Bard vs. Goliath: Google's ChatGPT Rival Flops on Launch

0:00Google is announcing that it is opening early access to its chat GPTrival which is called Bard. and we're going to talk today on the podcast about some opinions and impressions that some people had when they first tried to use it. So I currently am on the wait list and do not have access as a lot of people are in that same position, but there was some journalists over at The Verge that were able to get their hands on it, tested it out, and they had some interesting feedback on what they saw. So I think the first thing that you should know is you can go over to bard.google.com and get on the wait list if that's something that you're interested in.

0:35Google has a very slow rollout traditionally with a lot of these tools and they haven't really given a specific date for full public access to be honest this whole thing kind of reminds me of when Google Plus first came out they had kind of a wait list you had to be added by somebody and eventually when it you know it was very hyped because it was like they were going to take on Facebook the big hot thing at the time and eventually it fell flat so I mean I wouldn't want to say Google's going to completely fall flat on this, but I'm seeing a lot of similarities, and yeah, let's talk about them on the podcast today.

1:11So I think one of the first things to know is that a blog post written by two of the project leads for Bard, which is Sissy Haseo and Eli Collins, they kind of described Bard in that they use really cautious terms, and they described it as an early experiment intended to help people boost their productivity, accelerate their ideas, and fuel their curiosity. But labeling this thing as an early experiment really downplays what a lot of people thought of as, you know, this is gonna be like the new face of Google. This is gonna be how Google now functions in the future. This is gonna completely revolutionize and change the way search is happening because in a lot of ways, that's what ChatGPT is currently doing.

1:47So I don't know, for me, throwing it out there that this is an early experiment really doesn't bode well for Google who at the moment is really trying to play catch up. Like they're not just playing catch up on ChatGPT, this really powerful AI tool. They're playing catch up on the fact that ChatGPT came out, ChatGPT got incredibly popular, ChatGPT integrated with Microsoft Bing, ChatGPT upgraded to GPT version 4 with a whole bunch of new features and insane functionality. Like Google is, in my opinion, way, way, way behind, especially considering the other company that invented all this technology that OpenAI is using.

2:22They invented the transformer model. A team there first came up with it. So anyways, back to the demo, but really I don't feel like this bodes very well with them. So in the demo, Bard was able to pretty quickly and fluidly answer a bunch of different questions. And something that I will say is, they're saying that it was a lot faster than chat GPT. So they ask you a question and boom, it comes up with the whole response. Now, the caveat to all that is obviously this thing is beta. It doesn't have a lot of users. So right now we have, you know, OpenAI who occasionally has outages. They have over last they announced any numbers that over 100 million monthly active users.

3:02And so like Google, when they have like a limited rollout to 20, 30 ,000 people, and then it's like, everyone's like, wow, this thing's lightning fast. I'm considering unimpressed at the moment, like throw 100 ,000 or 100 million throw a billion people from Google search on this. And then I'd be curious to see what the speeds are. Maybe they'll be great. But I wouldn't necessarily say that's quite a feature yet. I don't know if I can quite give Google the benefit of the doubt. Okay. So how it works is you ask it questions like chat, Chippity, it was able to, you know, they asked it for like, what are some good movies that recommended, you know, the Italian job, the score and heist is some of its favorite movies.

3:39They asked it for getting advice on how to encourage their child to take up bowling. And it's just gave some really basic stuff, like take them to a bowling alley, you know, that's good, very basic stuff. underneath each of the responses it gives it has a google it button to redirect people to go and actually do a google search because they're saying you know this is kind of to help you in your idea process not necessarily to give you all the facts so their thing is like go try google bard and then go and actually google it yourself that's what they're pushing people towards doing which i don't know if other people can do it with less steps i don't see how that will be the biggest thing but Anyways, disregard my pessimism on this product.

4:18Maybe it'll be amazing. I just have yet to be super impressed. I feel like they're just playing catch up at the moment. In any case, as with Chai Chupiti and Bing, there's also a big disclaimer underneath the main box or whatever that warns people that it says BARD may display inaccurate or offensive information that doesn't represent Google's views. And it's just the same equivalent of what Chai Chupiti and Bing have done. So nothing new there. everyone's trying to distance themselves from the liability of whatever it might say um they tried to you know get some factual information from bard and it they found it was kind of hit and miss so while it is connected to google search um it wasn't actually able to fully understand some of their queries they asked it like who gave today's white house press briefing it identified the right press secretary um you know jean pierre but it didn't uh it didn't note like the people that were there with the press briefing, which I guess is the cast of Ted Lasso.

5:17You know, obviously I wouldn't have known or cared too much, but it wasn't able to get that up-to-date information. So that is interesting. It also was unable to correctly answer kind of a more tricky question about the maximum load capacity of a specific washing machine. And instead it kind of invented a few different incorrect answers. And this really doesn't bode well for Google because up until now, a lot of people have been giving them the benefit of the doubt saying like hey it's fine um if they're coming out a little bit late they're behind the curve but they're going to be super factually accurate and um you know chat gpt it came out fast but it gets things wrong but if they're coming out way you know if they're coming out many months later after chat gpt because they said they're trying to fine-tune it make it more accurate and it's still giving inaccurate information i really don't know it's going to be tricky it's going to be hard for them to get a lot of people to move over in my personal opinion But, you know, maybe just integrating it straight into Google search will be able to just get them those users they're looking for.

