Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,000 --> 00:00:01,860
and I think one of the reasons that chat
2
00:00:01,860 --> 00:00:03,600
gbt and stuff have been so hyped
3
00:00:03,600 --> 00:00:05,220
recently is because most people don't
4
00:00:05,220 --> 00:00:06,960
know what it is and so when you see it
5
00:00:06,960 --> 00:00:09,420
doing what it does you think this thing
6
00:00:09,420 --> 00:00:12,179
must basically be a person right because
7
00:00:12,179 --> 00:00:14,160
it's acting like one and and I should
8
00:00:14,160 --> 00:00:16,139
carry this by saying I'm not selling
9
00:00:16,139 --> 00:00:17,279
short these these incredible
10
00:00:17,279 --> 00:00:18,900
Technologies I'm just saying that it
11
00:00:18,900 --> 00:00:20,460
would be very silly to just completely
12
00:00:20,460 --> 00:00:22,080
use them blind and never check what they
13
00:00:22,080 --> 00:00:23,580
do right because we know they just make
14
00:00:23,580 --> 00:00:25,199
stuff up a lot of the time I'm glad you
15
00:00:25,199 --> 00:00:26,939
mentioned computer science do you think
16
00:00:26,939 --> 00:00:29,220
it's time to for more of us to learn
17
00:00:29,220 --> 00:00:31,320
computer science type stuff because of
18
00:00:31,320 --> 00:00:33,840
AI like maths and all these computer
19
00:00:33,840 --> 00:00:36,550
science stuff not really
20
00:00:36,550 --> 00:00:39,630
[Music]
21
00:00:40,100 --> 00:00:42,120
I've been saying that you need to learn
22
00:00:42,120 --> 00:00:44,520
artificial intelligence or AI question
23
00:00:44,520 --> 00:00:46,680
that a lot of you have been asking me is
24
00:00:46,680 --> 00:00:48,899
okay so how do I learn that so let's ask
25
00:00:48,899 --> 00:00:51,500
another friend
26
00:00:51,739 --> 00:00:54,000
yes you've mentioned this before but
27
00:00:54,000 --> 00:00:56,579
remind me which place do you recommend
28
00:00:56,579 --> 00:00:59,160
that I learn and others learn AI I
29
00:00:59,160 --> 00:01:00,960
really like brilliant it's one of those
30
00:01:00,960 --> 00:01:03,480
places where you can go and have a
31
00:01:03,480 --> 00:01:06,659
visual gamified way to learn Concepts
32
00:01:06,659 --> 00:01:09,659
and Mathematics behind Ai and machine
33
00:01:09,659 --> 00:01:11,939
learning you've recommended this a few
34
00:01:11,939 --> 00:01:14,100
times to me the way you said it was
35
00:01:14,100 --> 00:01:16,619
David if you want to learn AI I need to
36
00:01:16,619 --> 00:01:18,299
learn like statistics and stuff like
37
00:01:18,299 --> 00:01:21,180
that right yes I've got these road maps
38
00:01:21,180 --> 00:01:23,700
that actually helps you with Calculus
39
00:01:23,700 --> 00:01:26,100
and learning statistics and linear
40
00:01:26,100 --> 00:01:28,259
algebra all the stuff that you need to
41
00:01:28,259 --> 00:01:30,659
know for AI I'll say this David is on my
42
00:01:30,659 --> 00:01:32,520
team I'm really glad that he is David
43
00:01:32,520 --> 00:01:34,680
has strengths that I don't have and I
44
00:01:34,680 --> 00:01:36,000
think that's what's really important in
45
00:01:36,000 --> 00:01:37,979
life you need to learn from others the
46
00:01:37,979 --> 00:01:39,659
other will tell us you've done a lot of
47
00:01:39,659 --> 00:01:40,799
maths you've done a lot of computer
48
00:01:40,799 --> 00:01:42,600
science you've actually worked with AI
49
00:01:42,600 --> 00:01:45,180
stuff right I work in the medical field
50
00:01:45,180 --> 00:01:49,259
for with data science stuff so I really
51
00:01:49,259 --> 00:01:51,840
think like you need to know know all the
52
00:01:51,840 --> 00:01:54,240
statistics and calculus and linear
53
00:01:54,240 --> 00:01:56,460
algebra and the discrete mathematics
54
00:01:56,460 --> 00:01:58,439
that you need to learn which actually
55
00:01:58,439 --> 00:02:00,299
makes a lot of coding a lot easier for
56
00:02:00,299 --> 00:02:02,040
you that's brilliant so I'm looking on
57
00:02:02,040 --> 00:02:03,720
their website now the one that you've
58
00:02:03,720 --> 00:02:05,579
recommended that I go through is the
59
00:02:05,579 --> 00:02:07,979
data science foundations right that's
60
00:02:07,979 --> 00:02:10,679
like probability applied probability
61
00:02:10,679 --> 00:02:12,840
statistics fundamentals and then an
62
00:02:12,840 --> 00:02:14,520
introduction to neural networks and
63
00:02:14,520 --> 00:02:16,260
obviously me being me I just skipped all
64
00:02:16,260 --> 00:02:17,459
of that I went straight to learning
65
00:02:17,459 --> 00:02:19,620
neural networks but as David said what I
66
00:02:19,620 --> 00:02:21,180
really like about this website is it's
67
00:02:21,180 --> 00:02:23,099
gamified as he said so really great way
68
00:02:23,099 --> 00:02:24,599
to get started really want to thank
69
00:02:24,599 --> 00:02:25,980
brilliant for sponsoring this video
70
00:02:25,980 --> 00:02:27,900
brilliant as they say in the UK thanks
71
00:02:27,900 --> 00:02:30,060
hey everyone it's David Bumble back with
72
00:02:30,060 --> 00:02:32,700
Dr Mike pound Mike welcome thanks for
73
00:02:32,700 --> 00:02:35,640
having me back again Mike though it
74
00:02:35,640 --> 00:02:37,200
feels like the sky is falling again you
75
00:02:37,200 --> 00:02:38,940
know we had this interview previously
76
00:02:38,940 --> 00:02:41,280
and it was all this hype about AI but it
77
00:02:41,280 --> 00:02:42,780
seems to just getting be getting you
78
00:02:42,780 --> 00:02:44,700
know hotter and hotter so tell me is the
79
00:02:44,700 --> 00:02:46,200
sky falling am I going to lose my job is
80
00:02:46,200 --> 00:02:47,819
the future Bleak I think I think you're
81
00:02:47,819 --> 00:02:50,400
gonna be all right so relax
82
00:02:50,400 --> 00:02:52,980
oh it looks
83
00:02:52,980 --> 00:02:54,780
um are you just bring me into calm
84
00:02:54,780 --> 00:02:56,340
everything down a bit that's that's you
85
00:02:56,340 --> 00:02:58,379
know I think that the last six months
86
00:02:58,379 --> 00:03:00,239
particularly have been you know both
87
00:03:00,239 --> 00:03:02,580
unbelievable in terms of genuine hype
88
00:03:02,580 --> 00:03:04,080
like things that are really exciting
89
00:03:04,080 --> 00:03:06,060
appearing and also obviously totally
90
00:03:06,060 --> 00:03:07,860
overboard hype that's just getting
91
00:03:07,860 --> 00:03:09,599
really quite silly and everyone needs to
92
00:03:09,599 --> 00:03:12,060
calm down right so I think there's a bit
93
00:03:12,060 --> 00:03:14,400
of everything going on chat chat GPT is
94
00:03:14,400 --> 00:03:16,560
a is an incredibly impressive tool that
95
00:03:16,560 --> 00:03:18,840
works very very well I've I've done some
96
00:03:18,840 --> 00:03:20,700
really fun tests with it where I've
97
00:03:20,700 --> 00:03:22,920
pushed to see what it will do and some
98
00:03:22,920 --> 00:03:24,239
of the things it will do are quite
99
00:03:24,239 --> 00:03:27,060
amazing right on the other hand there
100
00:03:27,060 --> 00:03:28,560
are lots of things it doesn't do very
101
00:03:28,560 --> 00:03:30,840
well and one of the big problems we have
102
00:03:30,840 --> 00:03:32,700
at the moment is it won't always tell
103
00:03:32,700 --> 00:03:34,620
you it's one of those things and that's
104
00:03:34,620 --> 00:03:36,180
I think where we have something that
105
00:03:36,180 --> 00:03:37,920
needs addressing I've done some tests
106
00:03:37,920 --> 00:03:40,440
and I mean a lot of people I know have
107
00:03:40,440 --> 00:03:42,180
done tests and it's a it's amazing what
108
00:03:42,180 --> 00:03:44,459
it seems to be able to produce
109
00:03:44,459 --> 00:03:45,900
um I think the concern a lot of people
110
00:03:45,900 --> 00:03:48,659
have is like Mike I'm 18 years old or
111
00:03:48,659 --> 00:03:50,400
I'm let's say I'm older I want to switch
112
00:03:50,400 --> 00:03:53,040
careers to become a programmer or I want
113
00:03:53,040 --> 00:03:55,019
to get into cyber security or want to be
114
00:03:55,019 --> 00:03:56,220
a network engineer whatever some
115
00:03:56,220 --> 00:03:58,319
technical role and it feels like I'm
116
00:03:58,319 --> 00:04:00,120
just going to waste my time because chat
117
00:04:00,120 --> 00:04:02,040
GPT is just going to obliterate drop at
118
00:04:02,040 --> 00:04:03,840
jobs the first thing I would observe is
119
00:04:03,840 --> 00:04:05,459
that it's very nice for some of these
120
00:04:05,459 --> 00:04:07,019
big tech companies if that's the
121
00:04:07,019 --> 00:04:08,459
perception because it makes them look
122
00:04:08,459 --> 00:04:10,920
very very impressive right and so I
123
00:04:10,920 --> 00:04:12,840
think that the cynic in me a little bit
124
00:04:12,840 --> 00:04:15,299
is like this is there's no no pay R is
125
00:04:15,299 --> 00:04:17,519
bad PR kind of a situation they like to
126
00:04:17,519 --> 00:04:19,260
drop you know these tools get dropped as
127
00:04:19,260 --> 00:04:22,079
incredibly impressive Tech demos and and
128
00:04:22,079 --> 00:04:23,940
I'm not selling them short right very
