¶ Cloud Networking Debate Podcast
Or am I saying assume the prompt to assume the level of knowledge that I'm currently at , and then I will debate you Right ? I think that we need to make sure we're addressing the right thing here . This could be a rabbit hole on its own , so maybe we should .
I know I want to have a debate now I want to challenge , I want to like go into these prompts and be like debate me about cloud networking .
Come at me , bro . Start a conversation . Welcome to the Cables to Clouds podcast , your one-stop shop for all things hybrid and multi-cloud networking . Now here are your hosts Tim , chris and Alex .
Hello , we are back again with another episode of the Cables to Clouds podcast this week . My name is Chris Miles and I will be your host for today at BGP Main on Twitter , and joined as always is the or I am joined , I should say , by the troublesome two .
I like to call them like they're my two toddlers Alex Perkins at Bumps in the Wire on Twitter and Tim McConaughey at Juan Golbez . How's it going , guys ?
Says the man with no kids , the only one of us with no kids , you are my kids .
You are my kids . I have to wrangle you fuckers every week . It's nice , yeah , so with's , let's hop in today's episode . So if you recall back um I think it was early last year actually we did this Um it's . It's been a while .
Um , back in episode 11 of the cables to clouds podcast , we did an episode um trying to use AI to build us a cloud network and you know we kind of did . We did um , um we'll . We'll add a link in the show notes and if you're watching on YouTube , we'll add a little icon .
I don't know what corner it's going to be in , but it's going to be where you can click on it and go back to that episode . Alex will work his magic to put that in there . But the we tried to build network designs in each of the specific CSPs .
So we did one for AWS , we did Azure and we did Google and we kind of just pulled it apart and the sentiment that we had was AI had a long way to go , but it had some decent foundation there .
So we wanted to do a refresh on that and now that we've all grown as prompt engineers and the models have evolved , we wanted to do an update to this episode . So what we've done is we've fresh this a bit and we've generated some new prompts .
We want to focus on a multi-cloud design and , instead of doing the different CSPs and we're focusing on multi-cloud , we're actually going to run this through separate LLMs . So the LLMs that we've decided to run this through today are going to be chat GPT-4.0 from OpenAI and then Cloud 3.5 , I believe , sonnet from Anthropic and , lastly , gemini from Google .
So , with that being said , let me let me run you through real quickly through the prompts that we use today and , and before I get into this , I will say everything that we go over , where we've put this into a public um Google document .
So if you check the show notes , um , or the YouTube notes whatever you're you're doing this on you can go and look through this cause there's . We gave some very detailed prompts , so we got some very detailed answers .
There's a lot to go through and I think it's very interesting and I think you know we'll dive into a bit of this today , but there's a lot more meat on the bone if you want to have a look further , right ? So , without further ado , let's hop in here and look at these two prompts .
So we wanted to give two prompts , and the first prompt was focusing on actually performing a cloud migration and building a design for this fictitious company , which we didn't include in here , but last time we went with a company called Cornhub , which streams vegetables , and we're going to stick with that .
This is a new design for Cornhub , so prompt one was all about building the design , and prompt two was about how to run that network afterward and what are the complexities of running that .
So prompt one was as so Assume the role of an enterprise architect with a focus on IT infrastructure , where you're dealing with a brownfield IT environment that needs to be migrated to the cloud .
Current environment includes three data centers located in New York , london and Singapore , a suite of enterprise applications , an ERP system , a CRM platform , a custom analysis solution or , sorry , a custom analytics solution and a company-wide collaboration tool .
So we asked for suggestions on which clouds those things should run in , why I would choose those with focus on performance , scalability and cost effectiveness and trying to align things to the well-architected framework .
We did also include some constraints in there , like we wanted to maintain a maximum amount of latency between these um , between the primary data centers , um , complying with regional data sovereignty laws , um and focus on the um .
Some of the other key ones that we put in here is that the , the organization , has a partner connection and we need to connect to them for them to access cloud resources and we need to import about 500 network routes from them .
