2moon

⚔️📈 🛡️ 🌛Let's Moon!🌕 Professional Speculator. Super Trader. $ICP $ETH $BTC

[2moon @Crypto2Moon]

If I need to tweet only one. Buy and #hold $ICP.

[Hunter Kallay @lazukars]

I DON’T UNDERSTAND WHY PEOPLE DON’T USE GROK FOR STOCKS.

Most traders are looking at charts from 6 months ago.
Grok analyzes real-time sentiment on X to predict future.

Here are 20 prompts to find the next 10x stock:

[2moon @Crypto2Moon]

@dorloechter Where can we access to this dashboard ?

[ah kek @ahkek4]

Introducing Smart $PENGU Tracker 🐧

We’ve curated 100+ smart money wallets holding PENGU

Track their moves in one dashboard with instant Telegram alerts

For just 2888 PENGU (<$100), unlock full access: smart money insights, alerts, charts & data

Site: https://t.co/4wlXUnudy1

[2moon @Crypto2Moon]

@sjdedic @peaq Are they making any revenue yet ?

[2moon @Crypto2Moon]

@Robotexon Can you pls share the official news of @Robotexon participated the IBM's elite Startup Grant Program?

[2moon @Crypto2Moon]

@Robotexon Can you pls share the official news? Can't find any.

[gizzy.icp @gizzycrypto]

My @caffeineai journey, day 10:

🚀 near 2 weeks, 226 drafts, countless frustrations… and finally the breakthrough with #CaffeineAI.

At first, I struggled with prompts, slow builds (30 mins for a pixel change 😅), and endless retries.

But then it clicked. I changed my style, refactored big files, and suddenly Caffeine became powerful: compact code, fast drafts, real progress.

The real challenge? Integrating tokens on #ICP:
💠 ICP
💠 BOB
💠 ckBTC
💠 ckETH

After dozens of errors and help from friends, today my app finally reads real balances – I sent 0.001 ICP and it showed up 🎉

Just when I wanted to continue… the chat died 🫨😭🤯🤬. Support ticket opened. Waiting.⏳⏳⏳

This journey is far from over. But one thing is clear: building with #caffeineai isn’t easy – it’s intense, frustrating, but absolutely thrilling.

Anyone else faced similar issues with CaffeineAI or token integrations on ICP? Let’s connect.

#caffeineai #ICP #ckBTC #ckETH #Web3 #Crypto #BuildWithCaffeine

@JanCamenisch @dehypokriet @dominic_w @itsmejeremy77 @X2worldtech

[2moon @Crypto2Moon]

@dominic_w When native usdc ?

[John Tuld @BradHuston]

Internet Computer bros are mad that the U.S. Government went with $PYTH a crypto chain based in Baar, Switzerland to put government data on its blockchain. Once again, Dom and @dfinity have been out hustled as Dfinity continues to mock traditional marketing efforts. I have a $3.50 sell target...and I would expect heavy selling into the last part of the year. Hope $ICP holds the top 100 with a major crypto run, that they won't be participating in, this fall. Just jaw dropping the utter ineptitude of the Dfinity Foundation.

[2moon @Crypto2Moon]

@AaronTing8 We might running out of time in terms of Marco..

Good product needs marketing…

[Ben Z @batman_icp_maxi]

It took me approx. 20 minutes to code it with @caffeineai.

$ICP is far above all the blockchains.
One day everybody will know that!
Thanks @dfinity

[John Tuld @BradHuston]

Hey @dfinity, you can't just go dark now that you have launched Caffeine AI. Your competition has defined you for 4 years and are FUD'ing harder than ever. You need to communicate daily, not go on vacation. Marketing matters. $ICP

[2moon @Crypto2Moon]

@JustinJackBear Price is a best marketing tool to grab eyeball. Before that we need native USDC.

[geekchains | ICP ∞ 🇧🇪 @Geekchains]

ICP BREAKING 🚨 : Man clones sophisticated e-commerce website in just two prompts using Caffein AI.

@ericschmidt said it.
@caffeineai delivered it.
Internet Computer Protocol $ICP ♾️

[stoik.icp @stoik_eth]

A sample of @caffeineai from prompt to launch it took about 30 mins to launch but I wrote an extremely thorough an elaborate prompt. I wanted to QA test if I could crash it. Nope it did QA and solved its own issues for me. My mind was blown.

[Snassy.icp @SnassyIcp]

From watching the streams and talking to friends who were at the event, Caffeine seems to be what I expected.

Long post ahead, TL;DR Caffeine is great, ICP is greater.

I have tried to think of ways of saying this without coming off as negative when I am positive. So let’s start by saying I think Caffeine looks fantastic, that Dom did an amazing job demoing an extremely cool technology, which seems to really work and to really resonate with people! Congratulations are certainly in order!

I still think, however, that ICP is the real MVP.

What makes Caffeine so great is that it targets ICP as the deployment platform. That is what allows it with ease to create apps that don’t forget your data, that evolve with new features safely, that don’t require SQL databases and O/R Mappers, and everything else that makes apps created with Caffeine so great!

So what’s the problem with that? Nothing! It’s an amazing way to showcase and put the power of ICP into the hands of everyone! I am thoroughly positive about all this.

All I want to do is caution against suggesting that it’s *Caffeine* that makes all this possible - even if that may be trueish right now.

