YT YouTube2Text

YouTube2Text → Transcripts → [CS61C FA20] Weekly Lecture 03.LIVE - Floating Point & RISC-V

[CS61C FA20] Weekly Lecture 03.LIVE - Floating Point & RISC-V — Transcript

by CS 61C Departmental · 10,271 words · 1,669 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00perfect perfect thank you yeah perfect
0:02hey welcome everybody
0:04welcome everybody 61c
0:07still yeah let's start with a different
0:11class yeah
0:12all right a few things uh
0:16that we wanted to discuss today um and
0:18we're gonna go
0:19relatively quickly into that
0:22look my control works this uh you know
0:24that we're using this app
0:26and it's working on the ever release
0:28every other week and it works
0:30better and better so by the end of the
0:31semester we will be
0:34spectacular we might be out of beta we
0:37might actually be a
0:38be out of beta but that first time go
0:40ahead yeah
0:42anyways so we are in week three now uh
0:45what is happening is um we have
0:49covered floating point it's uh very fun
0:51so uh dan apparently had a lot of fun
0:53recording that lecture yeah it was great
0:55it was wonderful
0:56it took it took three days of my weekend
0:58to do one but one hour of lecture
1:00it normally would have taken me 50
1:02minutes to give the lecture i would have
1:03grabbed my slides
1:04give it three solid days to make the
1:06lecture so i hope you folks appreciate
1:07it but it's a lot of fun it's a lot of
1:08fun for these
1:09archival materials together and i want
1:11to check in to see if
1:13all of you have enjoyed the new clicker
1:16questions you know we made some mistake
1:17the first time we had too many we had 11
1:19and there was some confusion so we've
1:21reduced them down to hopefully being a
1:23much more mandible set of questions and
1:25each question the goal is to teach you
1:28something not
1:29did you watch minute 13 but did you
1:31teach you something so they're kind of
1:32actually in a way of writing exam
1:34questions into clicker questions so
1:36in a way we're trying to do this to
1:38again receive that feedback and always
1:39try to improve so thank you for that
1:41good feedback folks yeah we don't have
1:43to
1:44have those questions you know what size
1:46shoes does that
1:47mean on my third
1:50juggling did he drop it did he wink did
1:53it no none of that's going on so that's
1:54right
1:54yeah and
1:58risk five is uh in uh the first part a
2:02very simple introduction first five is
2:04in i managed to record the next
2:07lecture worked already so i'll try to
2:09post that relatively early along with um
2:13a bunch of risk five stuff by the
2:15weekend i know
2:16you're probably focused on quests at the
2:19moment but be
2:20really important to follow everything uh
2:23that is going on with with just five
2:24because we
2:25really need to get going with those labs
2:27next week
2:28yeah and we'll and we'll we'll this this
2:30is kind of a a switchover week this it
2:32is important that this
2:33this week's material is not covered on
2:36so we're not covering float or risk five
2:38for the quest this weekend
2:40but we're certainly going to be needing
2:41it very soon because there's a project
2:43coming up and there's more labs and
2:44projects after that so we'll do that as
2:47well and we are recording this we didn't
2:48record it last week and we got that
2:50record button
2:51we smashed the record button as the kids
2:52say so we've got that rolling for that
2:54um and again thank you for for all the
2:57support and all the great suggestions on
2:58piazza for how to improve
3:00good job down on smashing that record
3:03button
3:04so going into the you know jumping into
3:06the news
3:08there is here is us a chip that got a
3:11lot of press over the summer it was
3:12announced i think
3:13uh in june and then a little more
3:17details came out in a conference that
3:18was held
3:19in uh august it is um
3:23uh nvidia's new gpu but this
3:26you know they call it a graphics
3:28processing unit but this is
3:30pretty much everything other than a
3:32graphics
3:34computing unit so it is a
3:37domain specific machine
3:41however that accelerates multiple
3:43domains where we care about
3:46linear algebra so it's a very expensive
3:48uh
3:50chip it is a a big die and they do want
3:54to get
3:55i mean it's a foundation of you know
3:56multiple different products
3:58that nvidia will be offering
4:01they are taunting about its uh
4:05performance um and here are a few things
4:08that they're talking about
4:10um they're talking about their
4:14floating point performance
4:17that's why we picked this news for this
4:19week and what is interesting
4:21there they've introduced a brand new
4:24floating point
4:25format that has that is
4:29tf-32 now um dan has mentioned
4:32in the lecture that there are few that
4:35there is ie754 standard
4:38there are few other
4:41different standards that people who did
4:43not think uh
4:44floating point uh standard from is going
4:47to
4:48serve them well so they invented their
4:50own so we got b float
4:5116 that more than just google has
4:55adopted it is now adopted by intel's
4:58processors
5:00but nvidia has pushed something
5:02completely new
5:03they got tf32 someone would think that
5:07some kind of a 32 point floating point
5:09standard
5:10but no it's a 19-bit
5:13floating point standard and you're
5:16thinking 19 bits
5:18yeah and you start thinking about you
5:20know british machines of the 70s with
5:22you know i got a 35-bit int how about
5:26that anybody want a 35-bit int i mean
5:28what are we up to here it's ridiculous
5:31yeah they have good reasons it's worth
5:33reading about their blog posts why
5:35they want to do this i mean it basically
5:37took two parts
5:38from the you know appropriate mantissas
5:41and
5:42exponents from b float and
5:45and fp16 or from
5:48um and and and made it smaller so
5:51they're trying
5:52to save compute power but they're
5:53wasting uh quite a bit of
5:56data storage over there and that's kind
5:58of important
5:59we'll see how this plays out um this is
6:01not the first one
6:03there was a flex point that nirvana
6:05introduced a few years ago and that
6:07seems to be gone i mean it lasted for
6:08about a year they had it launched
6:10in one generation of products and gave
6:13up on that
6:14but nvidia claims spectacular
6:16performance in this
6:17um i am going to flip through a couple
6:20of these slides
6:22um and you know one of the first things
6:24that they're showing they're not talking
6:25about graphics here they're
6:27talking about accelerating high
6:28performance computing
6:30and they're showing um best in class
6:32numbers that are
6:35oh look at this is fun uh these bars are
6:37green in the presentation but what has
6:39happened
6:40is it's your it's the background
6:44this is the campanile technology
6:49there is a berkeley there is a berkeley
6:51benchmark on the bottom oh
6:53it's beautiful well what's better than
6:55the campanile for brentwood brooklyn
6:57benchmark
6:57more playful um yeah so
7:00and one of the big things that you know
7:02what sells a lot of these ambiguous
7:04chips is uh
7:04machine learning acceleration so
7:06accelerating different kinds of machine
7:08working
7:08machine learning workloads and what
7:11they've shown
7:12here is that they got some really nice
7:14performance for a single chip for a
7:16single gpu
7:17you know tends to beat everything else
7:20why they talk about that
7:21because these pods that google uses for
7:23their tensor processing units that we