129
00:04:23,940 --> 00:04:25,979
impressive but maybe not quite as
130
00:04:25,979 --> 00:04:27,600
impressive as they first appear on a
131
00:04:27,600 --> 00:04:29,400
service when you start to dig in and I
132
00:04:29,400 --> 00:04:30,780
think that's what's really important you
133
00:04:30,780 --> 00:04:32,699
know in science we spend a lot of time
134
00:04:32,699 --> 00:04:34,380
checking things and rechecking them at
135
00:04:34,380 --> 00:04:35,340
least that's what we're supposed to do
136
00:04:35,340 --> 00:04:37,979
right so I you know a PhD student comes
137
00:04:37,979 --> 00:04:39,360
to my office with some results and they
138
00:04:39,360 --> 00:04:41,280
say oh we've got 95 accuracy on some
139
00:04:41,280 --> 00:04:43,139
task and I think okay let's talk about
140
00:04:43,139 --> 00:04:44,580
which data you used and whether that's
141
00:04:44,580 --> 00:04:46,380
really true and whether when you use it
142
00:04:46,380 --> 00:04:48,060
on this new data you're going to get
143
00:04:48,060 --> 00:04:50,520
that same result and we spend ages going
144
00:04:50,520 --> 00:04:52,259
over and over the data again to make
145
00:04:52,259 --> 00:04:53,460
sure that when we actually publish it
146
00:04:53,460 --> 00:04:54,960
it's really as accurate as possible
147
00:04:54,960 --> 00:04:57,000
large language models are maybe not
148
00:04:57,000 --> 00:04:59,040
operating in quite that same way yes
149
00:04:59,040 --> 00:05:00,540
they release papers from time to time
150
00:05:00,540 --> 00:05:02,160
but mostly they release these big
151
00:05:02,160 --> 00:05:04,139
websites where you can try them out and
152
00:05:04,139 --> 00:05:06,060
they do incredibly impressive stuff and
153
00:05:06,060 --> 00:05:08,160
and they lie very impressively as well
154
00:05:08,160 --> 00:05:10,320
right I think that's the thing that we
155
00:05:10,320 --> 00:05:12,840
haven't quite uh got around so you know
156
00:05:12,840 --> 00:05:14,400
suppose you're a programmer and you've
157
00:05:14,400 --> 00:05:16,080
been using co-pilot and you've been
158
00:05:16,080 --> 00:05:18,900
using chat GPT also does code and you
159
00:05:18,900 --> 00:05:20,100
and you're a bit worried because it's
160
00:05:20,100 --> 00:05:22,560
just producing pretty decent code maybe
161
00:05:22,560 --> 00:05:24,479
you don't see it replacing you right now
162
00:05:24,479 --> 00:05:26,100
but you could see in 10 years maybe
163
00:05:26,100 --> 00:05:27,840
that's going to be a problem I think the
164
00:05:27,840 --> 00:05:29,280
problem is at the moment is it's very
165
00:05:29,280 --> 00:05:30,900
difficult to know where it's going I
166
00:05:30,900 --> 00:05:32,400
think a lot of researchers are
167
00:05:32,400 --> 00:05:34,080
suspicious of the idea that we can just
168
00:05:34,080 --> 00:05:35,759
make it continually bigger and bigger
169
00:05:35,759 --> 00:05:37,259
and more impressive and it will just get
170
00:05:37,259 --> 00:05:38,580
better and better you know when we talk
171
00:05:38,580 --> 00:05:40,560
to about how these models work they
172
00:05:40,560 --> 00:05:42,419
don't really have an internal model of
173
00:05:42,419 --> 00:05:43,680
what it is they're trying to do or
174
00:05:43,680 --> 00:05:46,080
anything really they just map text to
175
00:05:46,080 --> 00:05:47,520
other texts you know when I write a
176
00:05:47,520 --> 00:05:49,500
piece of computer code what I'm really
177
00:05:49,500 --> 00:05:52,500
hoping to do is is in my mind come up
178
00:05:52,500 --> 00:05:54,600
with an idea of the problem that needs
179
00:05:54,600 --> 00:05:56,160
to be solved so what the number the
180
00:05:56,160 --> 00:05:57,600
variables and things I'm going to need
181
00:05:57,600 --> 00:05:59,520
start to get them down on paper and then
182
00:05:59,520 --> 00:06:00,479
start thinking about how would I
183
00:06:00,479 --> 00:06:02,699
manipulate those variables using code to
184
00:06:02,699 --> 00:06:04,860
produce some result that I want chat GPT
185
00:06:04,860 --> 00:06:06,419
doesn't really work that way it just
186
00:06:06,419 --> 00:06:08,520
spits out code right and it happens a
187
00:06:08,520 --> 00:06:10,080
lot of the time to look pretty good at
188
00:06:10,080 --> 00:06:11,880
the moment it's a tool to be used quite
189
00:06:11,880 --> 00:06:13,860
carefully particularly with code I
190
00:06:13,860 --> 00:06:15,720
wouldn't push anything chat GPT has
191
00:06:15,720 --> 00:06:17,100
written straight into production without
192
00:06:17,100 --> 00:06:19,080
you know quite a few quite a few tests
193
00:06:19,080 --> 00:06:20,520
because at the moment there's no
194
00:06:20,520 --> 00:06:22,800
grounding in reality right the reality
195
00:06:22,800 --> 00:06:24,960
is for training data but once it's
196
00:06:24,960 --> 00:06:27,000
finished training you it's kind of
197
00:06:27,000 --> 00:06:28,860
random what it gets and these things
198
00:06:28,860 --> 00:06:30,660
actually I don't know if you've noticed
199
00:06:30,660 --> 00:06:32,639
this David but when you run it it can
200
00:06:32,639 --> 00:06:34,139
produce different answers each time yeah
201
00:06:34,139 --> 00:06:35,639
and that's because it uses something
202
00:06:35,639 --> 00:06:38,100
called temperature to somewhat randomize
203
00:06:38,100 --> 00:06:40,620
its output so instead of saying okay the
204
00:06:40,620 --> 00:06:42,600
next word in my in my output is going to
205
00:06:42,600 --> 00:06:44,220
be the it will say I think there's an 80
206
00:06:44,220 --> 00:06:46,139
chance that it's there but it's a 20
207
00:06:46,139 --> 00:06:48,780
chance that it's so and then what you
208
00:06:48,780 --> 00:06:50,580
what the machine will do is say well
209
00:06:50,580 --> 00:06:52,319
okay 20 of the time then we'll pick a
210
00:06:52,319 --> 00:06:54,120
different word and that way you can go
211
00:06:54,120 --> 00:06:55,740
in slightly different directions because
212
00:06:55,740 --> 00:06:57,000
if you didn't do that it would just
213
00:06:57,000 --> 00:06:58,740
produce the same output every time it's
214
00:06:58,740 --> 00:07:01,440
not it's not a random object so it's not
215
00:07:01,440 --> 00:07:03,600
a random Network in that sense and so
216
00:07:03,600 --> 00:07:04,800
you can imagine a situation where there
217
00:07:04,800 --> 00:07:06,419
is a really good version of this program
218
00:07:06,419 --> 00:07:08,100
that it could write but it randomly
219
00:07:08,100 --> 00:07:09,419
didn't and produced one with loads of
220
00:07:09,419 --> 00:07:11,819
bugs so you know assume that's what I've
221
00:07:11,819 --> 00:07:13,740
experienced yeah sure yeah and and you
222
00:07:13,740 --> 00:07:15,180
know I suppose there's a question in my
223
00:07:15,180 --> 00:07:17,039
mind about how is very inefficiency
224
00:07:17,039 --> 00:07:18,419
saving if you have to order everything
225
00:07:18,419 --> 00:07:20,520
you're reading right is reading code as
226
00:07:20,520 --> 00:07:22,620
fast as writing code or slower or faster
227
00:07:22,620 --> 00:07:25,500
I'm I I don't know right I'm undecided I
228
00:07:25,500 --> 00:07:27,180
think sometimes for boilerplate codes
229
00:07:27,180 --> 00:07:29,520
probably pretty effective um if it's a
230
00:07:29,520 --> 00:07:32,160
sort of code you know write me a a for
231
00:07:32,160 --> 00:07:34,500
Loop to do X Y and Z probably works
232
00:07:34,500 --> 00:07:36,300
pretty well as long as you're capable of
233
00:07:36,300 --> 00:07:37,740
quickly checking that but then it didn't
234
00:07:37,740 --> 00:07:38,819
take me very long to write four looping
235
00:07:38,819 --> 00:07:41,819
anyway I'm undecided I suppose as to how
236
00:07:41,819 --> 00:07:44,340
much of a game changer that will be this
237
00:07:44,340 --> 00:07:46,080
said I know there are developers that
238
00:07:46,080 --> 00:07:47,819
use it right and I know that the
239
00:07:47,819 --> 00:07:49,860
developers who claim or at least they
240
00:07:49,860 --> 00:07:51,660
think it's they're much more efficient I
241
00:07:51,660 --> 00:07:52,979
don't spend as much time coding as I'd
242
00:07:52,979 --> 00:07:55,020
like because I'm I see you know isn't as
243
00:07:55,020 --> 00:07:57,060
a professor in the University I spend a
244
00:07:57,060 --> 00:07:58,080
lot of time teaching a lot of time
245
00:07:58,080 --> 00:08:00,479
mentoring others right and teaching
246
00:08:00,479 --> 00:08:03,120
people so they do the coding and I sit
247
00:08:03,120 --> 00:08:05,340
there and look at it right so I haven't
248
00:08:05,340 --> 00:08:07,500
had as much experience as some yeah I
249
00:08:07,500 --> 00:08:09,060
mean I think the the concern is always
250
00:08:09,060 --> 00:08:10,740
you know younger young people are people
251
00:08:10,740 --> 00:08:12,780
trying to switch careers is you know I
252
00:08:12,780 --> 00:08:14,099
want to have a job for more than a year
253
00:08:14,099 --> 00:08:15,840
or five years
254
00:08:15,840 --> 00:08:17,580