In short the point that I want to make is that while ICP has a moat as deep and wide as an ocean, and is true alien tech, Caffeine may also be alien tech (modern LLMs are) but in this case the alien tech is not really unique to Dfinity - the moat is not nearly as substantial. And that’s OK, because Caffeine is only here to usher in ICP the wonder-platform, and if some other product takes over the role of ushering in ICP, then that’s fine too.

What I mean by Caffeine having no real moat is that I, and every ICP dev I know, already do what Dom did in his demo on the daily, on the ICP, but using Cursor.

Half a year ago it was still true that an app like Dom did would take a decent amount of negotiation with the AI. Being a developer definitely helped, although being patient and careful in explaining bugs to the AI would be enough.

But Cursor has a neat little dropdown that allows you to pick the AI you want to use, so when new, better AIs come out you just pick them, and the current ones are about at the level of what we saw Dom do.

Today’s AIs in Cursor can usually one-shot an application like that, with similar amounts of instructions from the user. On ICP, with all the bells and whistles.

The modern AIs are trained on a lot of information about ICP and Motoko and know it well, and when they don’t know they search the web to find out.

To Cursor, ICP is just one from a basically unlimited number of target platforms you can choose to deploy to, you just tell the AI what you want.

So is Cursor really quite as easy as Caffeine for a non developer that wants to go from zero to decentralized app on ICP? No. Almost, but not quite.

Dfinity has in my view done mainly 3 things that make Caffeine easier to use for now.

First of all, what I’m pretty sure they haven’t done: trained their own AI model. That’s not just because it’s economically infeasible but also because it doesn’t really - to the best of my understanding - work. And finally it’s because they don’t have to.

I’ll explain. When training an AI, you first feed it with a lot of data, then you start giving it inputs and tell it which of its outputs were good and which were bad, and the AI adjusts its enormous database of weights that it uses to classify the information you fed it.

If you see the AI as a big database, this phase involves both reading and writing data, refining it until it’s good enough to use. This is opposed to when users chat with the AI, which is just querying the database in a read-only way - no learning, or redistribution of weights, is going on here.

Most niche AI products do not train AIs, they only wrap existing LLMs with a hidden system prompt that they send along with what you typed, saying things like “you are a Gardening expert. Remember that roses are red and violets are blue.”

Real training is a very costly process if you do it for a big AI like ChatGPT. It’s unlikely Dfinity has done anything on this scale. But you can also train smaller AIs, which is a lot less costly, unfortunately these much smaller AIs are generally a lot less smart.

But what if you train a small AI on some narrow domain? Could you make a small ICP expert, that doesn’t know much else, at reasonable cost?

To some extent yes. But here’s what I understand from trying to keep abreast: If you train a big, ChatGPT scale, generalist AI to include your topic, say coding ICP, it will perform significantly better on coding ICP than a small model that was only trained on ICP. Knowing more things in general helps with doing specific tasks.

So even if Dfinity did train a small model to be an expert at ICP, it would soon be outcompeted by a large model that knew ICP. So Dfinity probably didn’t train a small model either.

And, as I said, they didn’t have to train any model, because they have put enough information online about their platform and their Motoko language that the latest AIs know ICP well.

Furthermore context sizes keep growing so they can put more and more information into system prompts.

What all this points to is that Caffeine calls one or some of the current large models - maybe the user will even be able to pick which one they want from a dropdown, like Cursor.

I thus feel fairly confident they have not trained their own AI. So what have they done? I think the following 3 things:

1) System prompts filling in the ICP knowledge gaps in the large models and telling them to focus on ICP deployment.

2) A very neat one-click to deploy feature, with staging and production environments.

3) A simple web IDE where you can type in what you want your app to do, and see the code the AI generated as well as links to test your app.

That’s it, I think.

And while Caffeine is a great product, most of that moat is crossable with a Cursor plugin containing similar system prompts and a button to deploy to staging and production.

The difference at that point is that Cursor is a far more capable IDE, so any dev would prefer it over Caffeine’s far simpler web UI. Granted, for non-devs the simplicity can be a selling point, but whipping up a very simple user interface is of course easily done if that’s what we’re after, and thus again not much of a moat for Caffeine.

So yeah, I’m not that impressed by Caffeine, exactly as I expected not to be, and for the reasons I laid out here and explained in several posts before.

But as I also said before, that doesn’t matter because I continue to be extremely impressed by ICP, and if Caffeine will put creating applications on ICP in the hands on non-devs a bit sooner that if we have to wait for more general tools like Cursor to make it correspondingly easy, then building it was well invested effort.

Just don’t be dismayed when Cursor catches up, and then surpasses Caffeine, in the “normal user creating ICP apps” space - rejoice, because it will be a good thing

[HERO @HEROonchain]

So many people told me that @caffeineai is not going to burn enough $ICP because of the offchain LLM Component.

Well, yesterday around 100 people tested ☕️and we burned 17K $ICP 🔥

Make your own conclusions...
$ICP = World Computer ♾

Data: @icterminal

[Adeyinka Joshua @Adebits_1]

With just 100 testers on #CaffeineAI yesterday, over 17,000 $ICP was burned.

Now imagine what happens when it opens to everyone. 😳
You're not bullish enough. #InternetComputer #ICP

[2moon @Crypto2Moon]

@AaronTing8 Hopefully the price will never be the same again.