7:25have shown in the first lecture
7:27do still beat collections of gpus by uh
7:31by a wide margin a few things that we
7:35have seen
7:35before um is
7:40um so i think we have shown this in the
7:42lecture and better get out of this
7:44of this slide over here
7:48so we have uh fp32
7:51which is an uh ieee standard
7:54we have uh fp16 which is a half
7:56precision half-precision ieee standard
7:58and then there are these two other
8:00things that have
8:01been showing up in uh particularly
8:04machine learning so google is using b
8:07float 16
8:08because they do want more of an exponent
8:10they want more of a range
8:12and they don't care that much about
8:13mantissa or significant
8:15and then now there is tf32 which
8:18basically
8:19plucks the exponent from b float or
8:22from fb16 and matches it with the
8:26you know mantissa from fp 16
8:29and puts everything together ends up in
8:3319 bits
8:34and just rolls on now when we look
8:38at a bunch of hardware that is out there
8:41that
8:41is used for different domains
8:44we end up with quite a bit of variety
8:48and this is going to make
8:49everybody's life real fun so if you
8:52target one of these
8:54how are you going to retarget your
8:56application to target something else
8:58if something is running with b float 16
9:02will it run with
9:03tf32 who knows
9:07i think a lot of us are going to have a
9:09lot of fun over the time again
9:11you know some people will think about
9:13you know we'll see that as job security
9:15i will keep
9:16all of this complete class this whole
9:19class and everyone else
9:20busy figuring this out now we are not
9:23neither dan nor i are experts in this we
9:25know a little bit but we do have in our
9:27department we do have a
9:28an expert so our
9:32you know not just a resident expert but
9:34a
9:35global expert is professor jim demille
9:39he has been in berkeley for a while he
9:42is a professor of mathematics and
9:44computer science
9:45and he was our former chair so he was a
9:48chair
9:49until a year ago but he also
9:51participated in the revision of 754
9:54standard
9:55and he knows a lot about numbers
9:59so jim do we hear you
10:02i hope so can you hear me absolutely
10:05thank you
10:06welcome jim for coming this is so great
10:08to have you here it's fun to be here
10:10tell stories about floating point
10:15very good so can you tell us a little
10:18bit about how you ended up in berkeley
10:20where where you came from how you became
10:21a
10:22computer scientist and mathematician at
10:24the same time and how did you
10:25end up loving floating point well uh
10:28so a brief history as i graduated from
10:31caltech
10:32when i was 19 and then i spent three
10:35years working at a startup and living on
10:37venice beach and doing a lot of body
10:39surfing and bicycling
10:41bicycling across the country and then i
10:43decided it was time to go back to grad
10:44school
10:45and i only applied to berkeley life was
10:47different back then
10:49and so i started as a grad student in
10:51the math department
10:52and professor william khan who i'll tell
10:54you more about was my advisor
10:56and he suggested that i move to computer
10:59science
11:00which is why i eventually got my phd
11:02here in 1983
11:04and just as professor khan who's retired
11:06now
11:07uh never you know was a joint faculty
11:10member in both computer science and
11:11mathematics
11:13when after i graduated i went and was a
11:15faculty member at nyu
11:16for a few years and then i came back to
11:18berkeley and i'm
11:20following his path and i'm also in both
11:22departments
11:25but while i was his student he was
11:28had established this ieee floating point
11:30standards committee
11:31because the community had decided for
11:33very good reasons
11:34that it was just too painful to have
11:36every uh company building computers
11:39have different rules for how floating
11:40point work and that's because if
11:42everybody does arithmetic slightly
11:44differently it makes
11:45writing portable code or debugging
11:47really really hard
11:49and so the time was right to have a
11:51standard so everybody
11:52could agree and that he was the world's
11:54expert
11:55and established that standard and shared
11:57it
11:58and uh i got to work on it while i was a
12:00grad student
12:01and write a bunch of papers on that as
12:03well as doing a bunch of linear algebra
12:05work
12:06and then after it became a standard in
12:09uh
12:10in 85 that he won the touring award
12:12that's how much recognition he got so
12:14the nobel prize in computer science in
12:151989
12:17and so uh while i was away
12:21in new york city i also became part of
12:23the team
12:24that produces all the basic linear
12:26algebra software that people use
12:27l apac and so it was a great delight to
12:30have
12:31be able to rely on one common arithmetic
12:33standard to write all these algorithms
12:35that everybody uses
12:37sounds fun and you're our department
12:40chair so how do you become a department
12:41chair is that like
12:43a rewarder of punishment so for many
12:45years my excuse was
12:47i can't be chair uh because i'm in two
12:49departments i owe half my teaching to
12:51both departments how you know how could
12:53i be chair
12:54but then that excuse broke down when
12:56mike jordan
12:57another colleague who's in joint two
12:58departments agreed to be the chair of
13:00statistics so i said okay
13:01that excuse no longer works and uh
13:04so then i became chair of each and it
13:07was a an amazing time to be chair
13:09there was just this you know extremely
13:11rapid growth
13:13um we ended up you know during my tenure
13:15with having the two largest depart
13:17majors on campus both l and scs and eeks
13:20and we added a third major data science
13:23which now has 900 majors signed up so
13:25maybe we'll have the three largest
13:26majors on campus
13:28and fortunately the you know the campus
13:31you know doesn't have unbounded
13:33resources as we all know
13:34but they were very generous and in terms
13:37of letting us hire
13:38and i had the pleasure of hiring 14 new
13:40faculty
13:41while i was chair so 10 in computer
13:44science
13:44and 4 in double e and i'm happy to say
13:47it's the most diverse crowd ever it was
13:49uh
13:49half men and half women and so
13:52a lot of those folks are you know have
13:55joined us just recently
13:56and so uh they're going to be making you
13:58you'll be seeing a lot of them in
13:59classrooms or in zoom rooms as you like
14:02and on top of all that uh we created we
14:05decided that the organization of
14:07the campus was not sufficient to deal
14:09with all the changes in the world
14:11and so uh we created new division of
14:14data science
14:15and you know we argued about the name
14:16for a long time that's what faculty do
14:18but
14:18if the official name is now the division
14:20of computing
14:21data science and society and uh jennifer
14:25uh chase from formerly microsoft is the
14:28new uh
14:29dean or associate provost of that
14:31organization
14:33and uh so we're reorganizing all sorts
14:35of things
14:36in the curriculum and research and
14:38fundraising and a new building
14:40to go around and that
14:44glad that uh that john kenny and jeff
14:46bochar have uh taken the reins to
14:48make all this work excellent and it
14:51looks like you managed to step down
14:53on time right before uh
14:56we got into into all of this what we are
14:59dealing right now
15:00yes um and and i got to go back to
15:02bicycling too which was nice so