um is it worth putting all the effort in
255
00:08:17,580 --> 00:08:19,259
to learn this stuff if AI is just going
256
00:08:19,259 --> 00:08:21,360
to take it away my gut tells me that AI
257
00:08:21,360 --> 00:08:22,740
isn't going to take it away anytime soon
258
00:08:22,740 --> 00:08:25,080
right because I think that I would argue
259
00:08:25,080 --> 00:08:26,520
that you need something more fundamental
260
00:08:26,520 --> 00:08:28,139
to understanding some of these problems
261
00:08:28,139 --> 00:08:29,639
if you're going to write code to solve
262
00:08:29,639 --> 00:08:31,919
them than just a text production
263
00:08:31,919 --> 00:08:34,320
mechanism that isn't to say that what it
264
00:08:34,320 --> 00:08:36,120
doesn't do it's very impressive what it
265
00:08:36,120 --> 00:08:37,679
does but I think that as you start to
266
00:08:37,679 --> 00:08:39,120
build up you know it's very it's all
267
00:08:39,120 --> 00:08:40,559
very well saying write me a for Loop to
268
00:08:40,559 --> 00:08:42,120
do this but if you want to write your
269
00:08:42,120 --> 00:08:44,339
class structure and a and a really
270
00:08:44,339 --> 00:08:46,500
complicated system that's such a more
271
00:08:46,500 --> 00:08:47,640
difficult you know it's like the
272
00:08:47,640 --> 00:08:49,080
difference between Lane assist and
273
00:08:49,080 --> 00:08:51,300
self-driving right and that's why we can
274
00:08:51,300 --> 00:08:53,459
we've seen Lane assist exist but
275
00:08:53,459 --> 00:08:55,260
self-driving seems to be so hard to get
276
00:08:55,260 --> 00:08:57,899
to because of how much harder that is as
277
00:08:57,899 --> 00:08:59,760
a problem and I think that it's very
278
00:08:59,760 --> 00:09:01,680
easy to fit a straight line upwards to
279
00:09:01,680 --> 00:09:02,820
these things right you say well they
280
00:09:02,820 --> 00:09:03,839
didn't do anything and now they're doing
281
00:09:03,839 --> 00:09:05,820
this which means they're going to be
282
00:09:05,820 --> 00:09:08,040
doing this it may get a lot harder and
283
00:09:08,040 --> 00:09:10,620
Plateau out right we you know it's
284
00:09:10,620 --> 00:09:12,899
difficult to say for sure I think that
285
00:09:12,899 --> 00:09:14,700
there's going to be a very strong need
286
00:09:14,700 --> 00:09:16,980
for people in the loop for a lot for a
287
00:09:16,980 --> 00:09:19,500
long time further right I mean as an
288
00:09:19,500 --> 00:09:21,480
example outside of programming in
289
00:09:21,480 --> 00:09:23,399
medical science AI is obviously used
290
00:09:23,399 --> 00:09:25,500
quite a lot to help with diagnoses and
291
00:09:25,500 --> 00:09:29,160
things but almost no AI systems are used
292
00:09:29,160 --> 00:09:31,140
just on their own with no human
293
00:09:31,140 --> 00:09:33,060
oversight because for a start because we
294
00:09:33,060 --> 00:09:34,380
don't trust them yet and also because
295
00:09:34,380 --> 00:09:36,480
patients don't trust them patients don't
296
00:09:36,480 --> 00:09:39,240
want an AI even if it's good making
297
00:09:39,240 --> 00:09:41,519
their health decisions right like not
298
00:09:41,519 --> 00:09:43,620
yet you know and so I think also
299
00:09:43,620 --> 00:09:45,060
culturally we're not quite we're not
300
00:09:45,060 --> 00:09:46,860
quite ready and I know a few companies
301
00:09:46,860 --> 00:09:48,660
that are not using copilot because
302
00:09:48,660 --> 00:09:49,620
they're not absolutely sure the
303
00:09:49,620 --> 00:09:51,540
copyright on on the code and think you
304
00:09:51,540 --> 00:09:52,800
know there's questions that haven't been
305
00:09:52,800 --> 00:09:54,420
answered I think if you're looking for
306
00:09:54,420 --> 00:09:55,620
it to be a software developer or you're
307
00:09:55,620 --> 00:09:57,600
looking for a career in security or
308
00:09:57,600 --> 00:10:00,060
career in AI there's still plenty of
309
00:10:00,060 --> 00:10:01,860
things to do so I wouldn't personally
310
00:10:01,860 --> 00:10:03,839
worry about that I think we we mentioned
311
00:10:03,839 --> 00:10:05,580
this last time and I I want I want to
312
00:10:05,580 --> 00:10:07,800
give people firstly you know a way to
313
00:10:07,800 --> 00:10:09,360
make themselves more valuable and then a
314
00:10:09,360 --> 00:10:10,920
path to get there
315
00:10:10,920 --> 00:10:13,320
um you mentioned that you know any if
316
00:10:13,320 --> 00:10:15,600
you attach AI to any skill that you've
317
00:10:15,600 --> 00:10:17,640
got it's going to make you more valuable
318
00:10:17,640 --> 00:10:20,100
um I I assume that's still the case and
319
00:10:20,100 --> 00:10:21,660
I want to ask you Mike how do I get
320
00:10:21,660 --> 00:10:23,160
there and it also makes you more
321
00:10:23,160 --> 00:10:24,540
experienced at dealing with things like
322
00:10:24,540 --> 00:10:26,040
this when something comes along you can
323
00:10:26,040 --> 00:10:27,899
you can sit back and you can say okay
324
00:10:27,899 --> 00:10:29,459
how impressive is this let's think about
325
00:10:29,459 --> 00:10:31,500
what it's doing and how it works and you
326
00:10:31,500 --> 00:10:32,760
know some understanding of how these
327
00:10:32,760 --> 00:10:33,899
things work you don't have to understand
328
00:10:33,899 --> 00:10:36,360
deep down transformer networks if you
329
00:10:36,360 --> 00:10:37,560
want to understand roughly what they're
330
00:10:37,560 --> 00:10:39,240
doing right and how they I've been
331
00:10:39,240 --> 00:10:40,860
trained yeah I would say some knowledge
332
00:10:40,860 --> 00:10:43,560
of statistical analysis and data data
333
00:10:43,560 --> 00:10:45,000
processing in general is really really
334
00:10:45,000 --> 00:10:47,399
important right people mock Excel Excel
335
00:10:47,399 --> 00:10:48,779
is I think one of the best products ever
336
00:10:48,779 --> 00:10:51,300
written it's totally ubiquitous it's
337
00:10:51,300 --> 00:10:53,220
very powerful and it underpins huge
338
00:10:53,220 --> 00:10:55,079
amounts of you know Financial systems
339
00:10:55,079 --> 00:10:56,760
and other systems I use it all the time
340
00:10:56,760 --> 00:10:59,760
from for student marks right so you know
341
00:10:59,760 --> 00:11:01,680
you get a table of data that comes in
342
00:11:01,680 --> 00:11:03,600
and it doesn't make any sense what we're
343
00:11:03,600 --> 00:11:04,800
going to look at how we're going to deal
344
00:11:04,800 --> 00:11:06,120
with this right and how we're going to
345
00:11:06,120 --> 00:11:08,100
make decisions based on this data and
346
00:11:08,100 --> 00:11:09,779
things like data science and machine
347
00:11:09,779 --> 00:11:11,760
learning will help you deal with some of
348
00:11:11,760 --> 00:11:13,680
these problems people who want to become
349
00:11:13,680 --> 00:11:15,899
experts in AI obviously need to delve a
350
00:11:15,899 --> 00:11:17,579
bit deeper but I think for a lot of
351
00:11:17,579 --> 00:11:19,440
people AI can just solve small problems
352
00:11:19,440 --> 00:11:21,060
in your pipeline that might make things
353
00:11:21,060 --> 00:11:22,800
a little bit easier having that extra
354
00:11:22,800 --> 00:11:24,899
string in your bow it's not it's not a
355
00:11:24,899 --> 00:11:26,820
terrible idea so in previous videos I
356
00:11:26,820 --> 00:11:28,560
told people you need to learn Ai and
357
00:11:28,560 --> 00:11:29,640
it's something that I want to really
358
00:11:29,640 --> 00:11:31,920
focus on this year and this is why I'm
359
00:11:31,920 --> 00:11:33,240
talking to you you know right in the
360
00:11:33,240 --> 00:11:35,279
beginning of the year uh have you got
361
00:11:35,279 --> 00:11:38,579
like courses places that I can go to
362
00:11:38,579 --> 00:11:40,260
books that I can read any
363
00:11:40,260 --> 00:11:42,000
recommendations of how do I go from like
364
00:11:42,000 --> 00:11:44,399
where I am now zero knowledge to yeah at
365
00:11:44,399 --> 00:11:46,440
least you know getting down that path to
366
00:11:46,440 --> 00:11:48,300
be able to put it online there are loads
367
00:11:48,300 --> 00:11:50,160
there's loads of books and resources in
368
00:11:50,160 --> 00:11:52,140
Python to learn machine learning and
369
00:11:52,140 --> 00:11:54,120
data science um and that would be a
370
00:11:54,120 --> 00:11:56,040
great place to start I you know I've
371
00:11:56,040 --> 00:11:57,480
said it before many times I have a love
372
00:11:57,480 --> 00:11:59,220
hate relationship with python I like it
373
00:11:59,220 --> 00:12:00,360
sometimes and I don't like it other
374
00:12:00,360 --> 00:12:01,800
times at the end of the day there are
375
00:12:01,800 --> 00:12:03,120
libraries in Python that do quite
376
00:12:03,120 --> 00:12:05,220
incredible machine learning and make
377
00:12:05,220 --> 00:12:06,899
your life a lot easier right so we've
378
00:12:06,899 --> 00:12:08,640
got things like psychic learn we've got
379
00:12:08,640 --> 00:12:10,800
tensorflow and Pie torch of course but
380
00:12:10,800 --> 00:12:12,360
