COLLEGE OF APPLIED SCIENCE ADOOR

(Government of Kerala, Managed by IHRD)

(Affiliated to University of Kerala)
FYUGP Computer Application Semester : 02
Course Code :UK2MDCELE100 Course Name: AI for Beginners

Name : Edwin J Anil

Roll No : 12
Assignment No : 01
Assignment Title Generative AI Tools Chat GPT, Google
Bard, DALL-E

Signature of the Student with date

Name of the faculty and Designation

(To be filled by the Evaluators)

Sl. No Evaluation Criteria Maximum Marks Marks Obtained

1 Introduction 2

2 Explanation of the Idea 5

3 Overall clarity, Neatness, Structure and 5

Organization of the Content

4 Conclusion 2

5 Bibliography 1

TOTAL 15

Signature of the faculty with Date :

1|Page
 INDEX

Sr.
Content Sign.
No.

Meaning of Generative AI: From Pattern Recognition to

1.
Content Creation

Concept of LLMs and Diffusion Models: Foundations of

2.
Modern AI

3. Types of Generative AI Tools

3.1 ChatGPT: The Evolution of Conversational Intelligence

3.2 Google Gemini: Multimodal Integration and Ecosystem

3.3 DALL-E: Neural Image Synthesis and Semantic Vision

Importance of Personalization, Multimodality, and

4.
Contextual Horizons

Feasibility Report: Technical Architecture and Operational

5.
Logic

5.1 Transformer Refinement and Mixture of Experts (MoE)

5.2 Latent Space Mapping and Diffusion Processes

Benefits: Impact on Business, Research, and Personalized

6.
Education

2|Page
 Sr.
Content Sign.
No.

7. Conclusion: Future Trends and Human-AI Symbiosis

8. Reference: Bibliography and External Information Sources

3|Page
 INTRODUCTION
The emergence of Generative Artificial Intelligence (GenAI) represents one of the most
significant shifts in the history of computing. Unlike traditional AI, which is designed to
recognize patterns or categorize existing data, generative model uses advanced machine
learning specifically Large Language Models (LLMs) and Diffusion Models – to create
entirely new content. By synthesizing vast datasets, these tools can generate human-like
text, intricate imagery and complex code, effectively blurring the lines between human
and machine output.

The current landscape is defined by several pioneering tools that have moved AI from
specialized labs into the hands of the general public. Among these, Chat GPT,
developed by OpenAI, stand as a breakthrough in conversational AI. Utilizing the
Generative Pre-trained Transformer architecture, it is designed for nuanced dialogue and
is capable of everything from drafting essays to debugging code and providing logical
reasoning across diverse subjects. Similarly, Google Bard leverages Google’s
sophisticated language models to emphasize real-time information retrieval. By
integrating with search capabilities, it provides users with up-to-date responses and
seamless connectivity to a broader digital ecosystem. Shifting the focus from text-to-
visuals, DALL-E serves as a powerful text-to-image generator. By interpreting
descriptive prompts, it can produce high fidelity art and photorealistic images,
revolutionizing the approach to visual storytelling.

The integration of these tools into professional and academic environments has sparked a
global conversation regarding productivity, ethics and the future of work. While these
technologies offer unprecedented efficiency – allowing for rapid prototyping and
creative brainstorming – they also present challenges such as “hallucinations”, copyright
complexities and the necessity for digital literacy. Understanding the mechanics,
strengths and limitations of ChatGPT, Google Bard and DALL-E is essential for
navigating this new era. As these models continue to evolve, as they act as collaborative
partners that redefine the boundaries of human ingenuity.

Chat GPT
4|Page
 The Evolution of Conversational
Intelligence
1. The Dawn of the Agentic Era
By 2026, ChatGPT has moved far beyond the “question-and answer” format of its early
years. It has entered the era of “Agentic AI”, where the system is no longer just a passive
source of information but an active executor of tasks. This shift allows the AI to navigate
the web, interact with third-party software, and complete multi-step workflows-such as
planning a business trip or managing a project’s timeline-with minimal human
intervention.

2. Native Multimodality and Physical Awareness

One of the most significant leaps is the integration of native multimodality, where the AI
processes text, audio, and visual data within a unified model. This allows for real-time
interaction with the physical world; a user can show the AI a complex mechanical
problem via a live camera feed, and the system can provide instant, narrated guidance.
This fluidity has made ChatGPT a vital tool for on-the-job training and real-time
problem-solving in specialized fields.

