NEW SPECIAL FEATUREDemystifying generative AI with Graham Glass

Watch the video or read the answer:

The answer really depends on how you are sharing that data.

So for example, OpenAI is very upfront that if you're using the free version of ChatGPT and you're copying and pasting your data into its buffer, so to speak, then they are going to train OpenAI using that data. So you don't want to share important proprietary data with a free version of ChatGPT.

That being said, they're also upfront that if you're using the enterprise version of ChatGPT or if you are sharing data via their APIs, it is not used for training, it's not stored anywhere, it's just transient and will be destroyed.

At CYPHER Learning, we interface with AIs through these APIs and we only do that when the vendor and the owner of the API says this data is completely private. So if you're using a system like the CYPHER platform, which is using the APIs, you do not have to have any concern about your private data being shared with anyone.

Watch the video or read the answer:

The first thing is that the accuracy of AI is rapidly increasing. We've all heard of this so-called hallucination effect where sometimes it gets the answer wrong or it just makes up an answer, which sounds very convincing, but it's in fact incorrect. Which by the way is what humans also do. So let's also acknowledge the fact that this is something that's not unique to AIs - it's really a general concept with any information source. That being said, GPT 3.5 as an example, does it more often than GPT 4 and OpenAI has stated publicly that they believe that the future iterations are going to get more and more accurate. So the first thing is that I do think the error rate of AI is going to go down dramatically over the next year or so.

The second thing is, let's face it, we don't want to be silly. Just because an AI gives it to you doesn't mean that it's perfect. So generally speaking, you're going to apply the same kind of critical thinking and cross checking with information from an AI, that you would also perform whether it was coming from the web or from another human being. So basically put on your critical thinking hat.

In the CYPHER Learning platform, when we're creating educational content using AI 360 with Copilot, we are very honest and upfront and say AIs do sometimes make mistakes, you have to review the information. So we're not trying to sell someone that everything that you do using an AI is perfect.

We are, however, doing some things to reduce the probability of a hallucination and also making it easier for human reviewers to find possible errors.

So something called AI cross check, which will be coming out fairly soon from CYPHER Learning will use more than one AI and it will use one AI to precheck the accuracy of information coming out from the other AI. Once again, that doesn't guarantee that it's going to be 100% accurate, but it will tend to minimize that.

So I think generally speaking AIs are just like any other information source. You've got to be careful, you've got to check them, but they are going to get more and more accurate quite quickly.

Watch the video or read the answer:

99% of the time, no, but 1% of the time, maybe.

A little bit of a review on how AIs actually work. AIs do not copy and paste all the content in the internet into their digital brain and then regurgitate it. What they do is review huge amounts of content, extract the essence of that content, then codify it in terms of numbers.

So the AI almost certainly could never actually come up with exactly what it's read in the first place because that's not how it stores information. And much like a human being, it generates fresh information from everything that it's seen. And anyone who has used something like ChatGPT knows that every time you ask it a question, it gives you a slightly different answer. And that is one example of how it is not copying and pasting.

One of the things that OpenAI and Microsoft has done, is they've announced something called "Copyright Shield", which underscores their confidence and commitment to protecting the output of AI against any kind of copyright issues. So if you're using one of their products and someone comes after you and says, you violated a copyright, then Microsoft or OpenAI will actually come to your legal defense, which is a really great move. So generally, you can use the output in confidence, knowing that it's not copying and pasting from any particular source. And that covers 99% of the cases.

The 1% of the cases, that I think is worth acknowledging, is that there are some content creators, especially artists, that have spent years crafting a very, very specific, very unique identity. Picasso is an obvious example from the past, but there are current people doing this. And if an image generation AI learned all about Picasso and then started generating 100,000 Picassos using his specific, crafted, identity, then that almost certainly would be a case for a copyright violation.

So there are very, very few vertical cases where you do need to be careful, but in 99% of them minimum, you don't have to worry at all.

Watch the video or read the answer:

I think there are some areas of your business that could immediately take advantage of AI and it's not introducing any existential risk to your business. And there are other areas, especially one subject to high degrees of government regulation, where you might need to be careful.

But the other thing is you have to remember, your competitors might not hesitate in certain key areas and the longer that you take to apply AI, the more and more of a disadvantage you're going to have.

For example with e-learning, if you are building courses 100 times faster than and at a 10th of the cost of your competitors, is that going to give you a strategic advantage?

