Yes that will be cool for sure.
But now Joe has gathered some users on his app and asks for a O(n^2) feature and starts wondering why things are slow and expensive…
Will Caffeine warn Joe? How far can Joe push it by talking at the frontend?
Is there more to software engineering than what is expressible in natural language? Will AI efficiently compile English to machine code? Will the user be specific enough?
I hope Ai can act as an assistant to debug and recommend things in that case.
(post deleted by author)
A system that protects data from LLM errors can do so without the overhead of consensus and replication
Perhaps that’s where the Caffeine dev can target certain functions to increase efficiency of functions. Of course, I’d like to think that it considers efficient moves first rather than complex, slow iterations. This is just conjecture though, which I’m sure those are the types of things the DFINITY team have been ironing out.
But we have to keep in mind that the target audience isn’t the power user, but any random Joe Shmoe for building apps. And they don’t really care about if their iterations are O(n^2) vice O(log(n)). They’d pay for it anyway in the short run and then target those functions in the long run. And if Caffeine can’t do it, any LLM familiar with Motoko can and they can slap the code into the app.
Future always belongs to top talents. Just because you know you want a button or can make a birthday app doesn’t mean it’s anything useful. Birthday app will be used max once a year and probably after one birthday gonna be forgotten and never used again.
I just hope that Caffeine wont be focusing on birthday apps but will target talents instead that can lead web3 with their ideas who can bring real attention and money.
Average Joe ain’t gonna bring anything to the table except an app that he used once and forgot.
That shouldn’t be focus.
The way I see it is even if Caffeine for some reason underperforms and doesn’t bring mass adoption, it will still help the ICP eco. When people complain there’s not enough dapps on ICP, we will now have the solution to that with Caffeine. Just use it to build your own dapps.
This will also help boosting key metrics, like dev. activity, token burns, revenue, etc.
It’s not the focus. It’s one of the usage scenarios. It’s just that the demo only covered that one.
What happened to the June 3 release date? Is it actually released and available now?
Since when june 3 was release date?
Caffeine will go from (1)Alpha to (2)Beta and then (3)it will be released to the public. We haven’t even completed(1) yet.
From what I’m guessing, It will go from Alpha to Beta and be released as “Beta” for a very long time in order to get the ball rolling as I’m assuming Caffeine will be collecting data from actual usage during Beta. I doubt Caffeine will exit Beta for quite some time - maybe even a full year or more?
I’ve seen other A.I. models be stuck in “Beta” for seemingly forever as training can take a very long time but for good reasons.
As for many of the comments esp. @Samer , can’t you simply just be positive and optimistic and grateful for what has already been built and achieved by the team?
What exactly have you contributed other than your negative attitude and commentary?
Makes me disgusted as to how this community behaves. So distrustful, judgemental, always expecting more and more no matter how much you give them, and just plain nasty.
When you train a child, do you not start with the A,B,C’s? Hello! You don’t teach the A.I. Shakespeare before you teach it the alphabet and some Dr. Seuss. I fully anticipate Caffeine to be able to produce massively complex apps eventually that will surprise everyone.
Bravo to DFINITY and DOM for an amazing demo and for all the hard work necessary to achieve it. Please stop being disrespectful and unappreciative. Root for the team.
I don’t know. Caffeine specifically uses the Motoko type system (stable signature) to guarantee no data loss on upgrades. I haven’t seen any web2 equivalent doing that kind of thing or even focussing on lossless automated backend upgradability. Doesn’t mean that there isn’t one or that there won’t be. Just doesn’t seem to be the focus of web2 companies at the moment.
Upgrades are rare. I don’t see any problem with those going through consensus. Noticeable overhead occurs during running the backend. Are you asking why decentralize the backend at all?
I’ve been Vibe Coding since GTP 3.5 / Q4 2022 so here’s the real feedback from an experienced user and developer.
It has to compete with other tools like Cursor, Windsurf, n8n, etc etc, and the list goes on, these tools allow you to use any LLM that is available on the market.
What’s its unique value proposition? If you need $ICP to deploy you’re still required to know about crypto overall
1- Centralized exchanges to buy ICP
2- Learn how to manage a Defi wallet to send it from the CEX
3- Learn the intrinsic basics of ICP → PIDs/AIDs/Canisters/Candid etc etc
Then you need to know some basics about development, a little bit of a good taste for a frontend and many more skills that you’ll be required to learn on the go.
Caffeine is not for everyone, this is only for a few specialized developers and IC community members that want to speed up their projects.
And if you’re a seasoned, oldschool developer, most likely you’re reluctant to use AI because you know you can code better, cleaner but maybe not faster.