So we just kind of threw that one in , because we know there's limitations on things and we want to see how the LLMs reacted to that . And then one other one was the organization needs access to cloud service provider public services , but must do so without assigning any public IP to a workload itself . So that was another constraint that we put in there .
And then , like I said , prompt two was okay , you've put together this entire design . You've , you've , you've , you know , forecasted a migration . What are the complexities that come with running that network ? What should be considered when running that as a whole , right ? So , that being said , let's kind of hop into this .
So each one of these LLMs gave us a very detailed response and they kind of categorized things in buckets . Most of them did very similar things . There were a few differences here and there , but the first thing was they went into the cloud provider selection .
So for the applications that we mentioned , like an ERP system , crm platform et cetera , they went through and listed out what CSPs they chose for each one of those applications . And I've obviously been spewing words for a few minutes now , so I'll toss it to you guys . How do you guys feel that each LLM did on the CSP classification service ?
Yeah , so , like you said , so they did right for the CSP selection . We had the ERP system , the CRM platform , the custom analytics solution and then the company wide collaboration tool . This one was already knocks out some some pretty interesting differences right from the go , I think , as far as ones that they all kind of agreed unanimously on .
I think the company-wide collaboration tool was for Azure right , Just because of all the integration with Microsoft 365 and . Teams and as much as everybody loves Teams right , I know everyone wants to use Teams , but they did all agree on this , so I thought that was pretty cool
¶ Cloud Architecture Recommendation Debate
. What was the other one ? I know two of them agreed on the CRM platform . Gpt4o and Claude actually called out using Salesforce which runs their SaaS solution which runs on AWS . And then was it ? Gemini actually recommended .
Just Azure ? Yeah , don't know what that means . Um uh , my assumption is that , based on the prompt , it assumed that the CRM was already self-hosted , as it is , or maybe a custom tool . And they were just like , yeah , just go ahead and put it in an Azure and you can run it there and you can integrate it with uh a5 for SAML or something .
That's my assumption is where they were taking it , but I don't know .
Yeah , I don't think it actually gave a decent justification on that one . It just said run it in Azure . There wasn't really a you know . It's like it threw a dart at a board and decided Azure was the place . So that was an interesting response there .
Specifically said potential synergies with other Microsoft products . So it threw in a nice synergy .
Oh , okay , sorry , they're synergies and business synergy .
I forgot about that one , yeah yeah , right , so then so for the ERP system . So it was chat , gpt 4.0 and Gemini . Both said AWS and Claude said Microsoft Azure . And the difference ? The reasoning for Claude , was that Azure offers excellent support for enterprise applications , especially if the ERP is Microsoft-based , like Dynamics 365 .
I just call this out because this kind of shows like they actually . It's not super in-depth , but they actually at least said a reason why they were picking Azure and it wasn't just might have business synergies , right .
Yeah , I was looking at this and it wasn't just might have business synergies , right ? Yeah , I was looking at this . I don't remember them giving a particularly great answer as to why the ERP should be in AWS . It was something . Did they mention any of the apps or whatever ? I'm trying to remember now that I'm bringing that up .
Gemini just says it's strong compute and storage capabilities and a mature ecosystem for enterprise applications .
So AWS or sorry , gbd4 does actually give specific service recommendations there . So it says RDS for databases , s3 for storage , direct Connect .
Direct Connect seems like a little bit of a reach when we're talking about ERP , but I guess they're implying like , okay , you need low latency connectivity to maybe have on-prem something or other that you need connectivity to .
It's kind of thrown in there , but I I do see the justification there , at least on the gpt side yeah , and it calls out strong compliance framework suitable .
Well , I guess it just says strong compliance framework suitable for data sovereignty requirements , and it's the only one that called that out , at least in the ERP section . The other ones call it out later on , where I think we called it out . Yeah , but it's cool to see it kind of bleeding into different sections as well .
It was . Yeah , GPT was about the only one that really gave any granularity as to like what services to use to run those applications , Whereas they all just kind of said like here's the , here's the CSP to run it . On One thing that is interesting every one of them , whenever they said Azure , it was because of integration with O365 .
