Right now, I am starting to have doubts about when (not “if”) Dfinity can deliver on game changing Ai applications, like caffeine.ai. I’ve made my thoughts known on other platforms but some have dismissed it as being rooted in price chasing and FUD.
To keep it short, my doubts stem from seeing missed deadlines on some of the major Ai stuff, like Wasm64 and Ai prompt engineering. Then the constant teasing that such stuff will be released “soon” doesn’t help when weeks and months have already passed. It is my view that taking too long to deliver might lead to ICP missing its moment to shine if other projects are doing similar. Then if the time for release keeps getting pushed back even further and further, some may start questioning IF these Ai features are even possible to build entirely onchain.
Btw, I’d love to be shown wrong by having Caffeine.ai and other major Ai features being released “soon” - as in days or weeks.
It’s not possible in days or weeks. It’s not possible to do it on-chain. Even doing it off-chain will only work for very simple dapps. So, it is basically just hype, and this is probably common knowledge to most technical people.
That said, implementing it off-chain could become feasible for moderately complex dapps within a year or so, while achieving it on-chain might take several more years.
At this point, I wouldn’t mind if the major Ai stuff are initially released in some hybrid model involving both web 2 infrastructure + blockchain infra. I say initially because then they can have the product out so that people can at least start using it. And meanwhile, Dfinity can continue working to bring it completely onchain. I would imagine that they can make the case that they can bring it onchain eventually by pointing to other things that they’ve been able to have onchain, like web applications and some Ai language models.
Maybe Dfinity will run into a limitation that they can’t overcome. Or maybe not. We’ll see!
When a contract is put on chain, the truth it provides is self evident and self effective. Whether it was generated on chain or off is of no effect.
Unless for some reason you don’t have transparency of the code in which case on chain generation can provide provenance and accountably and proof of non manipulation.
But transparency is probably better.
There are more applications for on chain generated AI in other domains.
I’ll add that since motoko is easily compiled and because AIs are actually pretty good at writing, running compile, provide errors, revise loop can be pretty strong. I have my doubts that a complete novice will be producing competent, secure applications before 2026, but toys and non-serious and non mission critical applications are certainly going to be possible. It is also the worst it will ever be right now.
As a reference AI wrote the first 80% of the test suites for our icrc-75 and 79 projects. They all take bit of oversight but the amount of work is massively reduced.
Also, teaching people to use Cursor to create simple dapps or modify existing ones with open-source code conveys a far more positive message than hyping the Caffeine AI dapp.
In a world with an abundance of block space, culture matters quite a bit.
How long should people wait, esp, if a project is passed a deadline? Months? Years? No set time?
I don’t think there’s a perfect answer as everyone’s patience levels will vary. In general, I think it’s reasonable for people’s patience to begin to dwindle the longer they have to wait for a project, especially if deadlines have been missed or keep getting pushed back.
Balancing the need for timely delivery with the need for quality should be a priority in software development.
In my opinion, it’s important for people to understand that when it comes to developing software, there’s always a trade-off between quality and time. While it’s true that there’s a point of diminishing returns where additional time invested doesn’t significantly enhance quality, the reality is that we often push to cut time as much as possible. This demand to meet deadlines frequently leads to sacrificing the quality of the product just to say it’s finished.
The solutions … from the same article…
Projects should start with approximate deadlines that can be adjusted as they progress, allowing teams to reflect actual progress and address challenges as they arise. As the project advances, these deadlines can be refined based on real-world progress, giving a more accurate prediction of when the project will be completed. By defining a realistic rate of progress early on, teams can better estimate the final completion date. This flexible approach helps to balance the need for timely delivery with the importance of producing a high-quality product.
Clear communication with stakeholders is essential, emphasizing the benefits of prioritizing quality over strict deadlines and setting realistic expectations from the start. Regular check-ins and transparent communication throughout the project help identify potential bottlenecks early, allowing timelines to be adjusted as necessary.
That last point I felt wasn’t done in a clear way. There were some Twitter posts made by Dominic that they are trying to make the product “sufficiently mature”, but I’m not sure everyone could see a deeply buried comment. Not to mention all of that might be drowned out with the constant teasing of “coming soon” shown with a picture of a studio as was the case in one tweet. That felt like an event where you are hyping the crowd but then keep them waiting long, and on more than one occasion.
