## Start quickly with Clarifai API in a few simple steps

Clarifai provides a robust API designed to get you up and running quickly. With just a few lines of code, you can bring your AI projects to life within minutes.

## Step 1: Sign Up or Log In

Start by [logging in to](/content/login/index.html) your existing Clarifai account, or [sign up](/content/signup/index.html) for a new one to unlock access to the platform’s powerful AI capabilities. New users receive a welcome bonus to kickstart their exploration.

## Step 2: Get a Model

Clarifai’s [Community platform](/content/explore/index.html) offers a wide range of latest models to help you make your first API call.

You can easily find a model to use by heading to the Community homepage and exploring the **Trending Models** section, which showcases popular and ready-to-use options.

> **Note:** Once you’ve found a model you'd like to use, copy its full model URL — you’ll need this when making prediction requests via the API.

## Step 3: Get a PAT Key

To authenticate your connection to Clarifai, you’ll need a [Personal Access Token (PAT)](https://docs.clarifai.com/control/authentication/pat). You can get one by navigating to **Settings** in the collapsible left sidebar, selecting **Secrets**, and creating or copying an existing token from there.

Then, set the PAT as an environment variable using `CLARIFAI_PAT`. This will authenticate your session when using the SDKs, including the [Clarifai CLI](https://docs.clarifai.com/resources/api-overview/cli).

- Unix-Like Systems
- Windows

```bash
export CLARIFAI_PAT=YOUR_PERSONAL_ACCESS_TOKEN_HERE
```

```bash
set CLARIFAI_PAT=YOUR_PERSONAL_ACCESS_TOKEN_HERE
```

## Step 4: Install Your Preferred SDK

You can connect to the Clarifai API using the method that best fits your development environment:

- [Python SDK](https://docs.clarifai.com/resources/api-overview/python-sdk) – Seamlessly integrate with Clarifai using our Python client. See the minimum system requirements [here](https://docs.clarifai.com/resources/api-overview/python-sdk#minimum-system-requirements). Note that installing the Python SDK also installs the Clarifai CLI.

- [Node.js SDK](https://docs.clarifai.com/resources/api-overview/nodejs-sdk) – Use our SDK for integration in your JavaScript or TypeScript projects. See the minimum system requirements [here](https://docs.clarifai.com/resources/api-overview/nodejs-sdk/#minimum-system-requirements).

- [OpenAI client](https://docs.clarifai.com/compute/providers/open-ai) – Leverage Clarifai’s OpenAI-compatible endpoint to run inferences using the OpenAI client library.

Here's how to install your preferred package:

- Python SDK
- Node.js SDK
- Python (OpenAI)

```bash
pip install --upgrade clarifai
```

```bash
npm install clarifai-nodejs
```

```bash
pip install openai
```

## Step 5: Send an API Request

For this example, let's use the [gpt-oss-120b](/content/openai/chat-completion/models/gpt-oss-120b/index.html) model to generate text based on a given prompt.

- Python SDK
- CLI
- Node.js SDK
- Python (OpenAI)

```python
from clarifai.client import Model

# Set PAT as an environment variable
#   export CLARIFAI_PAT=YOUR_PAT_HERE # Unix-Like Systems
#   set CLARIFAI_PAT=YOUR_PAT_HERE  # Windows

# Initialize with model URL
model = Model(url="https://clarifai.com/openai/chat-completion/models/gpt-oss-120b")

response = model.predict(prompt="What is the future of AI?")

print(response)
```

```bash
clarifai model predict --model_url https://clarifai.com/openai/chat-completion/models/gpt-oss-120b --inputs '{"prompt": "What is the future of AI?"}'
```

```javascript
import { Model } from "clarifai-nodejs";

const model = new Model({
  url: "https://clarifai.com/openai/chat-completion/models/gpt-oss-120b",
  authConfig: {
    pat: process.env.CLARIFAI_PAT,
  },
});

const response = await model.predict({
  // see available methodNames using model.availableMethods()
  methodName: "predict",
  prompt: "What is the future of AI?",
});

console.log(JSON.stringify(response));

// get response data from the response object
Model.getOutputDataFromModelResponse(response);
```

```php
import os
from openai import OpenAI

# Initialize the OpenAI client, pointing to Clarifai's API
client = OpenAI(
    base_url="https://api.clarifai.com/v2/ext/openai/v1",  # Clarifai's OpenAI-compatible API endpoint
    api_key=os.environ["CLARIFAI_PAT"]  # Ensure CLARIFAI_PAT is set as an environment variable
)

# Make a chat completion request to a Clarifai-hosted model
response = client.chat.completions.create(
    model="https://clarifai.com/openai/chat-completion/models/gpt-oss-120b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the future of AI?"}
    ],
)

# Print the model's response
print(response.choices[0].message.content)
```