there are tutorials and books written
381
00:12:12,360 --> 00:12:14,339
around these things and they take you
382
00:12:14,339 --> 00:12:16,200
from I don't know what this network is
383
00:12:16,200 --> 00:12:17,519
to I can actually get one of these
384
00:12:17,519 --> 00:12:19,680
networks running on a machine and it's
385
00:12:19,680 --> 00:12:21,120
often not that much code because of
386
00:12:21,120 --> 00:12:22,200
these libraries do a lot of heavy
387
00:12:22,200 --> 00:12:24,060
lifting for you often it becomes more
388
00:12:24,060 --> 00:12:25,920
plug-in building blocks together than it
389
00:12:25,920 --> 00:12:27,360
does right you know writing neural
390
00:12:27,360 --> 00:12:28,800
network layers from scratch which no you
391
00:12:28,800 --> 00:12:30,600
know we don't do anymore you know so you
392
00:12:30,600 --> 00:12:32,220
can start by just plugging some things
393
00:12:32,220 --> 00:12:33,779
together and I've got a rudimentary
394
00:12:33,779 --> 00:12:34,860
Network that I don't really understand
395
00:12:34,860 --> 00:12:36,959
that's doing this classification and
396
00:12:36,959 --> 00:12:38,100
before long you've made your
397
00:12:38,100 --> 00:12:39,360
classification problem a little bit more
398
00:12:39,360 --> 00:12:40,680
complicated and you've got multi-class
399
00:12:40,680 --> 00:12:42,000
classification and then you've got a
400
00:12:42,000 --> 00:12:43,260
slightly different data set and then
401
00:12:43,260 --> 00:12:44,700
you've solved a data augmentation
402
00:12:44,700 --> 00:12:46,200
problem and you can add these things in
403
00:12:46,200 --> 00:12:48,360
and slowly work towards a bit more
404
00:12:48,360 --> 00:12:50,700
experience you know I have you know a
405
00:12:50,700 --> 00:12:52,380
number of undergraduate project students
406
00:12:52,380 --> 00:12:55,260
every year so in University in the third
407
00:12:55,260 --> 00:12:57,120
year you often do a dissertation which
408
00:12:57,120 --> 00:13:00,360
is like a um like a focused project over
409
00:13:00,360 --> 00:13:02,300
a whole year after most of my
410
00:13:02,300 --> 00:13:04,320
dissertation projects are going to be on
411
00:13:04,320 --> 00:13:06,240
AI and something like this and you know
412
00:13:06,240 --> 00:13:07,980
these are students who've done some you
413
00:13:07,980 --> 00:13:09,420
know machine learning maybe a little bit
414
00:13:09,420 --> 00:13:10,500
in their modules throughout their
415
00:13:10,500 --> 00:13:11,880
undergraduate and they know how to code
416
00:13:11,880 --> 00:13:14,220
but a lot of it's new you know we pick
417
00:13:14,220 --> 00:13:15,720
it up and we run with it and we and we
418
00:13:15,720 --> 00:13:17,700
we drew some great stuff I've got some
419
00:13:17,700 --> 00:13:19,260
um I've got some students in the second
420
00:13:19,260 --> 00:13:21,240
year solving Rubik's Cubes using machine
421
00:13:21,240 --> 00:13:22,800
learning to detect where the colors are
422
00:13:22,800 --> 00:13:24,000
and things like this and this is from
423
00:13:24,000 --> 00:13:25,800
scratch right so this is this is people
424
00:13:25,800 --> 00:13:27,600
who haven't done machine learning before
425
00:13:27,600 --> 00:13:29,399
and like important minimized reaction I
426
00:13:29,399 --> 00:13:31,200
think it is very doable and I think it's
427
00:13:31,200 --> 00:13:33,120
it's you know and it's fun as well right
428
00:13:33,120 --> 00:13:34,500
there's nothing more satisfying to me
429
00:13:34,500 --> 00:13:36,540
than you've trained a network and it's
430
00:13:36,540 --> 00:13:38,160
just classifying really accurately
431
00:13:38,160 --> 00:13:39,959
whatever it was you wanted to do I
432
00:13:39,959 --> 00:13:41,279
basically my job is looking at numbers
433
00:13:41,279 --> 00:13:43,620
go up and I like when they go up so Mike
434
00:13:43,620 --> 00:13:45,480
I mean I'd love to come to you not in
435
00:13:45,480 --> 00:13:47,279
Nottingham University and attend your
436
00:13:47,279 --> 00:13:49,200
courses but obviously icon and so can
437
00:13:49,200 --> 00:13:51,240
you know most of us can't
438
00:13:51,240 --> 00:13:53,459
um do you have any like resources or
439
00:13:53,459 --> 00:13:56,399
ideas that things places I can go to to
440
00:13:56,399 --> 00:13:58,320
learn often the first course I recommend
441
00:13:58,320 --> 00:13:59,639
for everyone is to take Andrew Angus
442
00:13:59,639 --> 00:14:02,100
Corsair of course right very popular I
443
00:14:02,100 --> 00:14:03,420
mean I don't know how many times it's
444
00:14:03,420 --> 00:14:05,220
been taken now millions of times it's
445
00:14:05,220 --> 00:14:07,380
Andrew's course on machine learning
446
00:14:07,380 --> 00:14:09,000
there is a deep learning follow-up to it
447
00:14:09,000 --> 00:14:10,200
which I haven't I haven't done because
448
00:14:10,200 --> 00:14:11,940
partly I actually already know deep
449
00:14:11,940 --> 00:14:13,980
learning but um the machine learning
450
00:14:13,980 --> 00:14:15,660
course is really good it's a good
451
00:14:15,660 --> 00:14:17,820
understanding of some of the key
452
00:14:17,820 --> 00:14:19,800
Concepts in machine learning and not
453
00:14:19,800 --> 00:14:21,600
specifically about yes a little bit
454
00:14:21,600 --> 00:14:22,980
about how neural networks work and
455
00:14:22,980 --> 00:14:24,120
things like this and it can be a little
456
00:14:24,120 --> 00:14:25,920
bit mathematical it's my experience of
457
00:14:25,920 --> 00:14:28,139
it but if you if you watch it anyway
458
00:14:28,139 --> 00:14:29,880
you're going to pick up a lot of tips
459
00:14:29,880 --> 00:14:32,339
and tricks so things like watching your
460
00:14:32,339 --> 00:14:34,800
network train over time and reacting to
461
00:14:34,800 --> 00:14:36,480
how that works and doesn't work and
462
00:14:36,480 --> 00:14:38,399
making decisions based on this these are
463
00:14:38,399 --> 00:14:39,899
the things really I think that people
464
00:14:39,899 --> 00:14:41,519
who want to do machine learning in an
465
00:14:41,519 --> 00:14:44,399
applied way in a know in a business or
466
00:14:44,399 --> 00:14:46,380
in an industry that's what they need to
467
00:14:46,380 --> 00:14:47,760
better do a lot of them are not going to
468
00:14:47,760 --> 00:14:49,980
be writing neural networks from scratch
469
00:14:49,980 --> 00:14:52,139
or designing a number of layers in your
470
00:14:52,139 --> 00:14:53,639
network they're going to take a network
471
00:14:53,639 --> 00:14:55,380
that we know works and run it on some
472
00:14:55,380 --> 00:14:57,360
new data and if that works great the
473
00:14:57,360 --> 00:14:59,100
first time then that's fabulous but if
474
00:14:59,100 --> 00:15:01,500
it doesn't what do you do then and these
475
00:15:01,500 --> 00:15:02,519
are things that you're going to learn
476
00:15:02,519 --> 00:15:04,500
and start to learning that Coursera
477
00:15:04,500 --> 00:15:06,660
course Joshua bengio and others have
478
00:15:06,660 --> 00:15:08,639
written a book just called Deep learning
479
00:15:08,639 --> 00:15:10,740
which is very popular again obviously it
480
00:15:10,740 --> 00:15:12,720
can go into a little bit of heavy Mass
481
00:15:12,720 --> 00:15:15,300
detail but it's very popular I would say
482
00:15:15,300 --> 00:15:17,220
don't read it end to end it's one to dip
483
00:15:17,220 --> 00:15:18,779
into while you're doing some tutorials
484
00:15:18,779 --> 00:15:20,100
to understand a bit more about the
485
00:15:20,100 --> 00:15:21,839
theory and after that personally I would
486
00:15:21,839 --> 00:15:24,360
get I would do the pie torch tutorials
487
00:15:24,360 --> 00:15:26,639
or the psychic learn tutorials they can
488
00:15:26,639 --> 00:15:28,800
be directed at your own pace and they
489
00:15:28,800 --> 00:15:30,720
will include they'll give you experience
490
00:15:30,720 --> 00:15:32,220
and all those different things right
491
00:15:32,220 --> 00:15:33,839
there's there's tutorials on things like
492
00:15:33,839 --> 00:15:35,459
reinforcement learning but also just
493
00:15:35,459 --> 00:15:37,440
standard cnns and Transformers and
494
00:15:37,440 --> 00:15:39,240
things like this yeah and don't don't
495
00:15:39,240 --> 00:15:40,800
worry about you don't have to do all of
496
00:15:40,800 --> 00:15:42,660
those on day one on day one we're
497
00:15:42,660 --> 00:15:43,740
talking about about what is
498
00:15:43,740 --> 00:15:45,959
classification what is regression maybe
499
00:15:45,959 --> 00:15:48,000
get something little going right really
500
00:15:48,000 --> 00:15:49,680
you know start yourself off nice and
501
00:15:49,680 --> 00:15:51,360
slow and build up the complexity as we
502
00:15:51,360 --> 00:15:53,399
go right it's the same with any subject
503
00:15:53,399 --> 00:15:54,720
in computer science you can't learn
504
00:15:54,720 --> 00:15:56,160
everything on the first day so you just
505
00:15:56,160 --> 00:15:57,540
have to take it a little bit at a time
506