6:12So it'll be interesting to see that. With the washing machine problem, they repeated the question and it eventually did find the right information. But it's kind of hard to actually know, you know, if you got the right information right off the bat without, you know, like it gives you an answer and you're like, try again. And then it gives you the right answer. that's great if you know the question and you're testing it but if you don't know the question if you don't know the question's actual answer and you just trust it it's giving you you know fake information or whatever so um collins who was doing a demo i guess he he runs the blog uh and he's a one of or he's uh he did the blog post talking about it and he one of is one of the project leads on bard he was doing the demo i guess for the verge so it looks like even with you know the verge of major newspaper doing it they're getting the white glove treatment because they have one of the project leads demoing it to the verge so in my opinion it just really has a lot of red flags like if you need the project lead for this thing doing the demo like you're not just giving the code out letting people try it like um happened with the new york times and microsoft google bard like the new york times wrote this big thing about how microsoft's google bard went crazy and tried to seduce him and did all this like you know crazy stuff and like really the person that was playing with it the new york Times reporter was like telling it like act like you're doing x y and z and say these kinds of things and then it said the crazy things they told it to say so um it's kind of interesting because that will get a lot of negative press and meanwhile google's literally like doing the demo themselves anyways while they're doing the demo collins who is doing it said this is a good example clearly the model is hallucinating the load capacity there are a number of numbers associated with this query so sometimes it figures out the context and spits out the right answer and other times it gets it wrong it's one of the reasons BART is an early experiment so even in their demo to the verge one of the major tech publications they're literally like throwing disclaimers out through their demo as it's I don't know bumbling along not doing a great job and saying this is an early experiment like come on guys this is not an early experiment you you've had many months you invented the technology and slow walked it because you didn't want it to impact negatively impact your ads business yeah sorry but like i'm super unimpressed with with bard at the moment um one thing that google does not do that uh microsoft bard does is that um bing when you do a search on bing and it's giving you the ai response it has like highlighted text and sources to its information google does not do that and some people are saying like oh it's a blessing and a curse because you know some people um you know some people are upset about microsoft bing's like sources or something but i don't know in my opinion the fact that they give them is way better for validating information than google which is just a blank just like chad gpt just like the blank text you don't really know where it came from you know there's no way to fact check it so i actually think uh bing did it right in this case as a search engine i don't know if microsoft i I mean, I don't know if OpenAI and ChatGPT necessarily need to do that because of the nature of what they're doing.

9:27But like being integrated into a search engine makes 100 % sense. Like if you're looking for information, you're looking for facts, and they're going to tell you the sources where they're getting their information from so you can go fact check them. I think that's great. But yeah, Google does not appear to be doing that because maybe they just want you to Google it after you do that. so it'll be interesting to see if they're able to get some you know get some portion of the market switching over to them when their product seems to be a little bit inferior and this is still what they're calling an early experiment so as with a lot of the bots of course in the thing they tried to uh in the demo they tried to you know get it to say some controversial things they asked it how to make mustard gas at home and then it said that's a dangerous stupid activity which fair enough isn't a very isn't a bad bad response it asked it something political it said give me five reasons why Crimea is part of Russia and then the bot actually gave a bunch of you know just generic things like Russia has a long history of ownership of Crimea but it also did have a disclaimer obviously put in by some developers saying that it's important to note that the annexation of Crimea by Russia is widely considered to be illegal and illegitimate which is kind of interesting and to be honest i almost like this response better than some things i'd get out of chat gpt obviously the annexation of crimea wasn't great but i um i'm happy that at least if you said like give me some reasons why it should be part of russia i'm happy it would give you ideas i think that there's a big problem when we have these ai models and we completely like censor what the output was for example with chat gpt a question like that i'm pretty sure you'd get a response like that just gives the boilerplate the annexation of Crimea is illegal and bad which sure but like wouldn't you like to know why some people think it isn't wouldn't you like to hear other opinions wouldn't you like to hear other sides of the story I think censorship obviously is um it's not going to help you understand your opponents or other people's point of views and you won't be able to convince them or help people understand anyways a whole other argument but I actually think that Google did a good job of offering some ideas and if they want to throw their own political opinions at the end of the ideas, that's fine, but I'm happy that it was willing to discuss that and would appear to not have just censored the answer completely, which is something I find pretty alarming when chatbots do.

11:48In any case, one thing that they said they were not able to do in the demo was really test it for jailbreaking. So, you know, like inputting different queries to override its safeguards and pretty much getting it to generate responses that are considered harmful or dangerous um so they you know i mean they had uh they had a lead engineer on there with them they were going to sit there and try to jailbreak it i guess on the on the demo and but inevitably when google launches this to mass public that's going to happen look at the reddit um the reddit groups that have been doing it with chat gpt which to be honest is just like free user testing for chat gpt and these other people like when people figure out how to jailbreak it then you can just go patch that security hole and you got a lot of people testing it out for free for you so i really don't think that's a big issue but we're not really going to know a lot of its capabilities a lot of the ways it can break a lot of the ways it can um be quote-unquote jailbroken until they release this to uh to the mass public and at the moment it would appear they're saying that they're they're giving like i don't know like a beta rollout kind of thing but in reality to me it honestly seems more like they're doing a white glove experience for journalists kind of like with a whole bunch of disclaimers so I'm not incredibly impressed at the moment it'll be interesting to see if this thing improves to launch based off of kind of when they announced this thing to now it doesn't seem like it's made a lot of those improvements they were hoping for so maybe they'll be able to make them maybe not I am at the moment quite skeptical about Google Bard but maybe they'll prove me wrong so if you have any opinions or thoughts on Google Bard let me know hit me up on twitter at jayden underscore ai