As for me, I don’t really need it, I can build Motoko canisters with a nice frontend on any framework using any other AI tool.
That leaves the actual niche market to a small number of members on the ICP community, let’s be real, how many of us that aren’t shilling bots or paid accounts are still out there?
It sounds to me yet another expensive trick to boost the developer activity with no real products that can generate any revenue in the real world (outside ICP)
Happy to join the discussion and I hope you can prove me wrong on this one.
You’re right. Therefore, you build in a very easy way via Caffeine to purchase ICP on the fly and have it converted to cycles in a very simple way that the user doesn’t have to worry about in order for them to deploy their creations!
Don’t discount DFINITY and the smart people who probably are already one step ahead of you thinking about this issue.
ICP is a blockchain web3 web host that needs to be paid for hosting! Obviously.
There’s no need to complicate it. Accept all credit cards and do the cycles conversion on the fly and walla. You now have credit in your canisters that should last most people quite a long time unless they have huge bandwidth requirements and traffic to their easy peasy web3 dapps!
I COMPLETELY DISAGREE WITH YOU! This is a unique value proposition that will appeal to many, many people not just a few niche devs on ICP.
The easier you make it, the more influencers peddle it, the more people will want to try it, and it’s not ridiculously expensive and you can create some seriously powerful dapps. You may not think some of what DOM showed is spectacular enough but the majority of people do have a desire for simple dapps that improve their lives like an RSVP for a bday or a simple blog or yeah even the booking app! All great examples of simple things that people currently are paying quite a bit for in the web2 world. Now you allow them to do these extremely quickly in a web3 world and you can now have a personal family fundraising site to help fund your wedding for example setup in record time without having to pay a web developer or web 2 host. Not only do you save money but you get functionality that goes viral.
So easy to dismiss things that ordinary people see value in! Just a simple blog with crypto tipping x 1,000,000 is enough to catapult Caffeine to great heights! And I already predict people will develop multiple dapps not just 1 just like DOM mentioned. The most useful product in all of crypto is coming soon and you’re already downplaying and minimizing the accomplishment!
A more important question is what are the limitations of caffeine? What can it not do?
I bet you write code in assembly, right?
If I can use assembly, why would I bother with a high-level language?
If I can copy, paste, and deploy manually, why would I use a tool to automate it?
Everything’s pointless…
(End of sarcasm.)
It can’t fold your laundry → YET!
Anything that is possible to program in Motoko should be possible eventually. If DFINITY is using usage data to further train Caffeine, I don’t believe there is any limitation as to how much it can learn to program as the model continues to grow.
OpenAI has already stated that the growth of these models is exponential and already they have a model that is within the top 10% of the best programmers in the world and soon they predict within the next year or so, they will have an AI that will be ranked #1 in the entire world. Let that sink in.
I think the real limitation is Motoko itself. If latency is still a big enough issue for some dapps, then it will be necessary to utilize Rust however not necessarily for the entire dapp if you can mix the two together and have some functions utilizing Rust where necessary to improve latency.
Or maybe Caffeine will learn some Rust in the future as well?
Obviously in the beginning, Caffeine will not be able to build everything. That is just unrealistic but with enough usage and data fed back in for training over the next few years, I can’t see why it would not be able to build almost anything and complex dapps that no one could have ever anticipated eventually depending on the skill of the prompter. And due to the App store, even if you’re not a pro at prompt engineering, you can simply share or pay for whatever functionality you may require without reinventing a wheel someone has already prompted to perfection.
There will be thousands or tens of thousands or more on the dapp store. I can only imagine the permutations of every single possible dapp anyone could ever want and those prompt pros / engineers can make a whole business out of that in and of itself sparking a huge ecosystem where they can make a lot of money selling useful dapps.
And if you’re looking to start a business on ICP, you’re going to hire devs or do some dev work yourself and some surgery on the code anyways to get exactly what you want but prototyping and visualizing your ideas in almost real time is going to be a complete game changer for teams coming to ICP and for businesses of all sizes wanting to develop something quickly.
DFINITY could always add a “fund cycles” button onto each Caffeine project linked directly to banks via some trusted party. Sure, it’s not decentralized, but working with banks has never been and shouldn’t be expected. Either way, that makes it easy for someone to keep topping off their project. They wouldn’t need to know any of that third step since the canister managing the initial transaction would just fund the Caffeine project.
Also, not everyone can use other AI projects to build. Maybe you, sure, but how about the experience of going onto a website, speaking directly to it, and then having an app built for you with a link? Give the public that website, and all the other Ai tools are suddenly phased out of the picture and there’s no need to go to them.
So me and @jokerswild are on the same page with this.