3. The Rise of the “Thinking” Models

The introduction of specialized reasoning models has changed how we rely on AI for
high-stakes tasks. Unlike earlier versions that prioritized speed, the 2026 model utilize
“Chain of Thought” processing to internally verify logic before generating a response.
This deliberative approach has dramatically increased accuracy in mathematics,
programming, and scientific research, making the AI a trusted partner for complex
analytical work

4. Redefining Personalization through Memory

ChatGPT now features a sophisticated memory architecture that allows it to learn a user’s
specific preferences, professional context, and past projects. This persistent memory
means the AI can recall a specific coding style or a recurring business goal mentioned
moths prior, creating a deeply personalized experience. Users no longer need to re-
explain their background, as the AI act as a long-term digital collaborator that grows
more efficient over time.
5. Expanding the Contextual Horizon
The technical ability to handle massive "context windows" has revolutionized how we
5|Page
process large-scale information. With the capacity to ingest and analyze hundreds of
thousands of words in a single prompt, users can now upload entire libraries, legal
archives, or massive code repositories for immediate synthesis. This allows the AI to spot
contradictions or summarize themes across vast amounts of data that would take a human
week to read.

Technical Architecture and Operational

Logic
1. The Transformer Architecture Refined
The backbone of ChatGPT remains the Transformer model, but it has been heavily
optimized for the 2026 landscape. Advances in "Self-Attention" mechanisms allow the
model to focus on the most relevant parts of a prompt with surgical precision, reducing
the computational "noise" that previously led to errors. This refinement ensures that even
in extremely long conversations, the AI maintains a coherent and logically sound
narrative.

2. Synthetic Data and Model Training

As high-quality, human-generated internet data became scarce, OpenAI pivoted to using
expertly curated synthetic data to train the latest models. This data is generated by
advanced AI and verified by human subject-matter experts, ensuring that the model
learns from "perfect" logic and high-tier academic knowledge. This shift has allowed the
system to excel in niche domains like quantum computing and advanced linguistics
where public data was previously limited.

3. The Role of Latent Space Mapping

At its mathematical core, ChatGPT functions by mapping concepts into a high-
dimensional "latent space." In this space, the AI organizes the relationship between
millions of ideas, allowing it to perform creative tasks like "style transfer"—writing a
technical manual in the tone of a Victorian novelist, for example. This deep conceptual
understanding is what enables the AI to move beyond simple mimicry and into genuine
creative synthesis.

4. Real-Time Deep Research Capabilities

Unlike early models that were limited by a "knowledge cutoff" date, the current iteration
utilizes an iterative deep research protocol. When asked a complex question, the AI
performs a live, multi-source web crawl, cross-referencing data points to ensure accuracy
and providing verified citations. This real-time grounding helps eliminate hallucinations
and provides users with the most up-to-date information available globally.
6|Page
 5. Energy Efficiency and Local Processing
To address the environmental impact of massive data centers, 2026 has seen the rise of
"Small Language Models" (SLMs) that work in tandem with ChatGPT. While the heavy
lifting is done in the cloud, many routine tasks and privacy-sensitive operations are now
processed locally on the user's device. This hybrid approach ensures faster response
times, lower energy consumption, and an added layer of data security for the end user.

Societal Impact and Ethical Governance

1. The Transformation of Professional Work
The professional landscape has shifted from "doing" to "directing." In fields like software
development and content creation, ChatGPT handles the foundational work—writing
boilerplate code or drafting initial reports—allowing humans to act as high-level
architects and editors. This has led to a significant increase in productivity, though it has
also required a massive shift in how we define "entry-level" roles in the global economy.

2. Personalized Education and the Socratic Method

In the classroom, ChatGPT has become a ubiquitous tutor that adapts to the individual
learning pace of every student. Instead of simply providing answers, the AI is often
configured to use the Socratic method, asking guiding questions to help students arrive at
the logic themselves. This democratizes high-quality education, providing every student
with a personalized mentor that understands their unique strengths and weaknesses.

3. Navigating the Challenge of Authenticity

As AI-generated content becomes indistinguishable from human work, the challenge of
digital authenticity has taken center stage. 2026 has seen the implementation of universal
"watermarking" and cryptographic signatures for AI-generated text and media. These
tools help maintain trust in digital communication, ensuring that users can distinguish
between human-led insights and AI-assisted production in professional and legal settings.