15:04that's excellent news um
15:08so you're also a part of a four
15:11uh rather forces still so can you tell
15:14us what what's new with the
15:16in the 2019 so my first
15:19opportunity to contribute to that was
15:21when i was a grad student and
15:22the way the ieee standards work is the
15:24rule is every 10 years every standard
15:27has to be revisited to make sure it's
15:29still
15:29technically up to date and so the
15:33last revision was approved in 2019 and
15:36the chair of the committee turned out
15:38was another student of professor khan's
15:40uh david huff
15:41and i was a member and you know several
15:44a couple of my students and a lot of
15:45people from many industries in academia
15:47were in it
15:48and so after this many years what was
15:51worth changing
15:52and there's always a few bug fixes
15:54amazingly even after
15:5620 or 30 years so so here's one bug fix
15:59it turns out that the way min and max
16:01were defined
16:03were not right for everybody why is that
16:06i hope
16:06that dan told people about nan's yeah
16:08when he did i certainly did i how could
16:11you not
16:11yeah okay so what is the max of
16:14man and three uh it turns out that not
16:18everybody agrees on what the definition
16:20should be
16:20so now there are two definitions of max
16:22and men wow
16:23and they're both uh associative because
16:26it's really important to get the same
16:27answer no matter what order you do it in
16:29and one of them always propagates a nand
16:32so nan is treated as any possible number
16:35and in the other version of min and max
16:37it's treated as missing data
16:39so it's always passed over so
16:43and that's because that's what people
16:44wanted but the most interesting but so
16:47so they're little bug fixes but there's
16:48one
16:48important new mathematical instruction
16:51that we added
16:52and the motivation for this was
16:54reproducibility
16:56and and the fact that uh floating point
16:59and
16:59dan i hope told you about this is not
17:01associative if you add three numbers in
17:03different orders you can get different
17:04answers
17:05because of round off we showed that and
17:08so
17:08this has become and so it used to be you
17:10know you you type eight out out twice in
17:13a row
17:14your code you know makes you know does
17:16the same thing and a lot of people have
17:17depended on that
17:18for many years for both for debugging
17:20and for correctness
17:21but you can't depend on that anymore
17:23because all our computers are parallel
17:25and they schedule things dynamically and
17:26you have no idea what order
17:29results come out in and so a few years
17:32ago
17:33there was a an email to a mailing list
17:36where people
17:37share ideas about numerical software and
17:39there was a
17:40software developer from a company who
17:42said does anybody have
17:43some software that solves big sparse
17:46systems of equations
17:47and is guaranteed to be reproducible now
17:50why did he want it
17:51well he had customers who were civil
17:53engineers who
17:54who did you know made models of
17:57buildings and stuff
17:58and they had customers with lawyers and
18:00the lawyers said
18:02every time on your program that analyzes
18:05whether this building will fall down in
18:06an earthquake
18:07it can't change its mind it has to do
18:09that makes sense that makes sense yeah
18:11and so when i saw that i sent email to
18:14the 100 odd berkeley faculty on this
18:16mailing list to do you know large-scale
18:18simulations and i did i said
18:19did you folks know that reproducibility
18:22is dead
18:23you know is this going to be a problem
18:24for you and most people said how will i
18:26debug
18:27if i can't run my program again and get
18:29the same bug
18:30uh there was one person who said oh i
18:33know better
18:34i do error analysis that was professor
18:36khan but they're not many people like
18:37him
18:38and then there was one person who said
18:41oh um
18:42i do the way i do my science is i do
18:44these large-scale simulations looking
18:46for really really rare events
18:48and when they occur i have to reproduce
18:50them and then compute why they occurred
18:53and if i can't reproduce them i can't do
18:54my science
18:56that was professor govindji in civil
18:58engineering
18:59but the most interesting response was
19:02from professor stewart russell who's an
19:04ai professor in our department
19:06and it turns out that the united nations
19:08uses his software
19:1024x7 to detect underground nuclear tests
19:14which are all
19:14illegal and so he said well what if my
19:17my software says the north koreans did
19:20it and then they didn't do it that would
19:21just be a mess at the united nations
19:23so so there's all sorts of reasons to
19:25have reproducibility
19:26and so in the meantime a bunch of
19:28companies have responded so intel has a
19:30version of their
19:31math library that that slows down but
19:34gives you you know the same answer every
19:35time
19:36but the ieee floating point standard um
19:39decided to add a new instruction that we
19:41proposed and it's a new way to make
19:43floating point associative
19:45and and it's a very complicated
19:47algorithm but let me just say we had to
19:49add a new
19:50edition like instruction that rounded
19:52differently than before
19:53and so that's now in the standard wow
19:56fascinating that is fascinating my
19:59goodness
20:02i may have to add that to the lectures
20:04yeah well i actually wanted to ask a
20:06question from the lectures jim if you if
20:07you don't mind please um i i i know that
20:10there's been
20:11a fair bit of i'm not sure controversy
20:13but debate
20:14between uh professor khan and dr john
20:17gustafsson
20:18oh i can talk about that at great
20:20lengths i know
20:21i've been probably you've probably been
20:23in those debates i've seen some of them
20:24on youtube
20:24i i organized the great debate between
20:27khan
20:28that's right that's right at the uh
20:31computer arithmetic conference a few
20:33years ago yes
20:34right and i think that was taped on
20:35youtube i watched it it was fascinating
20:36to watch
20:37can you share your perspective on a unum
20:39i mean he claims
20:41actually a unum to a float is what a
20:43float is to a fixed point so let's
20:45let's have the final word on that let me
20:47start with just a little bit of ancient
20:49history
20:49when floating point numbers were first
20:52uh you know proposed as an idea for
20:55neumann
20:56thought they were a bad idea and that's
20:59because he said anybody who's
21:02competent using a computer should know
21:04what
21:05dynamic range their numbers are in and
21:07scale them all to fit in fixed point to
21:09fit an integer
21:10right right you know so so floating
21:12point is you know for people who aren't
21:13competent
21:14but 40.1
21:17obviously because it's much easier to to
21:20program and
21:20yeah yeah so
21:24so many floating point formats have been
21:26reinvented
21:27invented and reinvented over the years
21:29and
21:30unums were have the property
21:34that they have variable precision so
21:37they can use
21:37more bits for or fewer bits for the
21:39exponent depending on how big the
21:40exponent has to be
21:43so numbers near one have lots more
21:46mantissa bits and they're more accurate
21:48and numbers that are really large or
21:49really small have less precision
21:51so the first time this appeared was it
21:53was actually a homework problem in
21:55canoes volume 2.