00:15:57,540 --> 00:15:59,699
I'm glad you mentioned computer science
507
00:15:59,699 --> 00:16:01,860
um do you think
508
00:16:01,860 --> 00:16:04,199
it's time to for more of us to learn
509
00:16:04,199 --> 00:16:06,300
computer science type stuff because of
510
00:16:06,300 --> 00:16:08,820
AI like maths and all these computer
511
00:16:08,820 --> 00:16:10,800
science stuff not really I think that
512
00:16:10,800 --> 00:16:12,480
it's it wouldn't it's not necessary for
513
00:16:12,480 --> 00:16:14,699
everyone to do that I think that you
514
00:16:14,699 --> 00:16:15,839
know I would encourage everyone to do
515
00:16:15,839 --> 00:16:17,639
computer science because I would but of
516
00:16:17,639 --> 00:16:19,920
course I I think that sometimes both
517
00:16:19,920 --> 00:16:22,440
computer science and an industry have a
518
00:16:22,440 --> 00:16:23,820
sort of reverse snobbery about each
519
00:16:23,820 --> 00:16:25,440
other right which I don't like very much
520
00:16:25,440 --> 00:16:26,880
so for example computer scientists might
521
00:16:26,880 --> 00:16:28,199
say well if someone didn't do a degree
522
00:16:28,199 --> 00:16:29,579
you know what do they really know about
523
00:16:29,579 --> 00:16:31,380
computers right which is not true and
524
00:16:31,380 --> 00:16:34,139
someone who's who got on fine without a
525
00:16:34,139 --> 00:16:35,820
degree might go why would I go and get
526
00:16:35,820 --> 00:16:38,459
student loans and do a degree and
527
00:16:38,459 --> 00:16:40,620
different paths are all valid I don't I
528
00:16:40,620 --> 00:16:41,399
don't know why we're having this
529
00:16:41,399 --> 00:16:43,800
conversation right and I think um there
530
00:16:43,800 --> 00:16:45,180
are there are elements of maths in
531
00:16:45,180 --> 00:16:48,300
machine learning which help I suppose me
532
00:16:48,300 --> 00:16:49,920
to understand it a bit better when
533
00:16:49,920 --> 00:16:51,300
someone comes with a particularly weird
534
00:16:51,300 --> 00:16:52,980
problem that doesn't you know they've
535
00:16:52,980 --> 00:16:54,120
added another layer and it's not
536
00:16:54,120 --> 00:16:55,980
training why is that right they also
537
00:16:55,980 --> 00:16:57,660
help me sometimes when I'm reading
538
00:16:57,660 --> 00:16:59,519
papers because papers they can have a
539
00:16:59,519 --> 00:17:01,139
lot of mathematical notation in and
540
00:17:01,139 --> 00:17:02,579
sometimes that's not necessary and
541
00:17:02,579 --> 00:17:04,319
they've just added it in but often it's
542
00:17:04,319 --> 00:17:06,119
just it's just to be absolutely clear
543
00:17:06,119 --> 00:17:07,859
about what they've done and and often
544
00:17:07,859 --> 00:17:09,419
the mathematical notation is necessary
545
00:17:09,419 --> 00:17:11,280
to achieve that rather than writing it
546
00:17:11,280 --> 00:17:13,619
in in sort of flavorful text but to
547
00:17:13,619 --> 00:17:15,359
begin with machine learning you don't
548
00:17:15,359 --> 00:17:16,740
necessarily need to know those things
549
00:17:16,740 --> 00:17:18,839
you know you can train a network in pi
550
00:17:18,839 --> 00:17:20,400
torch with a knowledge of rudimentary
551
00:17:20,400 --> 00:17:22,199
knowledge of python and following some
552
00:17:22,199 --> 00:17:24,299
tutorials and you'll pick up the rest as
553
00:17:24,299 --> 00:17:26,220
you go the really complicated maths like
554
00:17:26,220 --> 00:17:27,780
back propagation which is how we train
555
00:17:27,780 --> 00:17:29,880
it that's all taken care of under the
556
00:17:29,880 --> 00:17:31,559
hood you don't see that it's not
557
00:17:31,559 --> 00:17:32,640
something unless you're really
558
00:17:32,640 --> 00:17:33,960
interested it's not something to concern
559
00:17:33,960 --> 00:17:35,400
yourself with but I mean the great thing
560
00:17:35,400 --> 00:17:37,740
is if I'm in Industry I let or I'm into
561
00:17:37,740 --> 00:17:41,940
cyber or Dev or whatever I can really
562
00:17:41,940 --> 00:17:45,419
enhance my career prospects and the
563
00:17:45,419 --> 00:17:47,460
future by just adding this on to my
564
00:17:47,460 --> 00:17:49,500
skills yeah but and I also think that
565
00:17:49,500 --> 00:17:51,240
and I mentioned it before I think the
566
00:17:51,240 --> 00:17:53,160
other thing is it makes you much more
567
00:17:53,160 --> 00:17:56,160
resistant to hype and to concerns over
568
00:17:56,160 --> 00:17:57,780
things and also when someone comes to
569
00:17:57,780 --> 00:17:58,740
you and says oh yeah I've trained a
570
00:17:58,740 --> 00:18:00,600
neural network to do X Y and Z you can
571
00:18:00,600 --> 00:18:02,460
start to think hasn't sound very likely
572
00:18:02,460 --> 00:18:03,720
right that sounds like the sort of thing
573
00:18:03,720 --> 00:18:06,179
but maybe is a bit fanciful right let's
574
00:18:06,179 --> 00:18:08,220
let's deal with let's look at their data
575
00:18:08,220 --> 00:18:09,419
and see if that's actually true what
576
00:18:09,419 --> 00:18:10,620
they've done and I think one of the
577
00:18:10,620 --> 00:18:12,480
reasons that chat gbt and stuff have
578
00:18:12,480 --> 00:18:14,220
been so hyped recently is because most
579
00:18:14,220 --> 00:18:16,080
people don't know what it is and so when
580
00:18:16,080 --> 00:18:18,059
you see it doing what it does you think
581
00:18:18,059 --> 00:18:21,000
this thing must basically be a person
582
00:18:21,000 --> 00:18:23,280
right because it's acting like one but
583
00:18:23,280 --> 00:18:25,080
actually it's only acting like one in a
584
00:18:25,080 --> 00:18:27,480
very narrow thing and we know how it's
585
00:18:27,480 --> 00:18:29,580
trained and how it's trained doesn't
586
00:18:29,580 --> 00:18:32,220
imply necessarily that it's got any
587
00:18:32,220 --> 00:18:35,100
human qualities right it might but I
588
00:18:35,100 --> 00:18:37,740
don't gut tells me not quite right but
589
00:18:37,740 --> 00:18:39,600
the point is that I I can I'm sort of
590
00:18:39,600 --> 00:18:41,100
more resistant to that in some sense
591
00:18:41,100 --> 00:18:44,160
because I I know how it works underneath
592
00:18:44,160 --> 00:18:46,020
and I sort of think I've trained all
593
00:18:46,020 --> 00:18:47,400
these networks and this is a bigger
594
00:18:47,400 --> 00:18:48,720
version of networks that I've trained
595
00:18:48,720 --> 00:18:50,760
myself I don't see what's different
596
00:18:50,760 --> 00:18:52,500
about what's so different about that
597
00:18:52,500 --> 00:18:54,539
that it would suddenly be unbelievably
598
00:18:54,539 --> 00:18:56,160
Insurgent compared to anything else if
599
00:18:56,160 --> 00:18:57,720
that makes sense some knowledge of house
600
00:18:57,720 --> 00:18:59,940
what some of these Technologies are just
601
00:18:59,940 --> 00:19:02,600
like knowledge of you know so some some
602
00:19:02,600 --> 00:19:04,440
companies trying to sell you a new
603
00:19:04,440 --> 00:19:07,080
firewall with Next Generation antivirus
604
00:19:07,080 --> 00:19:08,520
on it that has all kinds of machine
605
00:19:08,520 --> 00:19:10,320
learning well if you understand a bit
606
00:19:10,320 --> 00:19:11,820
about machine learning you'll know what
607
00:19:11,820 --> 00:19:13,679
it will and won't do right and that will
608
00:19:13,679 --> 00:19:15,059
allow you to make a better informed
609
00:19:15,059 --> 00:19:16,679
purchase decision and the answer is it
610
00:19:16,679 --> 00:19:18,419
will work pretty well right but it's but
611
00:19:18,419 --> 00:19:19,919
nothing's perfect and machine learning
612
00:19:19,919 --> 00:19:21,480
is only as good as the training data and
613
00:19:21,480 --> 00:19:23,460
so on so there's lots of things you can
614
00:19:23,460 --> 00:19:24,840
ask and you can ask really difficult
615
00:19:24,840 --> 00:19:25,919
questions instead of people that come
616
00:19:25,919 --> 00:19:27,660
and try and sell it to you especially
617
00:19:27,660 --> 00:19:28,980
with things like Twitter and the news
618
00:19:28,980 --> 00:19:31,020
it's very easy to get carried away in
619
00:19:31,020 --> 00:19:32,280
this hype cycle right lots of
620
00:19:32,280 --> 00:19:33,840
technologies have this it's in the
621
00:19:33,840 --> 00:19:35,039
interest of these companies to make
622
00:19:35,039 --> 00:19:36,360
these massive models of incredibly
623
00:19:36,360 --> 00:19:39,120
impressive performance I think we're a
624
00:19:39,120 --> 00:19:40,980
long way from Full automation of a lot
625
00:19:40,980 --> 00:19:42,419
of these tasks even if it might appear
626
00:19:42,419 --> 00:19:44,880
that way a sort of superficial level but
627
00:19:44,880 --> 00:19:45,900
on the other hand they're really
628
00:19:45,900 --> 00:19:47,700
promising in some other ways right so
629
00:19:47,700 --> 00:19:49,380
one of the things that I found that chat
630
00:19:49,380 --> 00:19:52,080
GPT is really good at is paraphrasing
631
00:19:52,080 --> 00:19:54,360
text and vice versa so you have a text
632
00:19:54,360 --> 00:19:56,340