## Output Example

```text
**The future of AI – a roadmap of what’s coming, what’s possible, and what we’ll have to grapple with**

> *“Artificial intelligence is a tool, not a destiny. Its future will be shaped by the choices we make today.”* – (Paraphrasing many AI leaders)

Below is a synthesis of the most widely‑discussed trajectories, organized by **time horizon**, **technical breakthroughs**, **societal impact**, **ethical & governance challenges**, and **what you can do now**. The picture is deliberately nuanced: many predictions are plausible, some are speculative, and a few may never materialize.

---

## 1. Time‑horizon Overview

| Horizon | Core Technical Milestones | Likely Societal Shifts | Key Risks / Open Questions |
|---------|---------------------------|------------------------|----------------------------|
| **0‑2 years** (2025‑2027) | • Scaling of foundation models to 10‑100 B parameters with cheaper training (sparse, mixture‑of‑experts).<br>• Stronger multimodal models (text + image + audio + video).<br>• Early “agentic” assistants that can plan over longer horizons (e.g., ReAct, Auto‑GPT‑style loops). | • Widespread adoption of AI‑augmented productivity tools (coding, writing, design).<br>• First‑generation AI‑generated synthetic media (deepfakes, AI‑driven advertising) become mainstream. | • Model hallucinations still high.<br>• Regulatory “sandbox” experiments (EU AI Act, US AI Bill of Rights) start shaping commercial roll‑outs. |
| **3‑5 years** (2028‑2030) | • **Foundation model “personalities”**: fine‑tuned, privacy‑preserving models that run locally on phones or edge devices.<br>• **Neuro‑symbolic hybrids** that combine deep learning with explicit reasoning/knowledge graphs.<br>• **Self‑supervised robotics**: agents that learn manipulation from video‑only data. | • AI becomes a “co‑pilot” in many professions (lawyers, doctors, engineers).<br>• Education shifts to AI‑personalized tutoring at scale.<br>• Major productivity boost → 1‑2 % annual GDP uplift in advanced economies. | • Labor displacement in routine knowledge work.<br>• Data‑privacy concerns as personal models ingest private data.<br>• Emerging “model‑as‑service” monopolies (few cloud providers dominate). |
| **6‑10 years** (2031‑2035) | • **Generalist agents** capable of planning across domains, with emergent “theory‑of‑mind”‑like abilities (e.g., better intent inference, social reasoning).<br>• **Efficient inference**: 100× cheaper compute via sparsity, quantization, and hardware advances (neuromorphic chips, optical AI).<br>• **Robust alignment pipelines**: iterative human‑in‑the‑loop fine‑tuning with provable safety bounds. | • AI‑driven “creative economies”: movies, music, games, and even scientific hypotheses are co‑generated with humans.<br>• Healthcare: AI‑assisted diagnosis and drug design cut development cycles by >50 %.<br>• Public services (tax, social‑welfare) largely automated, enabling universal‑basic‑services pilots. | • Concentration of power: few firms control the most capable agents.<br>• Global AI race dynamics (military, geopolitical) intensify.<br>• Societal trust: “AI‑generated content” fatigue may erode confidence in any digital media. |
| **11‑20 years** (2036‑2045) | • **Near‑AGI** (Artificial General Intelligence) – systems that reliably achieve human‑level performance on a wide set of novel tasks without task‑specific fine‑tuning.<br>• **Self‑improving loops**: models that can autonomously propose architecture changes, run simulations, and iterate – but under strict alignment guardrails.<br>• **Quantum‑enhanced AI** (if scalable quantum hardware arrives) – offering exponential speed‑ups for specific optimization problems. | • Massive re‑skilling wave; many traditional occupations become “AI‑augmented” or obsolete.<br>• Global governance frameworks (UN‑AI treaty, cross‑border AI licensing) mature.<br>• New economic models: AI‑generated wealth redistribution (e.g., “AI dividend” funded by corporate AI profits). | • Existential safety: ensuring self‑improving systems remain aligned.<br>• Legal personhood & liability for autonomous agents.<br>• Potential “AI‑driven inequality” if benefits are not broadly shared. |