21:57you know he said i did not
22:00invent this and the first uh proposed
22:03solution was published
22:04in 1971 um to do this
22:08and uh and then it sort of nobody paid
22:11attention then it was reinvented again
22:13in the 80s
22:14and in 1987 i wrote a paper
22:17uh which explained how it would mess up
22:19error analysis of all the standard
22:20algorithms
22:21so anyway gustafson decided to reinvent
22:25it
22:25but i think the thing that really made
22:27it uh people
22:29object was it was variable length so
22:31every number could have a different
22:33length depending on how many
22:35you know exponent mantissa bits it had
22:37and so that means how could you have an
22:38array you know
22:40if you have a 100 by 100 array where is
22:42a sub
22:43100 right right right right right
22:48the whole you know goal of saving memory
22:50which is gustafson's goal you know it
22:52doesn't work
22:52so so anyway he decided to go back to
22:55the drawing board and he invented
22:56something called
22:57posits and and that came out a few years
23:00ago
23:01and and those are also variable
23:03precision but it's fixed length
23:04fixed number of bits and um and so
23:07there's been
23:09several groups you know who've been
23:11who've been uh you know pushing it
23:13uh but let me just say i don't think
23:15it's a good idea
23:17um i have on my web page a link to a
23:20paper by some
23:21experts and the name of the paper is
23:24posits the good the bad and the ugly
23:27[Laughter]
23:29and so you know so they talk about all
23:31the things that can go wrong too but
23:32from the error analysis point of view
23:34making software correct
23:36it's just painful when numbers have
23:38different precisions depending on
23:39whether they're
23:40large or small or close to one or
23:42whatever right right and the other thing
23:43that's missing from posits
23:45it doesn't have nands and so if you you
23:48know divide zero by zero
23:50uh gus's and said oh you should just
23:51stop because your code is obviously
23:53wrong
23:54[Laughter]
23:56but you know this this just makes
23:57writing reliable code a little too hard
24:00right right yeah um one one uh
24:04comment about that you know risk five is
24:07uh
24:07open and anybody called invent an
24:10extension so
24:11this has been coming up uh multiple
24:13times by gustafson
24:15and others can we get a posits extension
24:18for x5 and they keep
24:20and then they keep showing interest in
24:22that well i had you know
24:23dave patterson about that last weekend
24:25and he wanted to know
24:27what's what are the arguments pro and
24:28con so i you know sent him a link to the
24:30good the bad and the ugly
24:32yeah very good but that's the beauty of
24:35this five it's completely democratic you
24:37can do
24:37whatever you want to do it including
24:40adding posits if it
24:41if it suits you well um so let me so so
24:44you you made the point that there are a
24:46lot of new
24:47formats being proposed not just unums
24:49and posits
24:50yeah b float 16 all these different
24:52things and and
24:54there was a paper just from a couple
24:56years ago called
24:57ternary arithmetic where there are only
24:59three numbers plus one
25:00minus one and zero so you know so it's
25:03wasting
25:04you know it uses two bits of course so
25:05it's wasting one of the patterns but
25:07right but there's a bunch of evidence
25:10that that's good enough for training
25:11neural nets all you need to know is you
25:13go up
25:13you go down or you not change
25:15interesting and
25:16so there is no one format that's going
25:18to make everybody happy
25:20and and it's a little bit of the wild
25:22west right now right but
25:23i think it's fair to say that you know
25:25not one format will make everybody happy
25:28but it makes sense final thoughts so
25:29flex point is kind of a distant relative
25:32of uh
25:34yeah so that that was from nirvana and
25:36just to tell a little side story on
25:38nirvana i was uh
25:40teaching 267 cs267 my parallel computing
25:43class
25:44you know this was a few years ago and i
25:46was teaching a new algorithm for
25:48multiplying matrices um that could go
25:51you know much faster than the previous
25:53ones and one of the students in the
25:54class thought it was a good idea
25:56and he founded nirvana
25:59a miraculous rashard and so
26:02so i was an advisor you know for that
26:04particular aspect of the work but not
26:06for uh for flex point so
26:10just as i mentioned everything is
26:11invented more than once so the old name
26:14for flex point is block floating point
26:16and it appeared
26:18in a 1963 book on air analysis by jim
26:22wilkinson who won
26:23the touring award in 1970 for his work
26:25and you know in that stuff
26:27so the idea has been around for a while
26:29and you know it has its pros and cons
26:32but again it puts a lot more burden on
26:34the programmer
26:35to figure out you know which numbers can
26:38i put together
26:40because they share a common exponent
26:42without losing too much precision
26:44and you know it sort of goes back to the
26:45neumann who said oh we should all just
26:47all scale stuff and
26:48use use you know integers you know
26:51versus let's make it easy to program
26:53you know life is trade-offs
26:57wonderful thank you uh so one final
27:00thought and then we will let you
27:02do whatever you need to do um so how are
27:05we going to handle all of these tf-32s
27:07and v-floats and
27:08and all like actually it's it's a soup
27:11yeah how do we handle the soup
27:13well all right um so we have a grant uh
27:16to keep maintaining la pac and scala
27:19pack
27:20you know these you know linear algebra
27:21libraries that everybody uses
27:23and one of the projects that we're doing
27:26is to ask
27:26are do these standard linear algebra
27:29algorithms work in b flow 16
27:31and so i propose this you know sort of
27:34half as a joke as a class project
27:36in my math 221 class this semester and a
27:38student said oh i want to do that
27:41so so what we're planning to do is uh
27:44port all of lapack to be float 16 and
27:48lapak comes with all this test code to
27:50see if whether it's correct or not
27:52and we're going to see whether it passes
27:53or fails all the tests
27:56interesting and so in the meantime
27:58though a lot of people are proposing
27:59mixed precision
28:01where you do you try to do a lot of the
28:03work in low precision
28:04say half precision and then you do a
28:07little bit of extra work at the end to
28:09clean up the answer
28:10and uh by doing it in higher precision
28:13and there's a bunch of algorithms
28:14been around for a while that work for
28:16some well-known problems
28:18um basically you solve the problem and
28:21do a few steps of newton's method to
28:23sort of clean it up
28:24yeah for some problems that's cheap and
28:26for some problems it's pretty expensive
28:28so it kind of depends
28:30it's it's like how i play golf you know
28:31you hit a big shot you get a rough thing
28:33and then when you're on the green you
28:34putt in at the higher resolution so that
28:35might make sense
28:36i like it nice yeah that's that's the
28:39claim for a100 so that you should be
28:42able to do multiple
28:43multi-resolutions so so so maybe a
28:46a lesson for the students is that you
28:48know you have to
28:49understand a little bit of enough math
28:52to understand where the errors can be
28:54so you're not surprised badly
28:59all right thank you jim and i believe