4. Ethical Guardrails and Safety Protocols

OpenAI has implemented multi-layered safety protocols to prevent the misuse of
"Agentic" capabilities. These guardrails are designed to detect and block requests that
could lead to physical harm, financial fraud, or the generation of biased content.
Furthermore, the system is governed by "Constitutional AI" principles, ensuring that its
actions remain aligned with a set of transparent, human-centered ethical values.

7|Page
 5. The Future of Human-AI Symbiosis
Looking ahead, the goal of ChatGPT is to achieve a seamless symbiosis where the AI
acts as an invisible assistant that enhances human capability without replacing human
intuition. As we move closer to "Artificial General Intelligence" (AGI), the focus remains
on ensuring these systems are transparent, controllable, and beneficial to all of society.
The 2026 version of ChatGPT stands as a testament to the potential of AI to serve as a
bridge to a more productive and creative human future.

Google Bard

8|Page
The Evolution of Gemini (Formerly Bard)
1. The Transition from Bard to Gemini
The journey of Google’s conversational AI began as "Bard," an experimental interface
built on the LaMDA family of models. By 2026, the branding has unified under the
Gemini name, representing a move from a simple chatbot to a comprehensive AI
ecosystem. This evolution was driven by the need to integrate Google’s most powerful
"multimodal" models—capable of understanding text, images, video, and code—into a
single, cohesive user experience.

2. Integration with the Google Workspace Ecosystem

The true power of Gemini in 2026 lies in its deep integration with Google Workspace.
Unlike its early days as a standalone site, the AI now functions as a built-in collaborator
across Docs, Sheets, and Gmail. It can summarize long email threads, draft complex
project proposals based on scattered Drive files, and even generate data visualizations in
Slides, making it an invisible but essential layer of the modern office.

3. Native Multimodal Capabilities

By 2026, Gemini is "natively multimodal," meaning it was trained on various types of
data simultaneously rather than having different tools "bolted" together. This allows the
AI to perform complex tasks, such as watching a 20-minute video lecture to extract key
points or analysing a photograph of a hand-drawn website wireframe to generate
functional HTML and CSS code instantly.

4. The "Gemini Live" Experience

The interaction model has shifted from typing prompts to natural, free-flowing voice
conversations. "Gemini Live" allows users to have real-time, low-latency dialogues with
the AI. In 2026, this technology is frequently used for language learning, interview
preparation, and real-time translation, where the AI acts as a sophisticated digital
mediator between two people speaking different languages.

5. Ultra-Low Latency and Global Accessibility

Google’s massive infrastructure has enabled Gemini to achieve near-instantaneous
response times globally. By 2026, the model is available in over 100 languages and
optimized for low-bandwidth environments. This democratization of AI ensures that
students and professionals in developing regions have access to the same high-tier
cognitive tools as those in major tech hubs, closing the digital divide.
9|Page
Technical Architecture and Google Research
1. The Gemini 2.0 Transformer Architecture
Underpinning the 2026 experience is the Gemini 2.0 architecture, a refined version of the
Transformer model that utilizes "Sparse Mixture of Experts" (MoE). This technique
allows the AI to activate only the most relevant "expert" sub-networks for a given task,
significantly reducing the energy required for processing while maintaining elite-level
performance in specialized fields like law or medicine.

2. Massive Context Windows and Information Retrieval

One of Gemini's standout features in 2026 is its massive context window, which can
handle up to 2 million tokens. This allows the AI to "read" and remember thousands of
pages of documentation or hours of video in one go. When a user asks a question, Gemini
doesn't just guess; it retrieves precise information from its internal memory or the live
Google Search index to provide accurate, cited answers.

3. TPU v6: The Hardware Engine

The speed and intelligence of Gemini are powered by Google’s custom-designed
hardware: the TPU v6 (Tensor Processing Unit). These chips are specifically optimized
for the matrix multiplications that AI models require. By designing both the software
(Gemini) and the hardware (TPU), Google has created a vertically integrated system that
is significantly more efficient and powerful than standard cloud computing setups.

