Thank you.

Okay, less than one minute to quiz start.

Okay, quiz is starting. Okay, I think the quiz has ended.

Okay, looks relatively good. I have a couple of announcements to make. One is that we will be

finally getting exam dates mid-December or 10th of December or something like this. The Academic

Records Office apparently wants to use the actual maximum number of students who will take the exam

or have registered for the exam as a scheduling help. So that's what we're waiting for. They have

taken out the wrong dates so that nobody actually has a wrong impression. So we'll have to wait for

two more weeks. We are looking for teaching assistants for the AI2 lecture and we've gotten

more money for them. So please apply if you think you can TA AI2. Some of you have already taken it.

Some faces look very familiar. We're always looking for good people and you can actually,

as TAs, make life for the next generation or these people who are with you in the class much

better. So just write an email to Florian Raber and or me. We're currently working on practice

problems for the quizzes, like similar problems in a kind of recreational practice quiz.

There's one problem with them though. They won't be identical obviously. So we're trying to give you

the problems but note that we have two channels of making problems. One public one and one private

one of course. So there's going to be discrepancies and if people start complaining too much, I'll just

shut down the practice problems on grounds that they're not doing the right thing. And finally,

I've looked at the numbers. About 700 people are registered for the course in Stuttgart. About 320

have registered for the exam. About 200, exactly 200 today take the quizzes and about 70 to maybe

80 are in the lectures present. Make of that what you want. I'm kind of thinking that somewhere

between 320 and 200 is the number of people who are actively following the course. And there seems

to be a very good correlation between people who are actually in the lecture, i.e. physically present,

even though it's snowing and having good grades. But that of course is only a statistical correlation.

Right. That's what I wanted to say. Are there any questions?

Yes.

Of the questions or the...

Yes.

Okay, then we have to think about that. All the devices we've tried on actually. So that's an iPhone probably, right?

Are you using Safari?

Well then you've got yourself to blame. It's just, it may be the coolest browser in the world, but it's just not good for math.

No. On an iPad? Okay. Well, then I don't know. I don't use iPads. Yes, we can try. Okay. Can you send me an email about that?

Then I'll remember. Better. More questions? Then the questions were hard, right?

Sorry. Yes.

Oh, there are different numbers. Yes. 700 are registered on Stodon. 320 are registered for the exam.

200 are taking the quizzes and 80 or so, depending on the weather, are here.

The 20th of December is when we will know the dates of the exam. About, yes. Maybe well or 15th or something like this.

Any more questions?

So we've talked about various algorithms for adversarial search, which is the special kind of variant of search that we use when there's kind of a two-agent situation.

Two-agent antagonistic situation. In our algorithms we've baked the fact that our opponent always tries to harm us into the algorithm.

Two agents running hand in hand to make the world better need different algorithms. These are adversarial search.

This kind of two collaborative agents is a different thing. You can do it with vaguely the same ideas.

But that's not what we've done here.

I want you to understand that what we learn in this class is actually a set of ideas and ways of thinking about things.

And the AI problems out there will need you to decide what of these techniques we're learning here is the right one for the task.

And then I'm sure you need to kind of think about it and do certain things differently.

Just like we adapted search techniques to this adversarial setting.

And sometimes you need totally new ideas, like AlphaGo.

The state of the art in gameplay is essentially right now defined by systems like AlphaGo or AlphaZero or AlphaStar.

Which kind of have the same idea of reinforcement learning by playing itself in a kind of a virtual arena and learning from that.