29:00you're you're teaching cs70 next
29:02semester so
29:03actually i just got changed to 170. oh
29:06perfect
29:07congratulations yes it is yes alessandro
29:10chase jason i will be teaching 170
29:12together wonderful
29:13all right so great i think people may
29:16uh some of the students may take 170
29:19next year may have you in their classes
29:21this coming spring that would be
29:22wonderful that would be great okay
29:26wonderful thank you very much thank you
29:28so much jim for what you've done for the
29:30department is and for joining us today
29:31thank you so much
29:32take care bye-bye wonderful take care
29:34that's great
29:36we have all right uh we have
29:40two um kind of polls to run
29:43um i think uh i'll hand them over to to
29:46then the first one
29:47sure i don't know how to start them but
29:49i'll first give the introduction for the
29:51first one
29:52um we have been
29:55thinking about how do we do this um
30:00how do we do the the clicker questions
30:02right
30:03because we do want to keep everybody on
30:05track but we
30:07are trying this time not to keep
30:09everybody on
30:10really too tight of a leash um we
30:13really need everybody to stay on track
30:17with the lectures such that we can
30:19proceed with homeworks and we can do the
30:21labs
30:22and they're all you know all whole
30:25thing or complete learning experience so
30:29um how do people feel so i'll hand it
30:32over to them
30:32about how do we administer these
30:36quicker questions sure so reason number
30:39three now
30:40yeah i mean again we're trying to
30:41optimize this is all new for us we want
30:43to basically
30:44be responding to what you'd like to see
30:46um
30:47and uh we want to have a little bit of
30:49time as well to talk about how to do the
30:51exam so we'll talk about that first but
30:52the first question is
30:54um you know these lectures sometimes
30:57they come in single one lecture
30:58pieces which we slice up into many
31:00pieces that's what i did for floating
31:01point and by the way that did grow a
31:03little bit longer than 50 minutes and
31:04that's okay because some lectures will
31:06go smaller than 50 minutes so it'll work
31:07out
31:08to be just about right we're not going
31:09to double load you in terms of lecture
31:11but this floaty was a little extra
31:12because i wanted to show you a little
31:12demo
31:13at the end of it um so that's an easy
31:16one that you know that one was you there
31:17but
31:18then we said you know do we have it due
31:19every single monday wednesday friday to
31:21keep you up like this
31:22or do we say let's try to do it once a
31:24week as a compromise of that and
31:26different people might
31:27thursday saturdays when it's time right
31:29exactly should they be do tuesday
31:30thursdays and saturdays or just kind of
31:32do thursdays
31:33and you have to do every other thing
31:34from last week up to every thursday so
31:36that's kind of where we
31:37how we're doing it and if there's any
31:38strong preference either way um you can
31:40do it and if by the way you would have
31:41preferred it monday wednesday friday
31:43we'll try to have it up in advance so
31:44that you could actually go ahead and do
31:46those things so that should that should
31:47work as well
31:48um that's the first part so thank you
31:50for that feedback it looks like people
31:51are really split on that i'm looking at
31:52the numbers here in the poll the poll
31:53looks like this was
31:55not tremendous help to you this is yeah
31:57they're all around the board
31:58every single bucket is there in the
32:00thing it's a very funny thing
32:02uh just so you know i mean every single
32:04uh
32:05all of us including tas had some
32:08opinions about how to do it right
32:10right and and some of that's borne out
32:14by data we actually surveyed the folks
32:15last summer
32:16and the folks last summer said almost
32:17overwhelmingly that they would be
32:19falling behind if it weren't for us
32:20keeping them
32:21up to pace every monday was a friday
32:23like that like a heartbeat uh
32:24summer does run a little bit faster than
32:26than spring and fall but still
32:28uh that was very useful to have that as
32:29a part of our conversation
32:31part two is um we want we wanted to have
32:35allow for you to have an asynchronous
32:38model
32:38uh that was the first the first pass you
32:40know the first draft is let's have them
32:41be asynchronous
32:42let's give them a whole day they can
32:44wake up sunday morning
32:45and then have a nice breakfast and pull
32:47out the exam the exams will be broken up
32:49into questions
32:50and every question will be a time where
32:52you jump in you're zoomed you know
32:53you're locked in
32:54and you can't stop you can't take a
32:56break mid question and then you gotta
32:58freeze it and hopefully that question
32:59doesn't take you more than 20 or 30
33:00minutes so
33:01hopefully you can give a good 23 minutes
33:03you know big stretch and then go take a
33:04stretch and then walk around and have
33:06some food hit the restroom
33:07and then come on back again so that's
33:09the model that was gonna be the model
33:11many people have shared on piazza that
33:13there's a concern about how to do
33:16the time lapse you know we we were
33:18thinking if you
33:19you need to have some way to verify
33:21people just keeping a copacetic you know
33:22having a nice thing and they're not
33:23going on google or not and you know
33:25their friends
33:25so how do you do this without the
33:27onerous and we've heard a lot of
33:29negative comments about
33:30synchronous zoom proctoring being really
33:32hard for people at different time zones
33:34and what happens if the thing goes down
33:36i know you're watching blah blah people
33:37are watching
33:38they're seeing your room what if your
33:39parents come in or if you're all the
33:41so that's a concern so we wanted to
33:43basically allow people to have an ace
33:45and by the way synchronous means you and
33:46you when you know you and the staff have
33:48to be on the same computer
33:50at the same time for some amount of time
33:52and you have to kind of limit what that
33:53time is
33:54unlike the asynchronous model which says
33:56just go home and take it for 24 hours if
33:58you want to work for 20 hours that's
33:59great you want it for two hours you work
34:00for 30 minutes that's fine
34:02it's just more relaxed in the
34:03asynchronous model but we do it we do
34:05understand
34:06that some people are worried about their
34:07phones not being able to do
34:09um a time-lapse video we did do a test
34:12and we found that the test
34:13for and remember we don't expect people
34:15to really because we're going to break
34:16this up into pieces
34:17you'll do one piece that's 20 minutes so
34:19people it may not be that much work
34:21on your phone in terms of the amount of
34:22storage you need and the amount of
34:24actual you know overheating your phone
34:25might do because you're not going to be
34:26there probably for an hour
34:27you'll be in for a set of small
34:29questions and then stop and then break
34:30and then come back in so it may not be
34:32and for each one of those we're going to
34:33time lapse video
34:35as well so that was the initial model
34:38but some folks said well we can't
34:40actually do that because my phone
34:41doesn't have a time lapse and nobody in
34:43my family has a towel up and i have no
34:44access to a time lapse
34:46phone that'll do that so how do i do