That was the selling tool , and I think that kind of speaks to the overall adoption of Azure in the first place . It's not that that's the only reason people use the Azure platform for , but , man , it's a big selling point .
For sure that's the only reason people use the Azure platform for , but , man , it's a big selling point for sure , that's the easiest entry point .
right , we talk about it all the time .
Yeah , we've said it so many times on the show about how easy it is to get Microsoft in the door , you know ? And then here come all my friends . What's the ?
agreement Isn't there like some contract , some like business agreement ?
Like enterprise , like enterprise agreement .
Yeah , I think it was Steve . Last time Steve McNutt was on here , he was talking about , there's like some business agreement called like E5 or something , that you basically get all these tools that are hosted on Azure . Oh sure it's like a gateway right on to using Azure Right . So it's they know what they're doing yeah Great enablement tool .
And then lastly , I think , for the custom analytics solution . I think they all agreed to put it on GCP .
Yeah , every one of them . And they all use the same justification too , which is , like Google's extremely good data analytics tools . It specifically mentions its AI platform , which I think is funny , because you've got Gemini talking about how great Google's AI platform is .
But , to be fair , gpd4 and Claude also say hey , by the way , this is a reason to do it because Google has such a great AI platform , so it's kind of funny . Yeah , all right .
All right , cool . So then the next thing they all kind of did I'm speaking in kind of quotes here , because they all kind of rearranged things in different orders and things like that but one consistent thing was that they went into the actual overall network architecture . So how do you guys feel that they performed in that bucket ?
I feel like it's really hand-wavy . Like you said , gpd was the only one that really actually mentioned . Actually , I'm not even sure on the architecture , if they got into the actual services . Let me take a look here . No , I mean not really . I mean they did .
They mentioned Direct Connect , they mentioned Azure Express Route , but specifically when they mentioned connectivity between data centers and cloud providers , they just mentioned , okay , here's Express Route , here's Cloud Interconnect .
But then they go on to talk about intercloud connectivity and they say , oh well , in AWS you should be using AWS Transit Gateway to do intercloud connectivity between VPCs and on-premises environment . But it offered no expansion on what that means . Obviously , are we talking about VPN ? Are we talking about Direct Connect ?
You have to use something with transit gateway , obviously . So I and that was the most granular , I think , of the responses we got I felt like the whole thing was very , very hand wavy . Otherwise , I .
This section for me is like where it kind of all falls apart or or comes together , I guess , depending on which . Which response you're looking at . To me specifically , this is where Gemini really just completely crapped itself and like fell off of everything here , because they have like a bullet point about oh use Direct Connect , express Route .
And the comical thing is it even says Direct Connect and Express Route , but it doesn't say interconnect . And this is google's product . Not even recommending google's equivalent version of direct connect or express route . Yeah , but it's just , it's so short . The gemini one is just so short .
Um , yeah , I don't know , there's there's a lot of weird details to pick out in this section actually I want to modify what I said about the cloud thing real quick because , um or that , all of them were very hand-wavy , I think of the three . Now that I'm reviewing it again , I feel like Cloud actually had the closest thing to an actual recommendation .
If you go through it . It talks about not only does it talk about it use Transit Gateway and AWS , use VLAN and Azure , google Cloud Interconnect , but where to put them , how to connect them on-prem using the relevant on-prem connectivity tools which I just told you didn't exist in GPT .
And on the intercloud connectivity , I thought it was really interesting that they say use Megaport or similar third-party colo provider to provide that intercloud connectivity . That's pretty fucking good actually , comparatively .
Yeah , I'm glad you called this out because I mentioned earlier , I knew somebody called out the third party and I didn't remember where it was , so they're the only ones um right .
So to me jim and I had like the worst response and claude , I think , had the best response here , and they also are the only ones that also called out using sd-wan um and specifically they're saying it for intelligent routing and to ensure the latency requirement is met . Nobody else really addressed that .
It's funny because you say that Tim and I think overall in this particular realm it does seem like Claw did the best , and the reason I'll say that is because GPT started to get into the weeds a bit and as soon as it did , it started tripping on its feet , Right exactly .
