@gt2978]

Excellent video!

[@kriskarlaonline]

Great presentation 👏

[@alialdubaisi9666]

🎯 Key Takeaways for quick navigation: 27:15 🚀 Scaling AI models with more parameters enables them to handle a wider range of tasks. 28:21 🧩 Fine-tuning is crucial to align AI models with human goals and adapt them for specific tasks. 29:54 🤖 Creating AI systems that are helpful, honest, and harmless is a challenge, requiring user instruction and bias avoidance. 31:28 🎯 Fine-tuning with human preferences is key to making AI models more aligned with user needs. 37:42 ⚙️ Despite fine-tuning, AI models like GPT can still struggle to follow instructions, highlighting alignment challenges. 38:12 📜 Regulating AI models is complex, with potential societal, energy, and job impacts. 43:27 🤔 Concerns about superintelligent AI are tempered; we'll likely have specialized AI tools, requiring risk mitigation, not prevention. 44:57 📝 AI regulation is on the horizon, akin to regulations for other risky technologies like nuclear energy. Made with HARPA AI

[@CoG7TVBirminghamUK]

Thanks, helped a lot

[@pradeepdesai3583]

Very simply and nicely explained Gen AI

[@bhushanparate9601]

amazing lecture!

[@bhushanparate9601]

indeed

[@dimakhovzun659]

I really enjoy this lection😀

[@padmajac7089]

Excellent presentation

[@psinha07]

A wonderful session.

[@slowpainful]

I learned a lot from this lecture and Ms Lapata was an engaging lecturer. Obviously this is a super-top-level view of AI, which is what I needed. The mind boggles. I feel her optimistic summary is maybe too optimistic as it assumes a lot of benevolent actors in the future and also that bad actors will always act logically (predictably) and mitigate harm, which seems almost child-like in its naivete....after all, we've known about climate change and its connection with human activity since the 1950's and look where we are today. Not terribly reassuring.

[@decembersunita]

Love it ❤

[@basantchoubey5137]

great presentation on GenAI..👍

[@justinwalker4475]

Well that clears that up

[@moderncontemplative]

Lovely presentation! I hope that others have realized that, given the model she presented of different types of brains compared to the human mind/brain, it is now clear that size doesn't necessarily equate with extreme intelligence. Certain LLMs are already smarter than humans regarding certain IQ metrics -even though they have way less parameters!

[@12polizei24wegvonhier]

Great lecturer

[@jimwebber5709]

Thanks!

[@Pooter-it4yg]

A little correction. Berners-Lee didn't invent the internet, he invented the web - the intuitive user environment interlinked and formatted by hypertext that is carried on the internet. An understandable slip though, since the vast majority of people have only ever interacted with the internet through this construct.

[@vr.108]

Great!

[@Lupinicus1664]

Regarding the ARC testing (which I presume was meant to 'allay fears') it clearly states that there was no task-specific fine tuning. Almost all the bullet point tasks seem simpler to train to than recognising and understanding natural language. It would appear that a rogue state could develop an AI, with such task-specific training. I'm not trying to be alarmist but it should be understood that virtually all academics are (rightly) excited by the technology but give very little thought to its applications outside their spheres of interest. Those with a 'hacker' mindset would likely be able to quickly pervert future LLMs or whatever comes next, to perform tasks it was never designed to do. As better understanding of the fundamentals of this technology spreads I hope we can have more in-depth discussions about real-world ways to avoid issues. (Very enjoyable talk, from someone clearly both knowledgable and enthusiastic about their field)

[@macc240038]

Most commenters just say how great the video is. Nice to see someone point out something other than that.

[@guyprovost]

Great talk... As a small correction, Tim Berners-Lee didn't invent the internet, he invented de World Wide Web (www) and protocols associtated.

[@vroomik]

many mistakes were made, I assume it's overexcitement on Mirella's part

[@AttilaAsztalos]

Exactly. A geologist making the distinction mistake would be entirely excusable - an AI specialist, not so much.

[@MOMO-YYDS]

THX

[@srirambm]

AI made simple. excellent lecture for simple folks like me.

[@NK-cn6gw]

nice session

[@AndreasEisele]

Excellent presentation! But I found one glitch at 43:11: Time Berners-Lee did not invent the internet, but the WWW.

[@divyaraj-rana]

An amazing presentation on one of the crucial topic in the age of Artificial Intelligence! So exciting to see the simplicity and flow of clearly understanding the core principles of the subject. Thank you Prof for this lecture!

[@IN-COgnitIONs]

“Understanding WHAT” level - flat flat flat and non-sense. Try to think.

[@mikeshawn5864]

A.I. , as it stands today, it is computational statistic model. Generative A.I. is statistic model of action based on the A.I. model.😊

[@aharx]

good one

[@dinamaanchannel]

This lecture actually is displaying the logical and creative limitations and boundaries of GPT. I am certain many people reading my comment will be outraged. That’s what is expected.

[@kencory2476]

Berners-Lee invented the World Wide Web, not the internet.

[@kunwei2005]

this is a very very general intro about gen ai

[@krittapaswongchai6095]

damn, that's so cool

[@353189]

As we are using data from dfrnt sources like quora etc ,how we are going to train the model in future because may we will have less data in future.as all people will move to gen ai for their ans.

[@nathanokun8801]

SCALE affects EVERYTHING in the world to a large or small extent. This is true for analog systems, mechanical systems, or here. EVERYTHING!!!

[@swaroopas5207]

Great presentation! But why does the last slide always focus on the implications of AI and then quickly move on? It's equally, if not more, important to delve deeper into this topic.

[@LifestyleMint14]

The thumbnail is wow

[@EinsteinsHair]

At 18:50, if I recall correctly, she asked how many knew about transformers, and got a good amount of applause. I really wonder what percentage thought she was asking about giant robots, and how many thought she was referring to devices that change electrical voltage. It was more applause than I would have expected.

[@franciscorompana2985]

Yeah, she's thinking her audience is from kindergarten.😅

[@NatalyRobles00]

I came to the comment section looking for this comment.... i completely agree :)

[@franciscorompana2985]

@@NatalyRobles00 I got confused when they apllauded

[@angstrom1058]

Thanks for the 1,394,302,003th explanation. At the pace of tech, this morning was so yesterday.

[@vedantraut327]

Wonderful presentation with clarity. However in the beginning she said 'She wont keep the audience hanging' however as no one can predict the future I felt hanging at the end of the session.