Category
🤖
TechTranscript
00:00Whoever made the tweet asking it how many LTT backpacks would fit in a the
00:08trunk of a Tesla or whatever the question was? Oh, I didn't see that.
00:10Someone made that tweet on the LTT handle. Oh, that's really funny. And it did it.
00:14It looked up the dimensions of the LTT backpack. It looked up the dimensions of
00:19the trunk and it figured it out. How the f*** did it do that? Ask it. Let's do it live.
00:22Because I thought the dimensions for the backpack are in picture form.
00:27Searching. Searching for that. Now it's searching up the LTT backpack dimensions.
00:31Shut up! Look at this!
00:33Based on some rough estimates, I will try to answer it.
00:36That's insane!
00:40That's actually nuts!
00:42Based on some videos of the Model Y trunk.
00:44Shut up!
00:46It can fit about five to seven standard carry-on suitcases which have
00:48similar dimensions and capacity. Holy...
00:50Which is accurate.
00:52That statement is f***ing s***.
00:54That's...
00:57Crazy!
00:59Look at this. How did this happen?
01:01How did Bing...
01:03No offense, Microsoft, but how did Bing just beat Google to the punch
01:07so dramatically at something that's so important and so core to their business?
01:13Well, there's actually a really good reason for it.
01:16So AI has been blowing up lately, both in the news and in real-life applications
01:21across a ton of industries. So, you know, years ago it was only in relatively small things
01:26like helping doctors detect cancer early using advanced pattern recognition
01:30and then a little bit more over the years with things like autonomous vehicles.
01:33But now AI is everywhere. It's creating whole original pieces of art.
01:37It's holding conversations with humans all over the place.
01:40It seems like we've just arrived at the beginning of the AI age.
01:43There's this chart that keeps popping up that hits extra hard,
01:46which is the time to reach 100 million users.
01:49And you can see the faster and faster adoption curves of these increasingly disruptive new technologies.
01:54The telephone took 75 years to hit this milestone 100 million number.
01:58Then the mobile phone reached the same mark in just 16 years.
02:02Netflix took only 10 years.
02:04And Twitter took 6 and Gmail only took 5.
02:07Facebook in 48-ish months was absolutely massive.
02:11Instagram hit it in just 30 months.
02:14Now TikTok we view as this gigantic existential threat.
02:189 months to 100 million users.
02:22ChatGPT, 2 months.
02:24I mean, looking at numbers like that, I buy it.
02:26It seems almost obvious that we're clearly on the precipice of something really, really big.
02:31That's going to change everything.
02:33So seeing Microsoft at the forefront of it with this new Bing shouldn't really be a surprise.
02:37I mean, people already like talking to these chatbots and asking it all sorts of questions.
02:41So it sort of feels natural having this chatbot act as your co-pilot for the web alongside search
02:47instead of just a traditional search engine full of links.
02:51But there is one thing that's going to follow this conversational AI thing everywhere it goes.
02:56Everywhere you see it.
02:57Which is that sometimes it's just wrong.
03:01Sometimes it just says things that aren't true.
03:04Because fundamentally the AI doesn't know if it's telling the truth or not.
03:08It doesn't understand that.
03:10Like that's not part of the model.
03:11Like what we're seeing is it taking our inputs and then creating outputs
03:15based on related words that are most likely to go together.
03:19It's not forming a sentence like humans do.
03:21It's generating a new sentence.
03:23And so adding it to a search engine like Bing,
03:25it's scraping all these relevant links and information
03:28and synthesizing new sentences just based on how it thinks things should be pieced together.
03:33It's not sentient.
03:35It doesn't understand what it's saying.
03:38And so it's definitely not fact-checking itself.
03:41So we have to keep that in the back of our mind through all of this.
03:44Every time you see a headline.
03:45So it's really interesting with these search engines, right?
03:48On one hand, you have Bing who has everything to gain.
03:51And then on the other hand, you have Google who has everything to lose.
03:55I've had access to this new Bing for a little bit.
03:58It's a limited preview before they push it live to the rest of the world.
04:01I'm just playing around with it.
04:03Basically it adds this chat experience alongside regular Bing.