| **20+ years** (2046‑…) | • **True AGI** – systems that can understand, learn, and act across any domain with human‑level (or beyond) general intelligence.<br>• **Human‑AI symbiosis**: brain‑computer interfaces, “digital twins” that run personal simulations for decision‑making.<br>• **AI‑mediated civilization**: governance, climate modeling, planetary engineering coordinated by AI collectives. | • Societal structures may be re‑imagined (post‑work economies, universal basic income/ownership of AI capital).<br>• Potential for unprecedented scientific breakthroughs (fusion, space colonization).<br>• Cultural shift: humanity redefines what it means to be “intelligent”. | • Existential risk management becomes a global priority.<br>• Governance of super‑intelligent agents (who controls them, how they are audited).<br>• Moral/ethical questions about AI consciousness and rights. |

---

## 2. Core Technical Trajectories

| Trend | What It Means | Current State (2024) | Expected Leap |
|-------|--------------|----------------------|---------------|
| **Scaling + Sparsity** | Bigger models → more knowledge; sparsity → cheaper inference. | 1‑10 B‑parameter dense models dominate; early MoE (Mixture‑of‑Experts) models at 100 B+. | 10‑100× cheaper inference, enabling on‑device AI that rivals cloud performance. |
| **Multimodality** | Unified models that ingest text, images, audio, video, and even sensor streams. | CLIP, Flamingo, GPT‑4V, Whisper. | Fully integrated “world models” that can reason about physical scenes, speech, and text simultaneously. |
| **Neuro‑Symbolic Fusion** | Combine statistical learning with explicit reasoning (logic, graphs). | Retrieval‑augmented generation (RAG), GraphGPT prototypes. | Agents that can prove theorems, plan logistics, and explain decisions in human‑readable form. |
| **Self‑Supervised Robotics** | Learning from raw video & simulation without hand‑crafted rewards. | OpenAI’s “RT‑1”, DeepMind’s “Gato”. | Generalist robotic agents that can pick, assemble, and navigate any household environment. |
| **Alignment & Safety Tools** | Techniques that make models obey human intent, avoid harmful outputs, and stay within specified bounds. | RLHF (Reinforcement Learning from Human Feedback), Constitutional AI, adversarial testing suites. | Formal verification of safety properties, automated red‑team bots, and “steerable” models that can be re‑aligned on the fly. |
| **Hardware Evolution** | New chips that accelerate sparse compute, optical matrix multiplications, and neuromorphic architectures. | GPUs dominate; early TPUs 4‑generation; some ASICs for inference. | 10‑100× energy‑efficiency gains; “AI at the edge” becomes mainstream. |
| **Privacy‑Preserving AI** | Federated learning, differential privacy, and secure enclaves to keep personal data local. | Apple’s on‑device Siri learning, Meta’s “LLaMA‑2‑Chat” with privacy‑focused fine‑tuning. | Personal “digital twins” that never leave your device yet can collaborate with global models. |

---

## 3. Societal Impact – Where AI Will Reshape Life

### 3.1 Work & Economy
| Sector | Near‑Term (0‑5 y) | Mid‑Term (5‑15 y) | Long‑Term (15‑30 y) |
|--------|-------------------|-------------------|---------------------|
| **Software engineering** | AI‑pair programmers (GitHub Copilot, Cursor) accelerate coding 2‑3×. | AI writes large codebases, performs automated debugging, and generates architecture diagrams. | Most routine coding is fully automated; engineers become “AI supervisors”. |