you don't quite understand say please
633
00:19:56,340 --> 00:19:57,840
can you read this and tell me what it
634
00:19:57,840 --> 00:19:59,940
means or please can you summarize these
635
00:19:59,940 --> 00:20:01,620
bullet points in an email or something
636
00:20:01,620 --> 00:20:03,480
like this you know these kind of
637
00:20:03,480 --> 00:20:04,740
functions I think are actually working
638
00:20:04,740 --> 00:20:07,020
really well right because those are
639
00:20:07,020 --> 00:20:09,960
functions that rely on the their text to
640
00:20:09,960 --> 00:20:11,760
text they're meant for text to text
641
00:20:11,760 --> 00:20:13,500
right they are that's kind of what
642
00:20:13,500 --> 00:20:14,820
they're for and I think that those are
643
00:20:14,820 --> 00:20:16,500
the ones that are really really good I
644
00:20:16,500 --> 00:20:18,179
think code completion is useful when
645
00:20:18,179 --> 00:20:20,400
you're asking limited things that you
646
00:20:20,400 --> 00:20:22,679
can carefully check quite quickly don't
647
00:20:22,679 --> 00:20:24,240
ask it to produce a thousand lines of
648
00:20:24,240 --> 00:20:25,740
code that you expect them to all be
649
00:20:25,740 --> 00:20:26,940
perfect because that's not what it will
650
00:20:26,940 --> 00:20:29,280
do right and and you'll end up with a
651
00:20:29,280 --> 00:20:30,840
lot of weird bugs or I mean there was
652
00:20:30,840 --> 00:20:32,160
this used paper just released just the
653
00:20:32,160 --> 00:20:33,539
other day actually from Stanford that
654
00:20:33,539 --> 00:20:36,059
said that they they audited code from
655
00:20:36,059 --> 00:20:39,000
about 30 to 35 researchers who some of
656
00:20:39,000 --> 00:20:40,260
them were using
657
00:20:40,260 --> 00:20:42,419
um AI to produce some of the code and
658
00:20:42,419 --> 00:20:44,880
some of them weren't and the AI produced
659
00:20:44,880 --> 00:20:47,160
code had more vulnerabilities in it and
660
00:20:47,160 --> 00:20:50,160
that's because when the AI produces code
661
00:20:50,160 --> 00:20:52,799
that works but let's say it uses ECB
662
00:20:52,799 --> 00:20:55,500
mode in as or it uses a slightly weak
663
00:20:55,500 --> 00:20:57,179
key derivation or something I don't know
664
00:20:57,179 --> 00:20:58,919
something subtle if they don't know
665
00:20:58,919 --> 00:21:00,780
about that subject already they might
666
00:21:00,780 --> 00:21:02,340
accept that change if that makes sense
667
00:21:02,340 --> 00:21:04,559
right they actually so you need this is
668
00:21:04,559 --> 00:21:06,000
why you need to still be an expert in
669
00:21:06,000 --> 00:21:08,340
your field because you can't just rely
670
00:21:08,340 --> 00:21:10,020
on it to do it for you yet you've got to
671
00:21:10,020 --> 00:21:12,360
be there saying I think that's okay or I
672
00:21:12,360 --> 00:21:13,679
don't think that's okay
673
00:21:13,679 --> 00:21:14,940
um and make those decisions for yourself
674
00:21:14,940 --> 00:21:16,919
yeah I mean you know it's a limited
675
00:21:16,919 --> 00:21:18,840
study but like it's not that limited and
676
00:21:18,840 --> 00:21:21,059
it makes a very valid point I think the
677
00:21:21,059 --> 00:21:23,460
real danger is people who and and I
678
00:21:23,460 --> 00:21:25,080
should carry out this by saying I'm not
679
00:21:25,080 --> 00:21:26,580
selling short these these incredible
680
00:21:26,580 --> 00:21:28,380
technology is I'm just saying that it
681
00:21:28,380 --> 00:21:29,880
would be very silly to just completely
682
00:21:29,880 --> 00:21:31,500
use them blind and never check what they
683
00:21:31,500 --> 00:21:32,940
do right because we know they just make
684
00:21:32,940 --> 00:21:34,740
stuff up a lot of the time I think a bit
685
00:21:34,740 --> 00:21:36,299
of domain knowledge is always going to
686
00:21:36,299 --> 00:21:37,620
help I mean it's interesting because I
687
00:21:37,620 --> 00:21:39,360
did I did some tests with like Cisco
688
00:21:39,360 --> 00:21:42,179
devices and um it's amazing like first
689
00:21:42,179 --> 00:21:44,280
time it got a perfect then I wanted to
690
00:21:44,280 --> 00:21:45,840
do it for a video and then it wasn't
691
00:21:45,840 --> 00:21:48,419
good and I did like five or six attempts
692
00:21:48,419 --> 00:21:50,400
and none of them were perfect yeah I
693
00:21:50,400 --> 00:21:52,140
think if I didn't know what it was doing
694
00:21:52,140 --> 00:21:53,940
I would have accepted it sorry go on
695
00:21:53,940 --> 00:21:55,740
yeah and the other thing is that you
696
00:21:55,740 --> 00:21:56,940
know if you think about a date of it
697
00:21:56,940 --> 00:21:59,220
it's trained on it's got some 40 40 plus
698
00:21:59,220 --> 00:22:01,559
billion tokens right it's just internet
699
00:22:01,559 --> 00:22:03,120
text we'll just leave it at that right
700
00:22:03,120 --> 00:22:05,520
loads and loads of text Cisco related
701
00:22:05,520 --> 00:22:07,559
text is only going to form a very very
702
00:22:07,559 --> 00:22:09,659
small fraction of that various vertical
703
00:22:09,659 --> 00:22:10,860
evidence because it's not got a world
704
00:22:10,860 --> 00:22:12,000
model because it's not got an
705
00:22:12,000 --> 00:22:13,679
understanding of the world where it can
706
00:22:13,679 --> 00:22:15,900
bring Cisco in and correct add it to its
707
00:22:15,900 --> 00:22:17,760
model it's just doing text completion
708
00:22:17,760 --> 00:22:19,200
and so when something is
709
00:22:19,200 --> 00:22:21,120
underrepresented in a training set it's
710
00:22:21,120 --> 00:22:23,400
gonna probably be worth performing when
711
00:22:23,400 --> 00:22:24,840
it comes to actually running it later
712
00:22:24,840 --> 00:22:27,419
right so when you say write me write me
713
00:22:27,419 --> 00:22:28,919
something in the style of Shakespeare
714
00:22:28,919 --> 00:22:30,659
it's going to do really well because
715
00:22:30,659 --> 00:22:31,980
there's Shakespeare all over the
716
00:22:31,980 --> 00:22:34,200
internet right some tasks are going to
717
00:22:34,200 --> 00:22:36,120
be very solvable because they've just
718
00:22:36,120 --> 00:22:37,380
they're hugely represented in the
719
00:22:37,380 --> 00:22:39,419
trading set they work really well and
720
00:22:39,419 --> 00:22:41,400
some tasks are really Niche and of
721
00:22:41,400 --> 00:22:42,299
course you don't know which one's a
722
00:22:42,299 --> 00:22:43,260
niche because you haven't seen the
723
00:22:43,260 --> 00:22:44,940
training set I say write me a link
724
00:22:44,940 --> 00:22:46,559
expression and it does it really well
725
00:22:46,559 --> 00:22:47,820
and when I say write me a link
726
00:22:47,820 --> 00:22:49,020
expression using some other thing and
727
00:22:49,020 --> 00:22:50,280
that isn't in a training set and it
728
00:22:50,280 --> 00:22:52,020
produces me a wrong answer and I don't
729
00:22:52,020 --> 00:22:53,580
know until I run it whether that's the
730
00:22:53,580 --> 00:22:55,980
case so I have to understand and be able
731
00:22:55,980 --> 00:22:58,020
to read that code because otherwise I
732
00:22:58,020 --> 00:22:59,820
can't possibly put it into my system and
733
00:22:59,820 --> 00:23:01,440
it's just it goes back to the exact same
734
00:23:01,440 --> 00:23:03,120
problem with medicine right it might be
735
00:23:03,120 --> 00:23:04,740
that we're absolutely confident but they
736
00:23:04,740 --> 00:23:06,419
say I will look at this image and make
737
00:23:06,419 --> 00:23:07,919
the correct decision but we're not
738
00:23:07,919 --> 00:23:09,480
absolutely sure and while we're not
739
00:23:09,480 --> 00:23:11,760
absolutely sure do we want to completely
740
00:23:11,760 --> 00:23:13,140
take a human out of the loop there
741
00:23:13,140 --> 00:23:14,640
there's questions to you know we have to
742
00:23:14,640 --> 00:23:16,020
think about so do you think it'll become
743
00:23:16,020 --> 00:23:18,179
like the AI might do a lot of the low
744
00:23:18,179 --> 00:23:20,419
level
745
00:23:21,320 --> 00:23:23,580
I think that's much closer to what will
746
00:23:23,580 --> 00:23:25,919
happen so I think in in there's a phrase
747
00:23:25,919 --> 00:23:27,780
in medicine called CAD or computer-aided
748
00:23:27,780 --> 00:23:29,820
diagnosis and the idea is that instead
749
00:23:29,820 --> 00:23:31,679
of the doctor not making a decision the
750
00:23:31,679 --> 00:23:33,360
doctor will be guided into a decision by
751
00:23:33,360 --> 00:23:35,640
the AI saying we've noticed these spots
752
00:23:35,640 --> 00:23:37,500
over here in this image is that relevant
753
00:23:37,500 --> 00:23:39,059
to you and it will speed them up right
754
00:23:39,059 --> 00:23:41,460
and if we can make doctors or Medics 50
755
00:23:41,460 --> 00:23:43,260
more efficient that's a huge that's
756
00:23:43,260 --> 00:23:45,059
that's a huge boost rather than try and
757
00:23:45,059 --> 00:23:46,919
put it all on the AI and similarly it
758
00:23:46,919 --> 00:23:48,360
works in code if you can produce