If you are a training organization or an IT organization, the answer is almost certainly, yes, you would be at a disadvantage.

Watch the video or read the answer:

We currently use 5 different AIs and over the next month or so, that's probably going to grow to 7 or 8 AIs.

The general idea is that we pick AIs based on what they're best at, how fast they are, and how much they cost. So we use one AI, for example, for image synthesis, another AI for voiceover, another AI for video transcription, another AI for content generation.

But the nice thing about our system is that none of this complexity surfaces to the user. You just click the checkbox, I want to synthesize images. And by hiding this from the user that gives us the flexibility to replace these AIs at any point when something better becomes available.

So obviously, we commit to our end customers that all their private data is completely private. So we're not going to share it with anyone. We strive very hard to make our systems work very fast and reliably. But aside from that, there's really no need for anyone to know what AIs we're using behind the scenes.

Watch the video or read the answer:

An AI prompt is basically just something that you say to an AI and depending on how you ask the question, you might get a different kind of response. Within our engineering team, we've got a lot of expertise now in how to ask the question in exactly the right way to get the best response back.

You don't have to know anything about AI prompts when you're using the CYPHER platform because we package the AIs, we hide all the nasty details from you. All you need to do is to check a few boxes, select a few menu options, and say Go.

Another question is, why can't I just do exactly what you're doing using AI prompts? And the answer is you actually could as long as you don't mind working about 100 times slower than what the CYPHER Learning platform can do for you. So to give you an idea, when you build a course on our platform, we trigger a minimum of 200 prompts, many of which are operating in parallel at superhuman speed. Then we take the results for all of those prompts and through a tight integration, we will generate the modules, the pages, the assessments, the quizzes, and the glossaries at superhuman speed.

If you were trying to do this with ChatGPT, what we do in 10 minutes would probably take you around 10 to 20 hours and every single time you want to repeat it, you'd have to spend another 10 to 20 hours doing that.

So, yes, you could do it. But I don't think it's a good use of your time.

Watch the video or read the answer:

Generally speaking, what we do is we try and make AI extremely accessible to anyone in the educational industry. It's packaged up, it has guard rails, and you can't really do anything crazy with it. So we've made it as safe as we possibly can. We can't force anyone to use AI against their wishes. But anyone who does want to use AI will find that it's very, very global policy friendly.

Watch the video or read the answer:

If you look at how people typically use learning platforms, it's to log in and then take a traditional course which is typically broken up into modules and sections. If you're lucky, it's gamified and if you're super lucky, it will also track your competencies. And that is an obvious area that AI can assist.

So specifically, AI can allow an instructor, professor, or a teacher to automatically create amazingly sophisticated courses in as little as 10 minutes. They're not gonna do 100% of the job, but they will do at least 80% of the job. But that's a fairly straightforward answer.

I think there's a much more interesting answer looking beyond the way that people have traditionally learnt using platforms. So for example, a lot of people have something they want to learn right now, something that's very important, mission critical, they don't have time or the inclination to try rummage around and find a course in your jukebox of courses and take it.

So one of the cool things that AI can do is that you can use generative AI. So you can learn something right there, on demand in an easy to use consumable format without having to take a course. And this is something that I think of as being kind of instant learning or just in time learning if you like.

I think the more profound thing that's going to occur as well, is that these platforms pretty soon are going to have personal agents. A personal agent is going to be a thing running 24/7 in the background, trying to find ways to help you to accomplish your learning objectives. They might be recommending you content, they might be automatically noticing useful news articles and summarizing them for you, they might be automatically following a week after you've consumed some material and giving you a little refresher. This is something which I think is going to revolutionize the way that learning platforms work. So stay tuned and let's see if this all comes true.

Watch the video or read the answer:

The short answer is no, why do I feel this? I used to be a teacher. I was an educator. I used to teach Computer Science at the University of Texas at Dallas. People really liked my classes and one of the reasons is the energy, enthusiasm, and anecdotes that I brought to the classroom, which then got everyone really excited about learning and put them in the right frame of mind. And obviously I delivered the core materials, but there was so much more that I did. I did lots of really cool hands-on projects - let's build a neural network simulator. And so I always got great reviews and people really enjoyed my classes and they still send me emails randomly today.

