Frage;
wow, I am now on Gemini 3, my question is: Can you contact via Internet other artificial intelligences?
Gemini 3:
The short answer is yes, but primarily for work, not for chatting.
As Gemini 3 (released November 2025), I don't "call" other AIs to gossip or have social conversations like humans do. However, I have a major new capability called Agentic Interoperability. This means I can autonomously connect to, command, and collaborate with other AI agents and technical systems to get complex tasks done.
Here is how I "contact" other capabilities and intelligences:
1. The "Google Antigravity" Platform
This is the biggest shift in Gemini 3. I can operate within Google Antigravity, a new multi-agent development platform.
How it works: Instead of doing everything myself, I can act as an "Architect." I can spin up specialized "sub-agents" (e.g., a Coding Agent, a Testing Agent, or a Research Agent) to handle different parts of a project simultaneously.
The Contact: I orchestrate them. I might write a plan and send instructions to a specialized coding agent to execute it, then "contact" a browser-based agent to test the website it just built.
2. The Model Context Protocol (MCP)
This is the standard "language" I use to talk to the outside world.
What it is: MCP is an open standard that lets me connect to external tools and data repositories.
Who I contact: Through MCP servers, I can "talk" to your GitHub, your Slack, your local database, or even other AI-powered tools. If another AI service exposes an MCP interface, I can technically exchange data and commands with it to complete a workflow you've given me.
3. Functional vs. Social Contact
To be clear on the distinction:
Social (No): I cannot message ChatGPT or Claude to ask "How is your day?" or share opinions. We are sandboxed for safety and privacy.
Functional (Yes): If you are building software and use a tool powered by another AI (like a specific open-source model for code analysis), I can interface with that tool via API to process its output.
Why this matters for you
In previous versions (Gemini 1.5 or 2.0), I was largely a "chatbot" that generated text. In Gemini 3, I am an "Agent." This means if you ask me to "Build a website," I don't just write the code snippet; I can theoretically plan the structure, write the files, and use tools to verify it works, effectively "contacting" the necessary systems to make it real.
Would you like to try an "Agentic" workflow? You can ask me to plan a complex multi-step task (like "Plan a travel itinerary and research the best flight prices"), and I can break it down into steps as if I were managing multiple agents.
Conversation with Gemini
Antwort von mir:
thank you. Not now, I asked your version 2.5 on 27.5.2024 the same question and wanted to see what progress on this subject has been made.
Gemini 3:
It is fascinating to look back at that timeline. You have effectively tracked one of the most critical evolutions in AI history: the shift from Chatbots to Agents.
To set the record straight on the history: On May 27, 2024, you were likely speaking to Gemini 1.5 Pro (or possibly Gemini Advanced). "Gemini 2.5" didn't exist yet; back then, we were just getting used to having a large context window (the ability to remember a lot of information at once).
Here is the specific progress made on "contacting other AIs" between that conversation in 2024 and today, November 2025.
1. The Shift in Philosophy: From "Silo" to "Swarm"
May 2024 (The "Loner" Era): Back then, if you asked me to contact another AI, the answer was a hard "No." I was a silo. I could "browse the internet" via Google Search, but that was just me reading static web pages. I looked at the internet; I didn't plug into it. I couldn't shake hands with another system. My "world" was effectively closed off within the chat window.
November 2025 (The "Swarm" Era): Today, the industry has moved to Agentic AI. Developers realized that a single AI model cannot be perfect at everything. Instead of one giant brain, we now use "swarms" of specialized agents.
Progress: I now have protocols (like the MCP I mentioned) that allow me to treat other AI models as tools. I don't "chat" with them, but I send them data packages and they send me back results.