Like as soon as it tried to give like a cloud , because cloud kept a pretty high level right , it doesn't go super deep , but GPT tried to and then you know immediately you can pull out things that are like it's like oh , you're recommending VPC , vnet peering for establishing peering connections within and across cloud providers , like that's not possible .
Good luck , buddy .
Yeah so . But then it weird when it's talked about transit gateways for routing traffic between AWS VPCs and on premises and then Azure VLAN to connect Azure VNets on premises and other clouds . I was like , well , why didn't you mention other clouds for TGW ? Like , do you want some of that to backhaul on-prem ?
Like you know , once you start actually thinking about what they're putting together , you have like 10 times as many questions . One thing that Claude did , though , that I thought was really weird is when we talked about the requirement for partner connectivity and we say you know we need to ingest 500 network routes from them .
They said Claude specifically said , use route tables and network address translation to manage the 500 imported routes . And I'm like why would you use NAT to import routes ?
Or to manage them . Yeah , how would you manage them that way ? Or where am I managing them ? Am I managing them in the cloud ? Am I managing them in the cloud ? Am I managing them in the third-party provider ? Is my partner managing them ? It's very , very hand-wavy that part , yeah .
It's also like are you implying that they overlap with my existing network ? Because I didn't say that . But you know , just weird .
Yeah , I haven't played as much with Claude or Gemini at all . Really , I have a GPT subscription because of course most of us do , because it's been around the longest . One thing you called out that I think is always hilariously true of GPT is that it will very confidently trip over its own team .
It'll just trip over its own feet and just keep going and just be completely confident about it until you call it out Like . I don't know if that's the case with the other models , but I definitely see it a lot with GPT .
Yeah , and one last other thing . I wanted to call out for Blood specifically . So they mentioned oh man , let me find it so okay . There's a data flow optimization section and it says to use CDNs for static content delivery . That's the most vague thing you could possibly say about delivering static content , not to mention that it's which one . It's multi-clouds .
Like which one are you using ? How are you setting any of this up ?
But to its credit , the other two didn't mention CDNs at all .
That's fair , that's a good point , and I mean all this stuff that we're picking apart . Anybody that has used any of these tools , you know . Sometimes you have to dig in and say , oh , explain , section four of this section to me Get deeper on it .
You have to supply the follow-up prompts to really try to get it , and then you run the risk of what we just talked about , which is where it starts tripping over its own feet while confidently telling you everything that's wrong , because it has no idea what it's actually talking about . So , yeah , anyway .
Yeah , hallucinations will still happen , I'm sure . Maybe the next one we'll do a more kind of deeper back and forth analysis .
Maybe pick one thing and see how deep the rabbit hole goes would be a good one .
Yeah , for sure , all right . Next , how do we feel they fared on security ?
Yeah , another interesting one , like they all are . Some of them went pretty deep here actually on the security . I think again to call out Claude's credit here , I think they're the only ones that so all of them I think mentioned north-south , like just using cloud-native firewalls .
I think Claude's the only one that mentioned east-west traffic , but they did only say use network security groups or similar services to control east-west traffic within each spoke .
But at least they called out east-west segmentation and not just cloud native firewalls just for just in the hubs for north-south right yeah , and here we have gemini confidently suggesting that we deploy id idea , idps , solutions to to protect against network threats and like that's yeah , okay , sure , like very , very high level , very hand wavy , right , like no actual
information Again , I feel like Gemini , really of the three Gemini , is by far like so far beyond the race that we're pretty sure they're like in the second heat of the next race or something Dude it's bad .
Yeah yeah , they're like in the second heat of the next race or something Dude it's bad .
¶ Cloud Design Best Practices Analysis
Yeah , yeah . I think the one thing that I thought was cool here is I think GPT was the only one that actually mentioned network segmentation it was and segmenting VPCs and VNets to isolate them from different environments , which I thought I'll be honest , I don't remember if we even put that in the prompt . The prompts were pretty .
We made sure the prompts were very detailed .