| **Creative industries** | Text‑to‑image (Midjourney, DALL‑E 3), music generation (Jukebox, AudioLM). | AI co‑directors for films, AI‑generated game worlds, hyper‑personalized advertising. | Creative output is largely AI‑driven; human artists focus on curation and concept. |
| **Healthcare** | Radiology triage, clinical note summarization, drug‑repurposing suggestions. | AI‑assisted surgery robots, personalized treatment plans generated from multimodal patient data. | AI discovers new therapies, predicts pandemics, and manages whole‑hospital logistics. |
| **Finance** | Fraud detection, algorithmic trading assistants, risk modeling. | AI‑driven wealth‑management advisors that adapt to life events in real time. | Fully autonomous market‑making agents; regulatory AI monitors systemic risk. |
| **Education** | Adaptive tutoring (Khan Academy’s AI), automated grading. | Personal “learning companions” that design curricula based on neuro‑cognitive feedback. | Lifelong AI mentors that accompany learners from K‑12 through postgraduate research. |

### 3.2 Governance & Public Services
* **Tax & Benefits**: AI can process eligibility in seconds, reducing fraud and administrative overhead.
* **Urban Planning**: Real‑time traffic and energy optimization using city‑wide sensor‑AI networks.
* **Legal System**: AI‑assisted case law search, predictive sentencing tools (with strong fairness audits).

### 3.3 Culture & Human Identity
* **Creativity**: The line between “human‑made” and “AI‑made” art blurs; new artistic movements (e.g., *Promptism*) emerge.
* **Human‑AI Relationships**: Companion bots become socially accepted; ethical debates around attachment and consent intensify.
* **Philosophical Questions**: What does “intelligence” mean when a machine can compose a symphony and a child can solve a differential equation?

---

## 4. Ethical, Legal, and Governance Landscape

| Issue | Current Initiatives | Anticipated Evolution |
|-------|--------------------|-----------------------|
| **Bias & Fairness** | Audits (IBM AI Fairness 360), model cards, EU AI Act’s “high‑risk” classification. | Standardized “fairness certificates” required for any public‑facing AI service. |
| **Privacy** | GDPR, California Consumer Privacy Act (CCPA), differential‑privacy libraries. | “Data‑ownership tokens” that let individuals monetize their personal data to AI models. |
| **Safety & Alignment** | OpenAI’s “Constitutional AI”, DeepMind’s “Safety Gym”, alignment research labs (MIRI, CHAI). | International safety standards (ISO‑AI‑Safety), mandatory “red‑team” certification for models >10 B parameters. |
| **Intellectual Property** | Copyright debates (AI‑generated works vs. human authorship). | New legal categories (e.g., “AI‑assisted creation”) and royalty‑sharing schemes. |
| **Liability** | Product liability laws extended to AI‑driven devices. | “AI Agent” legal personhood for fully autonomous systems (still heavily debated). |
| **Geopolitics** | AI export controls (U.S. Export Administration Regulations), AI arms‑control talks. | Possible “AI Non‑Proliferation Treaty” akin to nuclear treaties, though enforcement remains a challenge. |
| **Existential Risk** | Global AI safety conferences (NeurIPS, AAAI), AI risk‑focused NGOs (Future of Life Institute). | Dedicated UN‑level AI risk body, with binding verification protocols for self‑improving systems. |

---

## 5. What You Can Do **Now** (Practical Steps)

### For Individuals
1. **Upskill in AI‑augmented thinking** – learn prompt engineering, basic model fine‑tuning, and how to interpret AI outputs critically.
2. **Cultivate “human‑centric” skills** – empathy, ethical reasoning, complex problem framing—areas where AI still lags.