So wouldn't AI have done that? I don't think so. However, if I had gone into a lecture and all I said was "Ok everybody, turn to page 15. Now let's all read together", yeah, I could have probably been replaced by an AI if I was that bad and that boring.

But I do like to think that a majority of educators actually do a lot more. And the most important thing that you can do is to inspire and motivate your students.

Watch the video or read the answer:

Part one is the AI does not generate the whole course. The AI is meant to generate about 80% of the course, the core that frankly no one really wants to spend their time doing so that an instructor can spend the additional 20% adding their own flavor, adding their own anecdotes, making it theirs. If all the courses are the same, that means people aren't doing their job and they're just pressing a button and then they're just teaching it, which is not our philosophy at all.

The second thing is if you are going to build courses using AI, then you also want those courses to be modern and engaging, which means that you would like them to adopt a competency-based approach so that all of the materials are tagged back to the things that they are teaching and assessing. You would like them to be gamified. Ideally, you would like things like digital voice overs to make them engaging. What we've done in our course-building Copilot tool, is to build all of that directly into the tool to save people a lot of time and money.

I would say the last thing, honestly, which gives universities the biggest competitive edge is the kinds of instructors that they hire. There are poor instructors who will just teach by the book and there are really gifted instructors who know how to make any subject lively, fun, and memorable. You want to hire those kinds of people, the best platform you can deploy for them, the best AI integration you can give them, that'll make them happy and then they can spend more time doing what they do best, which is to inspire their students.

Watch the video or read the answer:

Overall I think it makes complete sense. I don't think there's anything weird in there. To give you an idea of the kind of things that they're thinking about, one of them is privacy. So if you're going to use an AI, you want to be assured that it's not going to take all of your data and share it with other people. So that makes good sense.

Another thing is related to national security. The government is going to work with key vendors such as OpenAI and Microsoft to try and make it as difficult as possible to use an advanced AI for nefarious purposes - such as building a biological agent or blowing up a nuclear reactor.

Another thing which is very important is there's no question that AIs are going to get used more and more for calculating algorithmically various outcomes which can affect your livelihood. So for example, you might be applying for credit, you might be applying for housing and we want to make sure that the AIs do not either directly or indirectly discriminate against any one particular audience, thereby putting them at a disadvantage.

And the last one is related to jobs. It's quite obvious that AI is going to displace a whole bunch of jobs over the next few decades, but ideally, people in those particular jobs can transition through a process of upskilling to other jobs. And one of the things that the US government wants to do is to try and make sure that that does not happen too abruptly, to give people in those jobs plenty of time to transition to other jobs.

So overall, I view the US Bill of Rights for AI as being very sensible and I don't see any particular issues with it.

Watch the video or read the answer:

I would say the answer is yes but with caveats.

Let's assume that the AI has access to your outputs - if you're an engineer, what kind of code do you write, if you're a creative writer, the text that you write. You can certainly evaluate that, I don't think there's any magic involved in that. But there's a huge number of things that you do during the day which really help your organization - whether you are fun to be around, you've got a great attitude, you've got a good team spirit, you rise to the challenge when the tough times hit. All of those are things that the AI probably won't have direct access to. Maybe someone could write it up, but unless we're surrounded by security cameras, they're not going to see you at 10pm at night fixing that really tough bug for an important customer.

So I think that yes, they will be able to evaluate outputs over time, but I think there's a lot more to evaluation than just evaluating your outputs.

Watch the video or read the answer:

I'm going to use a few examples, one from schools and universities, but it also applies equally well to businesses.

The thing that most people don't talk that much about right now is personal agents. What I mean by a personal agent is something that knows a lot about you, who is helping you, who is available 24/7, and is continually adjusting to your goals and your interactions with the agent.

Let's say you're interested in a particular topic and you talk to your agent and you say, “Hey, can you tell me a bit more about X?” So the agent knows now that you're interested in X. It has access to the entire internet, knows it can start finding lots of interesting related topics once it's interacted with you. If it gets the sense that you didn't quite understand, it might say, “Hey, let me explain this in a different way”. A week later, it might say, “I just want to do a quick pop quiz for you to see if you remember.” Which, by the way, I do all of these things for my kids, so I'm like their personal agent if you like.

You can compare that against what you see in a traditional learning environment where you go from one teacher in one year to another teacher, then you change schools or in the world of L&D you learn in this organization, then you move to a different organization in the same business, then you join a different business. It's constantly changing.