759
00:23:48,360 --> 00:23:49,980
boilerplate code if you can get it to
760
00:23:49,980 --> 00:23:52,260
bootstrap spring boots configuration
761
00:23:52,260 --> 00:23:54,960
files for you fabulous do that right and
762
00:23:54,960 --> 00:23:57,059
then that saves you half an hour to an
763
00:23:57,059 --> 00:23:59,039
hour of doing some actual code or making
764
00:23:59,039 --> 00:24:01,080
sure that it worked but what I would
765
00:24:01,080 --> 00:24:02,940
have avoid you know what I would avoid
766
00:24:02,940 --> 00:24:04,200
doing is trying to have it write
767
00:24:04,200 --> 00:24:05,940
everything for you and replace yourself
768
00:24:05,940 --> 00:24:07,919
because I don't think it will work I
769
00:24:07,919 --> 00:24:09,600
think you'll end up really frustrated
770
00:24:09,600 --> 00:24:11,460
that your code doesn't get part past any
771
00:24:11,460 --> 00:24:13,020
of your reviews because it didn't work
772
00:24:13,020 --> 00:24:15,179
right I was going to say I love what you
773
00:24:15,179 --> 00:24:17,640
said though because with that example at
774
00:24:17,640 --> 00:24:19,260
Stanford if if
775
00:24:19,260 --> 00:24:21,120
if people had just accepted the code
776
00:24:21,120 --> 00:24:23,880
there's hidden vulnerabilities in the
777
00:24:23,880 --> 00:24:25,740
code that wouldn't have been picked up
778
00:24:25,740 --> 00:24:27,840
yeah and and then there's a combination
779
00:24:27,840 --> 00:24:29,520
of issues right is it that the developer
780
00:24:29,520 --> 00:24:30,900
needs to know more about these subjects
781
00:24:30,900 --> 00:24:32,220
or is it that they're someone that would
782
00:24:32,220 --> 00:24:33,659
normally be on that team that wasn't
783
00:24:33,659 --> 00:24:34,980
auditing that code that would have been
784
00:24:34,980 --> 00:24:36,419
auditing that code at that time you know
785
00:24:36,419 --> 00:24:37,740
you know because you have security teams
786
00:24:37,740 --> 00:24:38,940
sometimes who are specialists in this
787
00:24:38,940 --> 00:24:40,799
but I think it's that same argument in
788
00:24:40,799 --> 00:24:42,480
some ways if someone has a small amount
789
00:24:42,480 --> 00:24:45,059
of knowledge of computer security that
790
00:24:45,059 --> 00:24:46,500
might allow them to be more resistant
791
00:24:46,500 --> 00:24:48,299
when code appears it does this and
792
00:24:48,299 --> 00:24:49,620
that's the same thing with the AI if you
793
00:24:49,620 --> 00:24:51,360
know a little bit about AI maybe you can
794
00:24:51,360 --> 00:24:52,740
be you can better deal with it when
795
00:24:52,740 --> 00:24:54,000
something comes along so I think a
796
00:24:54,000 --> 00:24:55,260
little bit of knowledge in lots of these
797
00:24:55,260 --> 00:24:57,240
things is is often useful for that
798
00:24:57,240 --> 00:24:59,280
reason micro so how's this affected like
799
00:24:59,280 --> 00:25:00,780
University life because I've heard
800
00:25:00,780 --> 00:25:03,419
people talk about how students can just
801
00:25:03,419 --> 00:25:06,120
get check GPD to write their essays and
802
00:25:06,120 --> 00:25:07,860
stuff like that so and you can't you
803
00:25:07,860 --> 00:25:08,880
can't see the difference between a
804
00:25:08,880 --> 00:25:10,559
student and a like a human sorry and
805
00:25:10,559 --> 00:25:11,940
yeah
806
00:25:11,940 --> 00:25:13,799
um I think it's very subject dependent I
807
00:25:13,799 --> 00:25:15,960
think that's one thing so um what we've
808
00:25:15,960 --> 00:25:17,640
done is we've done we've actually been
809
00:25:17,640 --> 00:25:19,919
running some tests right because so you
810
00:25:19,919 --> 00:25:21,480
know if I kind of open out and drop this
811
00:25:21,480 --> 00:25:23,520
tournament just before exams yeah yeah
812
00:25:23,520 --> 00:25:25,440
exactly
813
00:25:25,440 --> 00:25:27,179
um yeah we've run some tests and like I
814
00:25:27,179 --> 00:25:29,460
think it depends on if if I show suppose
815
00:25:29,460 --> 00:25:30,779
we're doing a computer security exam
816
00:25:30,779 --> 00:25:33,120
which actually I I teach so you know
817
00:25:33,120 --> 00:25:35,340
right and I ask a very simple question
818
00:25:35,340 --> 00:25:37,260
right a question like what's a good
819
00:25:37,260 --> 00:25:39,360
encryption algorithm to use track GPT
820
00:25:39,360 --> 00:25:41,400
can answer that so it'll be unwise of me
821
00:25:41,400 --> 00:25:43,140
to ask that question in an exam I
822
00:25:43,140 --> 00:25:44,940
suppose what we say in some sense I
823
00:25:44,940 --> 00:25:46,799
think it's another variant of a search
824
00:25:46,799 --> 00:25:48,539
engine so if a student could you know
825
00:25:48,539 --> 00:25:50,220
can we call it academic misconduct right
826
00:25:50,220 --> 00:25:51,360
if a student was going to use a search
827
00:25:51,360 --> 00:25:53,039
engine to do that they could also have a
828
00:25:53,039 --> 00:25:55,080
go at using chat GPT it has the
829
00:25:55,080 --> 00:25:56,640
advantage for that student but it's
830
00:25:56,640 --> 00:25:58,020
generating very plausible looking
831
00:25:58,020 --> 00:26:00,120
answers sometimes they're completely
832
00:26:00,120 --> 00:26:01,980
wrong right and those answers are going
833
00:26:01,980 --> 00:26:03,659
to get marked very far down when they
834
00:26:03,659 --> 00:26:05,220
come in front of of a of a convenience
835
00:26:05,220 --> 00:26:07,500
so I think your mileage may vary if you
836
00:26:07,500 --> 00:26:08,700
think you can get through a University
837
00:26:08,700 --> 00:26:11,520
degree using just AI tools it's
838
00:26:11,520 --> 00:26:12,960
something we have to consider right now
839
00:26:12,960 --> 00:26:14,760
some of our exams are face to face they
840
00:26:14,760 --> 00:26:16,080
aren't really affected right you know
841
00:26:16,080 --> 00:26:17,460
we're talking about coursework essays
842
00:26:17,460 --> 00:26:19,080
and we I don't know I haven't spoken too
843
00:26:19,080 --> 00:26:20,940
much to other other schools in the
844
00:26:20,940 --> 00:26:22,799
university in other subject areas but
845
00:26:22,799 --> 00:26:24,360
obviously there are lots of essay-based
846
00:26:24,360 --> 00:26:27,179
subjects but they require very well
847
00:26:27,179 --> 00:26:29,520
written essays trap GPT has a habit of
848
00:26:29,520 --> 00:26:31,679
producing general answers to things
849
00:26:31,679 --> 00:26:33,539
right which are sometimes very detailed
850
00:26:33,539 --> 00:26:35,039
but sometimes not quite so detailed
851
00:26:35,039 --> 00:26:37,260
again I think that your mileage would
852
00:26:37,260 --> 00:26:39,419
vary if you tried this I suspect that it
853
00:26:39,419 --> 00:26:41,279
is possible to tell that they're written
854
00:26:41,279 --> 00:26:43,919
by chat GPT to an extent because it has
855
00:26:43,919 --> 00:26:45,600
a way of phrasing things that's quite
856
00:26:45,600 --> 00:26:47,460
common I've noticed as I as I produce
857
00:26:47,460 --> 00:26:49,919
answers but that isn't sure that isn't
858
00:26:49,919 --> 00:26:51,900
necessarily all the time but that's
859
00:26:51,900 --> 00:26:53,400
going to be a problem it's something
860
00:26:53,400 --> 00:26:55,620
that every University on Earth is now
861
00:26:55,620 --> 00:26:58,559
looking at work so yeah it's had a big
862
00:26:58,559 --> 00:27:00,360
impact and you know when you consider
863
00:27:00,360 --> 00:27:02,520
that this is just version one right and
864
00:27:02,520 --> 00:27:04,500
you know there's going to be a chat gp22
865
00:27:04,500 --> 00:27:06,960
probably and Microsoft might release one
866
00:27:06,960 --> 00:27:09,059
and Google release one and so on and so
867
00:27:09,059 --> 00:27:10,320
forth there's gonna there's always going
868
00:27:10,320 --> 00:27:11,580
to be one of these tools floating about
869
00:27:11,580 --> 00:27:13,500
that we have to just be prepared and
870
00:27:13,500 --> 00:27:15,240
think about how that how that's going to
871
00:27:15,240 --> 00:27:16,740
work I think I mean the examples I've
872
00:27:16,740 --> 00:27:18,120
seen which have worked really well is
873
00:27:18,120 --> 00:27:20,220
like if I'm asked to write an essay
874
00:27:20,220 --> 00:27:22,679
about something I can get it to write
875
00:27:22,679 --> 00:27:24,000
something that gives me a lot of ideas
876
00:27:24,000 --> 00:27:25,860
and then I can just rephrase it in my
877
00:27:25,860 --> 00:27:28,500
own voice but it's it helps you a lot
878
00:27:28,500 --> 00:27:30,600
from a study point of view yeah I think
879
00:27:30,600 --> 00:27:31,919
actually does and I think so anyway
880
00:27:31,919 --> 00:27:33,419
that's a that's a big positive way and
881
00:27:33,419 --> 00:27:35,460
there are some academics in in this
882
00:27:35,460 --> 00:27:37,679
school for example and across across the
883
00:27:37,679 --> 00:27:39,720
world who operate in a kind of human
884
00:27:39,720 --> 00:27:42,539
computer interaction area who are very
885
00:27:42,539 --> 00:27:44,580