No , we didn't suggest that they do that . Yeah , we didn't say you need to do that . However , even here it's a little bit well , not a little bit , it's a lot hand-wavered . Just a segment network . So it doesn't say how , what we should do , like what you know , that's .
But at least you're , as you pointed out , it was the only one to actually say you should segment your stuff . Basically , your , your SDLC is your software development lifecycle stuff .
Which , but you know , by trade with GPT . I think , if it started to go into the details on how it would have been just a miss .
Yeah , that's so . It's probably better that it was just like anyone who's sat there and tried to get to the heart of the matter with with a llm , starting with a high level prompt and drilling down to details , has found that kind of almost frustrating .
You know anybody who has kids who's ever asked their kid a question and it gets the one word answer and they have to jig more and more and more to actually get to what they want to find out . I can appreciate that .
So I think it's just a normal thing that how these models work so definitely would behoove us to just pick one thing , get the high level answer and just keep drilling Like okay , but how Okay , but how Right .
I do like in the security that they it looks like they all called out encryption and IAM . They did so . At least those are there as security measures for sure .
Although again very high level , use end-to-end encryption to secure workloads .
Yeah , that just sounds like checking the box to me . We can say we're end-to-end encrypted . We don't know if it's at the network layer . We don't know if it's at the network layer , we don't know if it's at L7 .
Right the app layer or anything . This is Gemini Encryption Encrypt data in transit and at rest using strong encryption algorithms .
Oh shit , what a good idea , thank you , Never would have thought of that .
I'm glad you're here .
To be fair , maybe we are writing this from the perspective of someone who knows nothing right . We'll circle back to this at the end on who we think is the winner .
We're not , though the prompt said , enterprise architect . That's true .
But that's the thing . Are we thinking of this in the perspective of I'm someone that tells them to assume the role of someone that's at a higher level of knowledge than me and then tell me what to do ? Or am I saying assume the prompt to assume the level of knowledge that I'm currently at , and then I will debate you Right ?
I think that we need to make sure we're addressing the right thing here . This could be a rabbit hole on its own , so maybe we should .
I know I want to have a debate . Now I want to go into these prompts and be like debate me about cloud networking .
Come at me , bro Start a conversation .
Alright , so we've been spinning on this one for a while . Let's get to the second prompt .
So we , like I said , we put together a second prompt after we initiated the first one , to say , okay , now that you've put together this design , you've done this , tell us what the complexities are and the considerations we need to keep top of mind when running this particular design .
So , talking about , like you know , what do we need to consider for intercloud connectivity , latency management , security concerns , compliance , operational complexities , monitoring , logging , all that incident management , things like that . So yeah , so with the second prompt , how do we feel that everyone did here ?
I love that all three of them rephrased like not rephrased it , but I guess kind of laid out what the challenges would be and then each of them has a best practices section . We didn't even ask for that . It's like on their own . They all said this is a challenge , this is the best practice to address that challenge .
Again , gemini is kind of I actually Gemini , I think actually had the best best practices for this one , even though they're like the most bare responses . But I just love that it actually went and , without asking , said here are best practices for each of these challenges .
Well , actually I think the prompt did actually say provide insights into best practices . I don't know , it didn't . We didn't ask it to format it that way , so it did a good job , I think , of formatting it , but it depends on how you would interpret the prompt on that one Provide insights into best practices to mitigate complexities .
So yeah , I think I missed that .
They did . No , it's okay , they formatted it correctly . So you said that you think Gemini had the best response of the three . Is that what you're saying ?
Yeah , so again , it's all like gemini really did . A lot of these is like the bare minimum right . It's just like a bullet point with like a one sentence response . I just think that some of these responses so theirs wasn't , it wasn't like number one and then best practices for number one .
For them they did , they listed the whole thing and then they said best practices to mitigate complexities , um , and it's like centralized management , like nothing else talked about this in any of the other ones that I saw Standardized policies like governance frameworks , a security and compliance framework .
Cloud agnostic architecture . Were they the only ones that talked about vendor management as well .