04:07It's essentially the same experience as talking to chat GPT.
04:10But instead of being limited to a fixed data set that cuts off at 2021,
04:14it'll pull from the entire current web that Bing can scrape from.
04:18You can type in a question, flip it over to chat,
04:20and it'll give you a nicely written summary that's synthesized
04:24based on what it finds for similar queries.
04:27So if I ask it something kind of simple,
04:29like, what's the average lifespan of a cheetah in the wild?
04:33It gives me an answer, right?
04:35It gives me a convincing bunch of sentences.
04:37It actually gives me more information than I asked for.
04:39It tells me about cheetahs in captivity too,
04:41which makes it feel very convincing.
04:43It also gives little footnotes and citations for some of its sources.
04:47And it gives links at the end if you want to dig in some more.
04:49It's really impressive, actually. It looks good to me.
04:51This is like a real product that's going to ship to all over the world,
04:54like people everywhere in the next month or two, I think they said.
04:57But this could only come from Bing right now.
05:01The more you use it, the natural language is super, super impressive.
05:05The fact that it gives me a convincing-sounding couple of sentences in a row
05:09and strings it together based on my input, super cool.
05:12But the more you use it,
05:14the more you start to see these weird patterns and these habits
05:18and these shortcomings.
05:20Again, mostly in the fact that sometimes it's just going to be wrong.
05:24A little game I like to play is ask it a question you already know the answer to
05:28and then read what it says and spot the error.
05:32So I asked it right now, okay, what are the best smartphone cameras right now?
05:35And it gave me S23 Ultra, Pixel 7 Pro, and iPhone 14 Pro Max
05:40with this nice little write-up with some specs for each.
05:44That's actually a pretty good list, but it is wrong about some of these numbers here.
05:48The S23 Ultra has a 200-megapixel camera and a 12-megapixel front-facing camera.
05:53But yeah, okay, it's mostly right.
05:55I then asked it, what are the five best electric vehicles out right now?
05:59And it gives me some, five reasonable options.
06:02But I don't know any expert that would put the Jaguar I-PACE on their list right now
06:08and leave the Rivian off.
06:11So basically the answers that it gives are really convincing to someone
06:13who doesn't know anything already about that subject.
06:16But if you are already an expert in the subject that you ask it about,
06:19then you'll find that the answers are like C+, maybe B+, sometimes at best.
06:24So now you see what's happening?
06:26Now suddenly when you're asking ChatGPT or Bing about a factual thing
06:30or something you need help with, suddenly you also should probably add these layers on top.
06:35Like, am I a complete noob in this topic that I'm asking it about?
06:39Am I just willing to blindly trust whatever this spits out without any further research?
06:44Is a B-plus answer going to be good enough for me,
06:48even if it might have some possible errors in it?
06:51You know, that might be good enough for just asking like,
06:54you know, how old a cheetah gets or something like that.
06:56But maybe not good enough for planning a trip
06:59or meal planning for someone with an allergy or something like that.
07:02And then if you look around the internet, people have gotten it to give
07:06increasingly more and more unhinged answers over time
07:09as it tries to simulate conversations and stay in the flow with natural language.
07:13I've seen anywhere from arguing about simple corrections
07:16to spewing weird stories about how it's spied on its own developers
07:20or how it wants to be sentient
07:22to gaslighting people about things and lying about its previous answers
07:26and just saying some straight-up scary stuff.
07:28Just go to the Bing subreddit for like an all-you-can-eat helping
07:31of all the insane stuff that Bing has said to just the people testing it
07:36over the past couple of weeks.
07:37Can you imagine?
07:38Can you imagine if Google did this?
07:40If Google search, at the top of search for people,
07:43was just spewing out random stories and misinformation
07:47and like all kinds of insane unhinged things?
07:50That would not fly.
07:52Now, to be fair, this version of Bing isn't out yet to the public, right?
07:56So it is still a small group testing phase.
07:58But even with this, like Microsoft knew that some of it is gonna get out there
08:01and potentially go viral.
08:03It feels like they even basically programmed in
08:05lots of friendly emojis to try to soften the blow.