interested in could you end up writing a
886
00:27:44,580 --> 00:27:46,440
better essay if you worked with a
887
00:27:46,440 --> 00:27:48,000
computer to help you out right and in in
888
00:27:48,000 --> 00:27:50,100
a way is that not a win for the
889
00:27:50,100 --> 00:27:51,960
lecturers as well if that's the case now
890
00:27:51,960 --> 00:27:53,760
I I agree with that to an extent I think
891
00:27:53,760 --> 00:27:55,260
that's absolutely right I think that
892
00:27:55,260 --> 00:27:56,640
maybe we can't solve that whole
893
00:27:56,640 --> 00:27:58,440
discussion in a month right which is how
894
00:27:58,440 --> 00:28:00,480
long it is until our exams so you know
895
00:28:00,480 --> 00:28:02,760
the clock is ticking in somewhat it's
896
00:28:02,760 --> 00:28:04,919
somewhat in the short term for for these
897
00:28:04,919 --> 00:28:06,720
issues but in the longer term I think
898
00:28:06,720 --> 00:28:07,440
they're going to be really
899
00:28:07,440 --> 00:28:09,539
transformative in helping you know there
900
00:28:09,539 --> 00:28:11,460
are students who have who are very very
901
00:28:11,460 --> 00:28:12,659
intelligent and they know all the
902
00:28:12,659 --> 00:28:13,919
subject area but they're just not good
903
00:28:13,919 --> 00:28:15,600
at exams they really struggle to get
904
00:28:15,600 --> 00:28:17,580
their thoughts down on paper maybe those
905
00:28:17,580 --> 00:28:19,020
students could really be helped by
906
00:28:19,020 --> 00:28:20,520
something like this because if you give
907
00:28:20,520 --> 00:28:23,340
really specific prompts to chat GPT you
908
00:28:23,340 --> 00:28:24,779
get much better answers if a student
909
00:28:24,779 --> 00:28:26,340
knows what they're doing and can work
910
00:28:26,340 --> 00:28:28,320
with the AI I think that's going to be
911
00:28:28,320 --> 00:28:30,659
much better I mean I suppose the you
912
00:28:30,659 --> 00:28:31,740
could have said the same thing for
913
00:28:31,740 --> 00:28:34,020
Google or you know using search engines
914
00:28:34,020 --> 00:28:36,539
for yeah yeah that's that's the point
915
00:28:36,539 --> 00:28:38,520
that's been made I mean in some ways I
916
00:28:38,520 --> 00:28:39,960
see on Twitter a lot of people um
917
00:28:39,960 --> 00:28:41,640
compare these things to Google I would
918
00:28:41,640 --> 00:28:43,799
not because they're very different
919
00:28:43,799 --> 00:28:45,059
um and they don't have no source of
920
00:28:45,059 --> 00:28:46,559
actual data right that's the really
921
00:28:46,559 --> 00:28:48,539
important thing to remember but they
922
00:28:48,539 --> 00:28:50,640
they are a complementary tool in many
923
00:28:50,640 --> 00:28:52,020
ways and they are they operate in a
924
00:28:52,020 --> 00:28:53,400
similar way if you were going to try and
925
00:28:53,400 --> 00:28:55,260
answer an exam you know you would put
926
00:28:55,260 --> 00:28:56,940
the question in you'd rephrase it you'd
927
00:28:56,940 --> 00:28:58,740
see what came out you'd see does that
928
00:28:58,740 --> 00:29:00,000
look plausible I'm going to try again
929
00:29:00,000 --> 00:29:01,740
I'm going to edit it and so on in the
930
00:29:01,740 --> 00:29:02,700
same way that you would if you were
931
00:29:02,700 --> 00:29:04,140
doing if you're using a search engine to
932
00:29:04,140 --> 00:29:06,059
write an essay as well and right using a
933
00:29:06,059 --> 00:29:07,559
search engine to write an essay and I
934
00:29:07,559 --> 00:29:08,700
don't want to speak for every academic
935
00:29:08,700 --> 00:29:10,679
on the planet right but it's not
936
00:29:10,679 --> 00:29:12,960
necessarily plagiarism or misconduct it
937
00:29:12,960 --> 00:29:14,460
depends on how you use it right you know
938
00:29:14,460 --> 00:29:16,200
looking up sources online is absolutely
939
00:29:16,200 --> 00:29:18,960
to be encouraged it depends on how
940
00:29:18,960 --> 00:29:20,700
you're doing this I think in the long
941
00:29:20,700 --> 00:29:22,500
term we will get a nice balance actually
942
00:29:22,500 --> 00:29:25,320
between using it too much and not using
943
00:29:25,320 --> 00:29:26,460
it enough and I think actually there's
944
00:29:26,460 --> 00:29:27,779
another thing there's another aspect
945
00:29:27,779 --> 00:29:29,279
which is I think this plays into your
946
00:29:29,279 --> 00:29:31,320
this is relevant to your channels
947
00:29:31,320 --> 00:29:34,080
viewers is that you don't you shouldn't
948
00:29:34,080 --> 00:29:35,520
think of doing a degree or writing a
949
00:29:35,520 --> 00:29:37,260
coursework as just about getting a mark
950
00:29:37,260 --> 00:29:38,820
right that's very easy to think about
951
00:29:38,820 --> 00:29:40,740
that but actually it's about learning
952
00:29:40,740 --> 00:29:42,240
something that you can then take and use
953
00:29:42,240 --> 00:29:43,559
in your career or something like that
954
00:29:43,559 --> 00:29:44,820
right we don't teach people to program
955
00:29:44,820 --> 00:29:46,740
so they pass the exams we teach them the
956
00:29:46,740 --> 00:29:47,940
program so they can go off and be
957
00:29:47,940 --> 00:29:49,500
software developers if you use AI to
958
00:29:49,500 --> 00:29:51,179
write all of your work for you then you
959
00:29:51,179 --> 00:29:52,260
get out you wouldn't be able to get a
960
00:29:52,260 --> 00:29:53,460
job and you wouldn't better work in that
961
00:29:53,460 --> 00:29:55,020
job because you wouldn't be able to do
962
00:29:55,020 --> 00:29:56,700
any of the computer science actually I
963
00:29:56,700 --> 00:29:58,500
think that if you got a lot because I
964
00:29:58,500 --> 00:29:59,820
have quite a lot of learning and I love
965
00:29:59,820 --> 00:30:01,140
to learn about new soft topics
966
00:30:01,140 --> 00:30:02,460
particularly you know around computer
967
00:30:02,460 --> 00:30:04,799
science I would never use chat GPT to
968
00:30:04,799 --> 00:30:06,779
cheat because I wouldn't know any of it
969
00:30:06,779 --> 00:30:08,880
then right and you know and I like to
970
00:30:08,880 --> 00:30:10,440
learn about these things now if you want
971
00:30:10,440 --> 00:30:11,760
to become an expert in something then
972
00:30:11,760 --> 00:30:13,260
you're going to need to learn it you
973
00:30:13,260 --> 00:30:16,080
can't read what chat GPT wrote A lot of
974
00:30:16,080 --> 00:30:17,580
it comes down to hoping that students
975
00:30:17,580 --> 00:30:18,840
and trying to encourage students to
976
00:30:18,840 --> 00:30:20,159
think that it's about the process of
977
00:30:20,159 --> 00:30:22,020
learning and where they get to at the
978
00:30:22,020 --> 00:30:23,940
end rather than specifically about a
979
00:30:23,940 --> 00:30:26,760
series of of kind of barriers of exams
980
00:30:26,760 --> 00:30:28,260
that they have to get through right
981
00:30:28,260 --> 00:30:30,240
which I think is is not a good way to
982
00:30:30,240 --> 00:30:31,980
look at a degree or any course really
983
00:30:31,980 --> 00:30:33,659
you know it's much better to think about
984
00:30:33,659 --> 00:30:35,640
where you'll be at the end right and
985
00:30:35,640 --> 00:30:36,779
you'll be in that much better position
986
00:30:36,779 --> 00:30:38,520
to do what you want to do next that's
987
00:30:38,520 --> 00:30:39,899
exactly right I mean it's a like
988
00:30:39,899 --> 00:30:41,940
certification exams same thing you know
989
00:30:41,940 --> 00:30:43,020
you can go and get all the answers or
990
00:30:43,020 --> 00:30:44,520
the cheetah sites or you can actually
991
00:30:44,520 --> 00:30:46,200
learn something and and you haven't done
992
00:30:46,200 --> 00:30:47,760
yourself any favors if you get it off
993
00:30:47,760 --> 00:30:49,020
because you might get a job based on
994
00:30:49,020 --> 00:30:50,279
that it's not going to go well right
995
00:30:50,279 --> 00:30:52,080
because you don't you don't have any of
996
00:30:52,080 --> 00:30:53,399
the knowledge you'll always feel like
997
00:30:53,399 --> 00:30:54,539
you don't have any of the knowledge as
998
00:30:54,539 --> 00:30:56,100
well right you know actually it doesn't
999
00:30:56,100 --> 00:30:58,260
take that long to learn things if you
1000
00:30:58,260 --> 00:31:00,120
really put yourself to it and you'll be
1001
00:31:00,120 --> 00:31:01,020
in such a much better position
1002
00:31:01,020 --> 00:31:03,360
afterwards Mike as always I really want
1003
00:31:03,360 --> 00:31:04,919
to thank you for you know sharing your
1004
00:31:04,919 --> 00:31:07,380
knowledge and you know separating the
1005
00:31:07,380 --> 00:31:09,899
hype from like the worries about
1006
00:31:09,899 --> 00:31:12,000
people's futures thanks so much for
1007
00:31:12,000 --> 00:31:14,039
making a drill yeah it's no problem glad
1008
00:31:14,039 --> 00:31:15,120
to be on again it's been really really
1009
00:31:15,120 --> 00:31:18,230
good brilliant thanks Mike thanks
1010
00:31:18,230 --> 00:31:21,309
[Music]
73571
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.