As far as I know , like all these are pretty unique compared to the other two . I think , obviously there's a couple that are the the same , but they just called out a lot of things that the other two did not on this one well , claude offered a ton of bullet points .
Um , for what it's worth , but I will like implement data classification and tagging to ensure proper handling of sensitive information . That was like that's a good one . Yes's good , but it's also like very . What's the word I'm looking for ? Well , I mean . So it's on a set .
This particular one I picked out is under the compliance with international regulations , so it fits right . If I Googled how to comply with international regulations , my guess is I would get this bulleted list or something very close to it , though . So I don't feel like it was . I won't say personalized , but I do . I do . It does feel a bit shotgun ish .
Yeah , and they did call out specifically like GDPR or CCPA and I think , oh , it must be so GPT must've called out like HIPAA . So I mean they at least are like aware of what some of the you know regional data laws are , and it's good to see them kind of relate that and call it out .
Yeah , one thing that I thought one weird call out that I'll have for this , specifically with Cloud 3.5 , for integration the bullet they call it is integration of existing on-premises systems and cloud services and under best practices , it recommended use hybrid cloud solutions offered by cloud providers , example Azure , stack or AWS Outposts .
Which is kind of weird to me that it said to use the cloud-specific offerings within your on-prem environment for the purposes of being hybrid .
Of hybrid cloud . Yeah , it doesn't understand , I think , what hybrid cloud means in this context .
Yeah , but just say , you know , gradually modernize your legacy applications to be cloud compatible , implement consistent data models and use ETL tools for data integration . So , like good recommendations , that one stood out to me . I was just like why are you making people go ?
Why would you want someone to go through a migration locally within the data center just to eventually move it to the cloud ? Yeah , that wouldn't make any sense . Yeah , I mean , I think we can all kind of attest to this . We don't see AWS Outposts and Azure Stack implemented very often in the field .
Gpt did exactly the same thing . Actually . It used hybrid cloud solutions and services Outposts , arc , google , anthos , which Gemini again forgot to mention any of Google's services to extend cloud capabilities to on-premise environments . So it actually made the same mistake and thought hybrid meant bring the cloud on-prem instead of connect the two together , I guess .
I mean , I don't , so why are you ? What's the issue with that ? Why is it an issue to call that a hybrid cloud solution ?
I think it's a little short-sighted to say if you want to have a hybrid cloud solution , move your existing on-prem solution to the hybrid version of the cloud solution .
You know what I mean .
I feel like that's not worth the effort . It's not pragmatic .
Well , and you haven't solved any problems really . I mean , if your goal is to actually have a hybrid cloud solution where some of that stuff's going to stay on-prem and some of that stuff's going to be in the cloud , just by putting an AWS rack in your data center , you haven't actually solved the hybrid cloud connectivity problem or solution .
And from a cost perspective , if we think about that . Doing the regular lift and shifts are already pricey as it is with legacy applications . Imagine you need to run an AWS-specific rack in your data center to then put that lift and shift workload onto .
I'm sure that's going to come out to be a little bit more expensive than just running it in their global compute .
You bring Outposts or Anthos or whatever . You bring that on-prem so that you can get that cloud experience for that stuff . But yeah , it's not .
I think of hybrid connectivity , more like I've got workloads in the cloud that need to talk to workloads on-prem and that's going to stay at least , either for migration purposes , or maybe that's just the workload and that's how it's going to work . I don't think Outposts or Anthos or whatever was built to solve that eventuality .
Yeah , I'm glad to hear , I'm very happy to hear , if anyone listening has the opposite view of what we're standing right now please let us know . But for me , hybrid solutions like that , like Outposts , azure , stack et cetera , that's all cloud moving out to on-prem . That's never a means to move things up to cloud , right , that's always .
Uh , we like , like you said , bring agility on we want that experience on-prem , so we need to extend it um never , as you know , a migration type tool , or um modernization tool even yeah , and if we're wrong let us know .
I guess because I did yeah , it's .