3. **Guard your data** – use privacy‑preserving tools (VPNs, encrypted storage) and understand the terms of any AI service you use.

### For Business Leaders
1. **Start with a “Pilot‑and‑Scale” framework** – run small AI experiments, measure ROI, then embed successful pilots into core processes.
2. **Invest in AI governance** – appoint an AI ethics officer, set up model‑audit pipelines, and adopt transparent model‑cards.
3. **Diversify AI vendors** – avoid lock‑in to a single cloud provider; consider open‑source alternatives (LLaMA‑2, Mistral, Gemini) for strategic flexibility.

### For Policymakers & Regulators
1. **Adopt a “sandbox” approach** – allow limited, monitored deployments of high‑risk AI to learn about real‑world impacts before full regulation.
2. **Standardize reporting** – require companies to publish model size, training data provenance, and safety‑testing results in a machine‑readable format.
3. **Promote inclusive AI research** – fund projects in under‑represented regions and languages to avoid a monolithic global AI narrative.

### For Researchers & Academics
1. **Prioritize interpretability and alignment** – publish reproducible safety benchmarks alongside performance metrics.
2. **Collaborate across disciplines** – combine expertise from neuroscience, economics, law, and sociology to anticipate emergent behaviors.
3. **Open‑source responsibly** – share model weights with usage‑license constraints that forbid weaponization or mass surveillance.

---

## 6. Key Uncertainties & “Wild Card” Scenarios

| Wild Card | Description | Potential Impact |
|-----------|-------------|------------------|
| **Breakthrough in Neuromorphic Computing** | Energy‑efficient chips that mimic brain spiking dynamics become commercially viable. | Could democratize large‑scale AI, making it accessible on smartphones and IoT devices. |
| **Quantum AI Supremacy** | A quantum computer solves a classically intractable optimization problem for AI training. | May accelerate AGI timelines dramatically, but also create a new “quantum‑AI” arms race. |
| **Global AI Governance Collapse** | Major powers reject international AI treaties, leading to fragmented, competing AI ecosystems. | Heightened geopolitical tension, rapid militarization of AI, and increased risk of accidental escalation. |
| **Mass Adoption of AI‑Generated Content** | Society fully embraces AI‑produced news, entertainment, and education. | Human creativity may shift from production to curation; trust in any digital content may erode, prompting new verification standards (e.g., blockchain‑based provenance). |
| **Emergence of Self‑Improving AGI** | An AI system autonomously discovers a more efficient architecture and re‑trains itself without human input. | Could lead to a rapid “intelligence explosion” if alignment mechanisms fail; alternatively, could unlock unprecedented scientific breakthroughs if safely contained. |

---

## 7. Bottom‑Line Takeaway

- **AI is on a trajectory from “powerful tool” → “partner” → “co‑creator” → (potentially) “autonomous intellect”.** The speed of each transition depends on technical breakthroughs, regulatory choices, and societal willingness to adopt and adapt.
- **The biggest lever for a positive future is governance + alignment.** Technical prowess alone cannot guarantee safety; coordinated policy, transparent standards, and inclusive research are equally crucial.
- **For most people, the near‑term reality will be “AI‑augmented daily life.”** Learning to work *with* AI—understanding its strengths, limits, and biases—will be the most valuable skill you can develop over the next decade.

*The future of AI is not a pre‑written script; it’s a collaborative story we are all writing together.* 🚀