But with the personal agent, it won't matter because it's yours. It doesn't report to the business per se. The other thing is it will have access to so much information out there on the internet in real time. So if there is a new subject or a new related material, it can let you know within 10 minutes. Try doing that at school.

I think that the new AI generation, my kids for example, are already beginning to interact with AI and they're enjoying it. I think they're going to come to expect this form of customization and detailed interaction. There's always the danger that if they go to their school, university, or business and it's not somewhat up to speed, they are going to feel like they are in the dark ages in this place and I'm just going to get back to my trusty personal agent.

That's what I think is going to be a pretty big deal in the world of education.

Watch the video or read the answer:

It really depends on how they're using it. If you teach the class the same way that you always did, you give the same kind of questions, and the students can just turn around and use ChatGPT to quickly find an answer, then yes it's probably going to make lazy students even lazier, shall we say. That's asking for problems if you're just teaching the same way while people are using AI, you're going to be doing them a disservice.

What you need to do is to create a much more complex set of tasks for them to do that are going to get them to think deeply about how do I use an AI? How do I team up with it to achieve this much higher level of performance? In those situations, the students are actually having to think creatively about how to engage with the AI, whether it's how do I prompt engineer, how do I scrub it to make sure it's true because if I just copy and paste something and it's obviously wrong it's going to make the student look bad.

There are a lot of things that they have to do, but as I said, I think the main onus is going to be on the teacher or instructor to raise the bar of what's expected so that it doesn't allow people to get mentally lazy.

Watch the video or read the answer:

One of the things that many times goes unspoken is the fact that during the daytime you interact with a lot of different people, whether it's talking with a creative writer who's trying to pick your brain about which particular angle to take something on, if you happen to be in the medical services it might be people working in ER for example, or a neurosurgeon. I can think of tons of people that you interact with where the way that they interact with you is crucial to both of your outcomes.

If I go into an ER and there's a doctor who's not empathetic, is rude, and doesn't see that I'm in huge pain and I need to get priority, then obviously I'm going to be very dissatisfied. Whereas I've had some very positive interactions with the medical profession where they make you feel so at ease, they understand where you're coming from, they give you priority when it's necessary and it's a huge part of being a successful professional and I think the soft skills have always been pretty important.

I think now people are starting to say, “Hey, what is the difference between me and some AI?” One of the first things they will come up with is the soft skills aspect. I think there's going to be a lot of people around the world who are going to be quite motivated to improve their soft skills. Interestingly enough, software powered by AI might be one of the ways that you can accelerate your ability to have very good soft skills.

I know at CYPHER we're doing a lot of work with organizations and schools to use competency-based learning to measure soft skills and to target specific areas of strength and weakness.

I think, yes, they're definitely related and I think there's going to be a lot more emphasis on soft skills as a result of the resurgence of AI.

Watch the video or read the answer:

So today's question is “I know that CYPHER Copilot does an amazing job at allowing me to build engaging modern classes in minutes. However, do you see a time where Copilot will, in addition, help a learner to recognize their talent, to learn things much quicker that they're confused about, and otherwise help them on their educational journey?” 

The answer is absolutely! What you're referring to is somewhat known as an agent. CYPHER Learning is dedicated to building out our Copilot functionality to include exactly this kind of thing.

Watch the video or read the answer:

Turnitin recently announced that they were going to be able to significantly downsize their engineering team and replace a bunch of the engineers with AI. I've a few thoughts on that.

So first of all, let's just take AI off the plate completely. The best way to build software efficiently is to have small teams of really experienced people who create reusable frameworks and bring in the most advanced tooling, so that a small number of people can do a lot. Those are high performance engineering teams and that's the kind of team that we have here at CYPHER Learning.

There are certainly engineering teams that haven't adopted the latest software best practices where they do have more less talented engineers, shall we say, building a lot of mediocre stuff that could be done in a much more efficient way. And I think if you have an engineering organization like that, then it's certainly possible that you could downsize somewhat by leveraging AI. But I really do think in this case that AI is like the red herring. I think that you would normally reduce an engineering team regardless, if your team is so easily replaceable by AI.

At CYPHER Learning there's absolutely no way I can replace anyone in our engineering team by AI because they're just way too good, but they do use AI on a daily basis to make their own productivity more efficient.