34:48that so
34:49bore i i would give bohr the great the
34:51great pro the props for thinking of this
34:54he said well what if we allowed for a
34:56backup with zoom proctoring
34:57so that means we know people usually
34:59don't like zoom proctoring
35:01usually although i'm looking at the
35:02survey of the survey it's
35:04equal preference for zoom proctoring an
35:06equal preference without doing proctor
35:08so that the idea is for people who don't
35:11have access to that
35:12um you you would be able to get online
35:15and we're going to only
35:16you know maybe have one time where you
35:18have to get in there and it's going to
35:19be
35:20a fixed time we're going to be there you
35:21have to lock in we have to start at the
35:23same time
35:23and it's not that whole flexibility is
35:25gone you're in there
35:27to you know you're in there stuck and we
35:29have to watch you when you're watching
35:30your thing and you're watching the
35:31screen
35:32and bruh and we've got the whole
35:34situation and in that situation your
35:36camera has got to show your whole
35:38i mean we have to then to have a
35:40conversation maybe on our side of how we
35:41get the screen
35:42and you and your desktop and how all of
35:44that together if we if we do zoom
35:46proctoring you have to then
35:47have your camera above maybe you have to
35:49have your camera as the zoom camera
35:51showing both the screen and the desktop
35:53so we can see the screen because you
35:54obviously could have another window open
35:56with some bad stuff so we need to be
35:57able to see
35:58what's on your screen and we didn't want
35:59to have a time lapse of both the screen
36:02so we didn't want to have both zoom
36:03proctoring and a screen recording we'd
36:05like to be able to have just one movie
36:06for each of these things or a zoo a live
36:08zoom that is able to see the whole
36:10desktop
36:11so this is a concern this is where we're
36:13working with this fortunately this this
36:15quest is not worth that many points
36:17so it's not like well this is a 90 you
36:18know worth 90 of your grade this is
36:20worth a very small amount of your points
36:21um and and we're going to try i think we
36:24might have to try both because i'm
36:25hearing that enough people
36:26cannot do the time lapse that we have to
36:28be able to to give a little bit
36:30so i want to be able to have some
36:32questions and concerns bro you want to
36:33jump in yeah
36:34there are some good questions that that
36:36came through the chat
36:38i don't know if i'm uh i i'm paying too
36:41many
36:41too much attention steven if you see
36:43some maybe you can
36:44read a few um there are comments over
36:47there that uh i think some people
36:48didn't realize that if it is a zoom
36:50proctor that it's going to be a fixed
36:52time we are not going to be correcting
36:5424 hour
36:55correct yeah dan and i are not going to
36:57be sitting in front of screens for 24
36:59hours
37:00right um so so so we we almost want to
37:03encourage people to take the advantage
37:06of the asynchronous
37:07do it whenever you comfortable
37:09time-lapse model if that works
37:11and encouraging people if they can't if
37:14if you know you really prefer to just
37:15the way you work or because of technical
37:17reasons you really need to go with the
37:19zoom proctor
37:19we're going to have to find the time for
37:21our staff and us to be there and
37:23uh it'll be on sunday so we'll we'll
37:25both be on piazza answering questions
37:27and we'll be have you know the official
37:29exam starts here an official exam ends
37:30there and it'll be the same idea
37:32either way by the way each person is
37:34going to get hopefully a randomized exam
37:36and so your exam should be different
37:37than your neighbor's exam and you'll be
37:38doing it
37:39and being able to we'll be able to watch
37:40you and have to be able to answer any
37:42questions
37:42so it's not like you'll get more support
37:44if you're zoomed you're going to be me
37:46here's the here's the thing about the
37:47zoom proctoring you're going to have to
37:49have your camera on
37:50again that's a privacy concern which is
37:52why we wanted to not have that as the
37:53default
37:54and your microphone has to be on both of
37:56those have to be on so we're hearing
37:57what's room you can't have
37:59people in the room doing that so that's
38:00very different you have to be on a fixed
38:02time
38:03with the camera microphone on but your
38:05speakers are off so you're not hearing
38:07what so if someone coughing you don't
38:09want to hear everyone coughing in this
38:10room but essentially it is like we're
38:12all watching you at the same time
38:13like that it's not any different so
38:15that's i hope that i don't that affects
38:17people's votes
38:18but that's that's what the space of that
38:20zoom proctoring situation is going to be
38:21and it's a lot more
38:22it's now we have to now manage two of
38:24these exams to do this but most people
38:26in the summer did not want to have zoom
38:27proctoring and most people
38:28were doing that so this is always a
38:30backup for people who don't want to have
38:31for
38:32whatever reason the time lapse technical
38:34reasons or or
38:35or their own the way they study so yeah
38:38there is a question that i saw multiple
38:40times you know what
38:41yes everyone's in the same room that's
38:42the idea no no the time lapse time lapse
38:44it's
38:45like there's a standard setting in
38:46iphone if you have an iphone it's like
38:49takes a picture of you every five
38:50seconds that's correct you want a
38:52picture of you from behind your back
38:54um and your screen and that's it it's
38:56not a very big file
38:58but you don't have an iphone then you
39:00have to have an app
39:01on android that will do that and we have
39:03a policy in this class which is unlike
39:05every other policy which is our exams
39:07clobber
39:08so if you do better in your midterm than
39:10your quest then
39:11that'll be that'll be the the way so
39:14because of this clubber possibility
39:16um you don't also need a tripod you can
39:18have anything that can a box anything
39:19that can place your camera above and
39:20behind you
39:21so it sees your screen and your desktop
39:23a little an amazon box
39:25tape it to the wall like literally you
39:26could sit with your back against the
39:28wall put on a windowsill i just
39:29responded on piazza about that
39:31but if you decide to go to the zoom
39:32you're all in the same zoom meeting
39:34everyone microphone and camera must be
39:36on and but your speakers are off so you
39:38can't hear anything you're just
39:40delivering the data to us and we're
39:41watching the whole room and capturing it
39:43on recording so we have that there
39:45um but no one's able to no and you're
39:46able to see everybody else that's the
39:48part of this
39:48the privacy thing you have to find a
39:50quiet space that you can do that
39:51and that's a lot harder um we thought
39:53that this was a lot easier a lot more
39:55flexible
39:56because you can do it you know do a a
39:58couple of questions
39:59go visit the refrigerator or do it yeah
40:02you need to
40:02exactly so so we we think you we thought
40:06that the students would just eat it up
40:07they would just oh we love thank you so
40:09much
40:09but enough people are like wave blah
40:11blah blah blah technical technical
40:12yeah i gotta have a box i don't have a