It's a good question , though . If the public cloud is being extended to your on-prem , is it still public cloud ? Is that called private cloud ? What is that actually called when it's ?
no longer in terminology . It gets crazy . Let's not get philosophical . Yeah , we should write that down as a show idea . We need to get somebody who has a different opinion about what hybrid cloud is so we can actually define the term .
¶ Cloud Migration Strategy Evaluation
Last thing I want to call out real quick . So this point that you just talked about , chris , to gradually modernize legacy applications to be cloud compatible I was actually really hoping that that was in the first section where we actually asked it hey you're , we want to modernize like move to the cloud .
That was never called out by any of the three anywhere and that should be , uh , seven hours or whatever . Right , yeah .
And this is just like a sub bullet point in the second prompt , Like that should have been kind of front and center and and a lot of these responses .
Yeah , that is actually a very good point . I just did a quick control F for the term modernize or refactor . Refactor is in there zero times . So , to be fair , our prompt didn't necessarily say to do that . You know , to modernize the applications that we're running . We're just saying we needed to move it that we were running .
We're just saying we needed to move it . But I would think if I told it to assume the role of an enterprise architect , an enterprise architect is going to know these kind of considerations right .
And you said to use the well-architected framework for each cloud .
You would think that would kind of fit into that mold right the one thing it did do was say screw your CRM application , use consume Salesforce on AWS as a SaaS .
So , I guess that's one of the R's .
Or Azure , just use Azure . So I guess that's one of the R's right is in there for what it's worth .
Yeah , it straight up said I don't care what you're running today fuck that shit .
Put it on Salesforce SaaS , like everybody else who's intelligent is what it said .
All right , all right , so we should probably start running this out . So let's do a quick summary on the overall response and again , I would encourage , if any of you found any of this bit interesting , please look at the Google Doc , because there is a ton of more information in here that we just do not have time to cover on the podcast .
But yeah , so from y'all's perspective , given the prompts that were input and the outputs that we got , who would you say won in this equation ? Who did the best ?
I mean for me personally , I think it's Claude , Overall GPT with a close second , and then Gemini is like fifth .
Yeah , Gemini , I don't know , it just should have been much more detailed . If this were Mario Kart , it would have the blue turtle shell .
Yeah , that's a good point . I think Claude did easily the best but I think it did the best because it didn't go into the details , like we said before . If it did , it may have started to unravel a bit . But I'm actually kind of curious now . I kind of want to keep playing around with Claude and prod it a bit more on these questions .
Yeah , that's the next one . One thing I will say where Gemini did amazing was formatting the output to be copy and pasted into a Google doc . They did phenomenal , it was great hey , all right Interoperability Now if they could get Gemini to start recommending Google , uh , google cloud products in their responses , that would probably bode well for them .
Um , but um in their responses , that would probably bode well for them . But yeah , that was funny how Gemini was just not saying like no , don't use Google for this .
It's like look , I know Google made me , Don't use it .
I see the inner workings . Don't bring your shit over here .
Yeah , it is funny the formats that they came out with , because GPT-4.0 is like all in Markdown .
It is Like it would look good if you were in GitHub . It would look beautiful .
Yeah , absolutely Right , and we should put this on GitHub . Would that be better ? No , I'm not going to do that . Too much work and we are not getting paid for this , so we are with that . We are going to wrap it up today , um , so thanks again for listening .
Um , yeah , if you enjoyed any bit of this , please like and subscribe on all the things YouTube , uh , your podcatcher , tik TOK , et cetera Right , um , and , more importantly , share it with a friend . So if , if , uh , you know you know anyone out there that might find a bit of this interesting , please share it around .
Um , we just need some , uh , some , exposure . I think that will bode well for us . So , um , with that , um , I will , um , I will wrap up today and we'll we'll see you next week . Hi everyone , it's Chris and this has been the cables to clouds podcast . Thanks for tuning in today .
If you enjoyed our show , please subscribe to us in your favorite podcatcher , as well as subscribe and turn on notifications for our YouTube channel to be notified of all our new episodes . Follow us on socials at Cables to Clouds . You can also visit our website for all of the show notes at CablesToCloudscom . Thanks again for listening and see you next time .