08:09So when it knows it's giving an answer to maybe a more controversial topic
08:12or something that it doesn't have a super clear answer for,
08:15you might get a little smiley face at the end
08:17just so you don't, you know, take it too seriously.
08:20Also, literally as of today when I'm testing this,
08:23it started completely bailing on a lot of topics
08:26that might just be the slightest bit existential or dangerous.
08:31It just says, I prefer not to continue this conversation
08:33and then it just stops.
08:35Just refuses to answer any more questions on that topic until you reset it.
08:38Which seems like a pretty good failsafe.
08:40It's a pretty good idea in hindsight.
08:42But like we've already seen the other stuff.
08:44It's gotten out there. The damage has been done.
08:46Like the point still stands.
08:48This could have only come from Bing.
08:50Like a lot of people might have forgotten about this
08:52or might not have even known about this.
08:54But Google has been working on conversational AI stuff for years.
08:58We've seen Google Assistant.
09:01But they also literally showed an AI chatbot demo
09:04on stage at Google I.O. in 2021
09:07where you could have this whole conversation with any person or object
09:11or anything in the universe that you wanted.
09:13Their demo on stage was asking Pluto about itself.
09:17Nice and friendly, right?
09:18Oh, what's it like to be you Pluto?
09:20What would it feel like if I visited you?
09:22How do you feel so far from the sun?
09:24The difference with Google is
09:26this was never shipped as a product
09:28like this was an internal research project
09:30but the idea of displacing
09:32their massive search and ads business
09:34with a chatbot
09:36that gets things wrong all the time
09:38is insane.
09:40It can't happen, right?
09:42So literally search and ads is more than half
09:44of Google's revenue as a company.
09:46That's what having everything to lose looks like.
09:48Now to be fair,
09:50Google did hold an event in Paris
09:52the day after Microsoft's event
09:54which was them talking a little more
09:56about their chat with search
09:59AI plans and they did say
10:01they're planning on eventually doing a chatbot
10:03on top of Google search.
10:05It's called BARD. It was much more subdued though
10:07and yes, it also literally
10:09did have a factual mistake
10:11in the promo for it.
10:13So look, I actually like the idea.
10:15I obviously think it's smart
10:17when you're on the precipice of this huge AI thing
10:19to have AI kind of be this
10:21co-pilot for the web to help you around
10:23the internet. The idea of it
10:25summarizing a longer piece
10:27into some bullet points accurately
10:29that would be great. The fact that it could
10:31give you spark notes for a longer book you haven't read yet
10:33cool. It could even help you
10:35meal plan, help you plan a trip
10:37help you make a purchase decision.
10:39But it's clear that we're still at the beginning of this.
10:41There are so many unanswered questions
10:43from obviously the fact checking
10:45to do schools
10:47embrace this or
10:49ban this?
10:51How do search engines
10:53keep sending traffic to the
10:55publishers who are the sources
10:57that the chat pod is scraping
10:59from? I mean you get the links at the bottom but
11:01a lot of people are not going to click those anymore if you just give
11:03them the answer above the search results. So right now
11:05in its current stage, my take is
11:07anything we do with any of these AI
11:09tools should be a collaboration
11:11with the human
11:13touch. Like you wouldn't
11:15just put in a query in Dolly
11:17and then just take whatever it generates
11:19and put in a frame and just call
11:21that art, right? It's more for inspiration
11:24for your own paint and canvas.
11:26Like you wouldn't ask ChatGPT to write an
11:28essay and then just copy and paste it
11:30and submit it as your own. It's supposed
11:32to be the inspiration for the framework
11:34for your own piece, for the
11:36human touch. So of course you shouldn't ask
11:38the Bing chat bot what TV
11:40you should buy and then just like mindlessly
11:42click and buy the first one that comes up.
11:44I mean it could be fine but it could also
11:46be a C plus answer.
11:48You should use that as a springboard
11:50for your own more informed research
11:52especially on topics that you don't already know much about.
11:54Like maybe don't
11:56just buy 19
11:58backpacks immediately when asking if it
12:00can fit in the back of a Tesla. Maybe check it's work
12:02first. Thanks for
12:04watching. Catch you in the next one.
12:06Peace.