40:14thing i'd rather be in the zoom just
40:15make it easy
40:16all right we'll give you that option too
40:18so that was borah's great idea and i
40:19think we'll support that as well
40:20and we'll have to decide very quickly
40:22what time that'll be and let you know
40:24it'll be sunday so it's four days away
40:25but we'll let you know as well as
40:27revealing
40:27uh sharing the official document uh
40:30about the logistics of
40:32you know here here are ten different
40:33ways to put your camera above your
40:35shoulder
40:35here's a box here's a tape on a wall
40:37here's sitting behind a windowsill
40:39here's uh you know that kind of thing so
40:40we'll have to do you know
40:42here's my neighbor holding it up but
40:44here's my neighbor holding it up
40:45for an hour you know above my thing so
40:48um
40:48that's the place of that uh happy to try
40:52to talk to more this is the chance for q
40:53a this is the
40:54ask me anything situation um
40:57both of us yeah anybody can see the the
41:00questions
41:01actually they're in the chat so stephen
41:04if you see them
41:05they're also in q a and it's hard to
41:07follow all of this yeah
41:09a lot here i got a lot here um
41:14and zoom proctoring you can't just
41:16cannot disconnect audio i'll just i'll
41:17click on that one
41:18you'll have to have your audio on and
41:20your video on both and
41:21so we're hearing sound from the whole
41:23from every room and we can mute you so
41:25we can mute you and unmute you
41:26uh but you'll have to have that on so if
41:28someone were kind of whispering over
41:29your shoulder telling you the answer we
41:30would know that that's that's the idea
41:32um so that's that one there's a question
41:34though is the quest
41:36digital only will it require submitting
41:38any paperwork
41:39uh currently no you will not need to
41:41submit any paperwork no paperwork
41:42exactly no paperwork and cs70 had a
41:44whole policy about
41:45you print the paper and you work on
41:47paper none of this is there you get a
41:48randomized exam as i think 70 did as
41:50well
41:51or to try to at least um and then so
41:53this is everybody's a randomized exam as
41:55much as we can
41:56give randomization and it's all digital
41:58you're going to be in in the system
41:59called prayer learn and
42:00here's the important thing we're going
42:01to have a practiced quest
42:03that's going to be released very very
42:04soon if it's not already released uh
42:06steven do we give a sense of eta for
42:08that at all i don't know because
42:10cece is the one who's spending okay so
42:12so yeah but so but it's going to be very
42:13soon we're going to have a practice exam
42:15which means and by the way in the
42:17practice exam i would encourage
42:18everyone every single person to try to
42:21put their phone above them and try to do
42:22this and put it on for an hour
42:24i don't think it's gonna take any of the
42:25sub questions an hour but try it you
42:27know give that kind of a high a high
42:29order a
42:29high water mark put your phone above you
42:32try to capture that and get a sense and
42:33we'll also try to share as many as much
42:35as we can
42:36examples in the we're going to have this
42:37again the quest logistics stock that'll
42:38have
42:39all the examples of that um because it's
42:42it's going to be a a real pain for folk
42:44for folks to
42:45to have to do zoom proctoring if
42:47everybody has to get on the same time
42:48zone so if they make the more people
42:50that can
42:51do this is asynchronous thing we think
42:52it gives you by the way the zoom
42:54proctoring we're going to have a finite
42:55time
42:56you know maybe double time so we'll have
42:57two hours we're going to sit with you
42:59to do that but you'll have 24 hours if
43:01you do the
43:02the the time lapse that's the other
43:03difference is you can take a rest and
43:05obviously not going to ask you 24 hours
43:06to work
43:06probably it's going to be we think it's
43:08going to be less than an hour that's
43:09what we think we're going to try to have
43:10the exam that's doable in less than an
43:12hour that's what we always try to do
43:13um so that's the idea yeah and again we
43:16are all we haven't done nobody's
43:18administered an exam like this to 1200
43:20people
43:21right so we're trying our best right and
43:24by the way
43:25i should i should add somebody that had
43:26an interesting point which is
43:28you could do zoom if you have a second
43:29computer if you have a second computer
43:32that computer behind you could be your
43:34your thing so you if you wanted to zoom
43:36from a second device
43:38you could zoom record if you can record
43:40that you can record that it doesn't have
43:42to be time-lapse if it's actually the
43:43full video that's fine it'll be a bigger
43:44data file but we can live with that too
43:46so if you have something else to record
43:47you like if you have a gopro or a second
43:49computer doing zoom you can do that the
43:50problem is
43:51the normal zoom is only from out the
43:53computer we need to see your desk space
43:54we need to see what you're going
43:56to see your screen and the zoom wouldn't
43:57be able to do that you can do it from an
43:59android phone you can just call in
44:00into you know from a zoom app and right
44:03so
44:04so so the cloud recording yeah that you
44:06could do that if you wanted to do that
44:08so there are ways to you know be
44:09creative about this and hopefully we'll
44:10add this to all these logistics and the
44:12logistics stock we're trying to support
44:14everybody in this kind of new space but
44:15again we're trying to be as flexible as
44:16we can for everybody
44:18i'm at 501 i'll make sure um so zoom
44:21proctoring okay disconnect audio no you
44:22cannot
44:23um uh
44:26again the camera is going to be showing
44:28so i'm answering this one live
44:30the camera is going to be showing your
44:32desk your hands and your screen that's
44:34the most important thing okay so we see
44:36so on our table and by the way usually
44:38these things are high resolution because
44:39they're free they're just a picture
44:40every time
44:41not a whole full movie but it's a full
44:42picture so we can actually be able to
44:43read your screen
44:44and see what you're doing for every
44:46single frame we'll be able to go through
44:47that and the smaller file format
44:49so that's there um asynchronous can do
44:52zoom yes you can uh and i answered that
44:55live
44:56because you could use from a second
44:57computer for that um
45:01a lot of questions boy um
45:04the i'll do the answer there is no time
45:06limit when you start the exam if you do
45:08the asynchronous model you can start and
45:10stop whenever you want to you just have
45:11to
45:11only you cannot start and stop within a
45:14question
45:15you do one question you cannot start and
45:16stop and that question should be less
45:18than
45:18should be about 30 minutes max for each
45:20of the small parts i think probably 10
45:21minutes for some of them five minutes
45:23for some people
45:24to go take a break so you can't stop
45:26once you started a question
45:28but then you can take as much break as
45:29you can for that so there's no time
45:30limit 24 hours to do all the whatever
45:32parts we have i think it's like three or
45:33four
45:34or maybe five parts we have to the exam
45:36um
45:42one question that they can take about
45:43these due dates right i mean we
45:45have among ourselves the the debate