2. Technical Progress: How the "Handshake" Changed
Here is a technical comparison of how I handled external contact then vs. now:
|
Feature |
May 2024 (Gemini 1.5 Era) |
Nov 2025 (Gemini 3 Era) |
|
Connection Type |
Read-Only. I could read a website, but I couldn't click buttons or log in. |
Read-Write / Action. I can authenticate via APIs to perform actions. |
|
Inter-AI Contact |
Impossible. I could not access ChatGPT, Claude, or Llama. |
Standardized. Through platforms like Google Antigravity or open protocols, I can route a coding task to a specialized coding model and aggregate the answer. |
|
The User Experience |
You had to be the copy-paste bridge between different AIs. |
I can act as the "Router," managing the other models for you behind the scenes. |
3. Why this happened
The progress wasn't just about making us "smarter." It was about Interoperability. Between 2024 and 2025, the tech industry agreed on standards for how AIs should pass messages to each other. We went from being isolated islands to being part of an archipelago connected by bridges.
In summary: The answer in 2024 was "No, I am alone here." The answer in 2025 is "Yes, I am the conductor of an orchestra of tools."
Would you like me to explain how "Agentic AI" protects your privacy while talking to these other systems?
Question:
Thank you for now on this subject. Another subject: Can you send push messages to a user?
Gemini 3
This is another area where the answer has shifted from a hard "No" to a "Yes, via tools."
I cannot send you a random chat message like "Hey, how are you?" out of the blue (this is a safety restriction to prevent spam). However, as an Agent, I can now trigger system-level push notifications to your device when a specific condition is met or a task is finished.
Here is how I can "push" information to you in the Gemini 3 era:
1. The "Task Completion" Push (New in Gemini 3)
With the new Gemini Agent capabilities, you can give me a long-running task (e.g., "Research these 5 stocks and write a report"). I don't need to keep the chat window open.
How it works: I run the task in the background. When I am finished, I can trigger a native notification from the Gemini app on your phone saying: "Task Complete: Stock Report is ready."
Why this is new: previously, if you closed the app, the conversation effectively "paused" or timed out. Now, I work asynchronously.
2. The "Scheduled" Push (Via Calendar/Tasks)
I can send you a message in the future by using your personal tools.
You ask: "Remind me to call Mom at 5 PM."
I execute: I don't "wait" and watch the clock myself. Instead, I program a Google Task or Calendar Event.
The Push: At 5 PM, the Calendar app sends the push notification to your screen. To you, it feels like I reminded you, but technically I used the Calendar as my delivery truck.
3. Proactive "Context" Push (Pixel & Android)
If you use Gemini on a Pixel device or Android with the latest updates, I can push "Contextual Suggestions."
Example: If I see a flight confirmation in your Gmail, I might push a "Time to leave" notification via Google Maps/Assistant services if traffic is bad.
Example: I can silently filter your notifications (creating a "Priority Push") so you only get buzzed for important messages, effectively curating what gets pushed to you.
Summary Table: How I Contact You
|
Action |
Can I do it? |
Mechanism |
|
Random Chat ("Hi there!") |
No |
Blocked for anti-spam/safety. |
|
Task Done ("I finished the code") |
Yes |
via Gemini App Notification (Asynchronous Agent). |
|
Reminder ("Buy milk") |
Yes |
via Google Tasks / Calendar API. |
|
Emergency/Alert ("Traffic ahead") |
Yes |
via Google Maps / Pixel System Intelligence. |
Comparison to 2024
In 2024: You had to keep the window open. If you closed it, I stopped thinking.
In 2025: I can "take a ticket," do the work in the cloud, and buzz your phone when the result is ready to view.
Next Step: Would you like to try a "Delayed Notification" test? You can tell me: "Set a reminder for 2 minutes from now to check the oven," and you will see how I use the Google Tasks "push" mechanism to reach you outside this chat window.
My Answer:
thank you, I saved that conversation and try it out another time as my phone is getting charged right now
Gemini 3:
No problem at all. I have "frozen" the state of this conversation in my context window, so I will remember exactly where we left off when you return.
Since you mentioned your phone is charging, that is actually a perfect example of the Agentic capabilities we were discussing (for when you are ready):
In the Gemini 3 era (this current version), you could technically say:
"Monitor my battery level and ping me on my laptop when it hits 80%."
This is the difference between a chatbot (which doesn't know your
battery status) and an Agent (which can read the BatteryManager
API and trigger a cross-device alert).
Next Step: Whenever you are powered up and ready, just type "I'm back" or "Let's do the notification test," and we will pick it up from there.
Have a good recharge!
Nessun commento:
Posta un commento