45:48there are different people who have
45:49different learning styles so
45:51you know it may work fine if everybody
45:54if we have everybody uh watch every
45:57lecture
45:58when it's supposed to be monday
45:59wednesday friday but some people may
46:01prefer to watch two or three
46:03um as long as they're not falling behind
46:05and they're ready
46:06for the labs and and homeworks and so on
46:09so we're trying to accommodate that we
46:11made a due date on thursday for two
46:14reasons first
46:15honestly if you look at risk five
46:16lectures they are not three times 50
46:19minutes they're less
46:20and you should be able to there's
46:22slightly more than two
46:24um you know 50 minute slots that are in
46:27there
46:28even though i've thrown extra stuff in
46:30there i still wrap it up in less than
46:32three times 150 minutes i'm done in 120
46:35minutes about two hours total
46:40but we would like you to be ready for
46:43the for the labs that are on wednesday
46:47thursday and friday
46:48and that's the big goal that's what we
46:50are trying to set so we can try to
46:51adjust uh
46:52we'll try to see what we can do uh
46:54without should we you know capture
46:57um you know there is a strong preference
46:59now i think to go
47:00for um you know three times a week two
47:03days for
47:04due dates for clickers but the option
47:07for that perhaps maybe to go friday
47:09monday wednesday and then make sure that
47:12every
47:12everybody is done by wednesday and ready
47:14for the labs that week
47:17and we can try to adjust to that
47:19actually there's a quick question can
47:21you
47:21can it's a proposal could you possibly
47:23do the zoom recording for the time lapse
47:25on the device you're taking the test on
47:26the answers no
47:27because we kind of have to make sure
47:28your hands and your phone and everything
47:30is clear so it's like
47:31the model is the time lapse covers your
47:34space your workspace
47:35your desktop your phone your your what
47:37computer you're using
47:39if you only record the screen then you
47:40could be on your phone texting your
47:42friends and that's not cool
47:43we basically try to reduce all the kind
47:44of outside ways we're trying to lock the
47:46house
47:46and prevent all the outside obvious ways
47:48for this and that's that's the reason we
47:50have the phone over the shoulder
47:51situation so you can't do that one um
47:54dan what about if they have multiple
47:55monitors
47:57um i'm actually okay as long as this
47:59camera can see the multiple monitors i
48:00mean if you have to if you
48:02if you want to be able to have by the
48:04way we're going to ask you to do some c
48:05programming in the exam
48:06the practice exam this is very important
48:08before this is really crucial
48:09the practice exam is just to practice
48:12with prairie learn the process exam is
48:15not to simulate
48:16exactly the difficulty or exactly the
48:18topics that are on the exam
48:20so there is a floating point question on
48:22the practice exam
48:23do not panic there's not floating point
48:25on the quest
48:27okay so that's just to give you we just
48:28to throw you out kind of the kind of
48:30questions
48:30that how do you answer one is a multiple
48:32choice one's a fill in the blank
48:34we threw a floating point in there
48:35because we had it in our back pocket but
48:37it's not on the quest so
48:38do note that the practice exam is and
48:40we're going to send it out there very
48:41soon
48:42is going to is going to be not exactly
48:44the model
48:45or the difficulty with the question be
48:46don't come to me say hey the practice
48:48exam
48:48was too easy compared to yes it's much
48:51easier we're just throwing some basic
48:52baby questions out there
48:53that are not not just trivial what color
48:55is the sky but actually it's maybe some
48:57learning that happens in there with a
48:58thing
48:59but we threw floating point in there
49:00because we had in our back pocket but
49:02floating point is not on the exam so
49:03it's not a model of the difficulty
49:05but it is a model of kind of how you
49:06interact with the system and you should
49:08please do that before sunday
49:09so you can practice it yes uh to to
49:13to finish on a light note we are not
49:16asking to calculate avogadro's number
49:18from the color of the sky
49:21you know how to do that that's right
49:26six point something 6.023 to the
49:29power of ten to the power of twenty
49:31three twenty three right
49:32from the color of the sky today because
49:35there are too many colors
49:36in the sky but you know on a clear day
49:39you totally can
49:40right it actually got brighter like i
49:42can't believe that it's it's five
49:43o'clock and it's
49:44brighter you know why it's brighter by
49:45the way you know why
49:47because here's the cloud layer here's
49:48the sun up here and now the sun is from
49:50and now it's hitting from a skew angle
49:52so it's cutting under it so it's
49:53actually getting major
49:54brighter at five o'clock than it is at
49:56noon unbelievable um
49:58that's it we'll try to answer more
50:00questions on piazza we're going to
50:01launch the practice quest
50:02we'll have the logistic exam any last
50:05thoughts stephen there was one more
50:06important one was is the quest open book
50:08can we have lecture slides textbooks
50:10uh open on our computer you can have the
50:12answer is
50:13yes you just cannot do google searches
50:15you can't ask your friends and you can't
50:17do
50:17stack overflow but anything we've given
50:19you it's all open book anything we've
50:20given you you can use you can use a book
50:22pull the book out anything we've given
50:23you or anything from a resource book but
50:25you just can't
50:26search on the web that's the idea it's a
50:28big difference from past
50:30uh yeah very different by the way and
50:32this and this would be true but this
50:33would be true for
50:34either either model whether you decide
50:36to go asynchronous or zoom proctored
50:38that same policy of being open book to
50:40be for everybody that makes sense
50:43all right good stuff
50:46well we're late for our five o'clock
50:48meeting bar thanks everybody
50:50we hope you hope we have this we hope
50:53this helped good to see everybody
50:55we'll see you next time there is a
50:57question before we close it there is a
50:58question about what happened to the
50:59group
51:00tutoring uh my group tutoring is on
51:02wednesday my individuals on friday
51:04borah's individual wednesday group is on
51:06friday so on friday
51:08bori will be having his group tutoring i
51:09believe or you can give the thumbs up to
51:11that or not
51:12let's try to organize that so we can
51:14probably it's it's very close to the
51:16to the quest so we will just talk about
51:19that now and we'll let you guys know how
51:20okay okay perfect perfect a serious
51:23review session so topical review
51:25sessions such that
51:26yeah that makes a sense compressed
51:28version of what we have done so far
51:29perfect perfect good stuff thanks
51:31everybody good to see you all take care
51:33and we recorded it successfully we did
51:34this all right
51:36we remembered thanks steven

About this transcript

This page contains the full transcript of [CS61C FA20] Weekly Lecture 03.LIVE - Floating Point & RISC-V by CS 61C Departmental, generated from the public captions YouTube serves with the video. The transcript has 10,271 words across 1,669 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.