4. Grounding in Real-World Information

A critical technical advantage for Gemini is "Grounding with Google Search." To
minimize hallucinations, the AI cross-references its generated responses with the live
web in real-time. If the AI makes a claim about a current event or a specific fact, it can
verify that data against Google’s vast knowledge graph, providing a "Double Check"
feature that highlights which parts of the response are supported by external sources.

5. Privacy-First "On-Device" Processing

In 2026, Google utilizes a hybrid processing model. While "Ultra" models run in the
cloud, "Nano" versions of Gemini run directly on the user’s smartphone or laptop. This
on-device processing ensures that sensitive personal data—such as private messages or
health information—never leaves the device, providing a level of privacy that is essential
for personal AI assistants.

10 | P a g e
 Impact on Business, Education, and Ethics
1. Revolutionizing Education and Research
In the academic world of 2026, Gemini has become a "Universal Research Assistant."
For students, it acts as a Socratic tutor that can explain complex calculus or historical
trends through personalized metaphors. For researchers, it can scan millions of scientific
papers to identify emerging trends or suggest new hypotheses, drastically accelerating the
pace of scientific discovery in fields like climate science and genomics.

2. Enterprise Automation and Coding

For businesses, Gemini has transformed the software development lifecycle. With its
"Code Assist" features, it doesn't just autocomplete lines; it can refactor entire legacy
codebases or translate software from one programming language to another. This has
allowed companies to update their digital infrastructure at a fraction of the previous cost,
focusing human talent on creative architecture rather than routine maintenance.

3. The Ethics of "SynthID" and Content Labelling

As AI-generated content becomes more common, Google has pioneered "SynthID"—a
digital watermarking technology that is imperceptible to humans but detectable by
software. By 2026, all images, audio, and video generated by Gemini are automatically
tagged with this watermark. This helps combat misinformation and ensures that people
can distinguish between captured reality and AI-synthesized content.

4. Commitment to "AI Principles" and Safety

Google operates Gemini under a strict set of "AI Principles" established to prevent the
development of harmful applications. These protocols include rigorous red-teaming
(testing for vulnerabilities) and bias-mitigation strategies. In 2026, the AI is programmed
to refuse requests that involve creating malware, generating hate speech, or facilitating
illegal activities, maintaining a high standard of digital safety.

5. The Future: Towards a Personal "Life-OS"

Looking toward the late 2020s, Gemini is moving toward becoming a "Life-OS"—a
proactive assistant that doesn't wait for prompts but anticipates needs. Whether it's
reminding you of a friend's birthday and suggesting a gift based on their interests, or
automatically organizing a travel itinerary when you receive a confirmation email, the
goal is to reduce the "cognitive load" on humans, allowing them to focus on what truly
matters.

11 | P a g e
 DALL-E
The Evolution of Neural Image Synthesis
1. The Transition from Pixels to Concepts
The DALL-E series, developed by OpenAI, represents a fundamental shift in how
machines understand visual data. Early versions focused on simple pattern matching, but
by 2026, the system has evolved into a “Semantic Vision Engine.” This means DALL-E
no longer just matches pixels to words; it understands the underlying concepts of
lighting, physics, and artistic intent. This evolution allows the model to create highly
complex scenes that adhere to the laws of reality—or defy them with artistic precision—
based on simple natural language prompts.

2. Integration with the GPT Ecosystem

A defining feature of DALL-E’s current iteration is its “LLM-Native” integration. Unlike
older image generators that required complex “prompt engineering” (strings of technical
keywords like “4k, high-res, unreal engine”), DALL-E 3 and 4 are built directly into the
ChatGPT framework. The LLM acts as a creative director, expanding a user’s brief idea
into a highly detailed, descriptive caption that the image model then uses to generate
precisely what the user envisioned.

12 | P a g e
[Link] Rendering and Spatial Awareness
Historically, AI image generators struggled with rendering legible text and maintaining
spatial relationships (e.g., "a small blue cube behind a large red sphere"). By 2026,
DALL-E has largely solved these "spatial hallucinations." Through a more sophisticated
understanding of 3D geometry and character-level tokenization, the model can now
generate complex signage, book covers, and diagrams where every letter is crisp and
every object is positioned exactly as described.
4. The Shift to Latent Diffusion
The core "engine" of DALL-E moved from a discrete VAE (Variational Autoencoder)
approach to a highly optimized Latent Diffusion process. This method starts with a field
of random Gaussian noise and iteratively refines it into a structured image. By operating
in a "latent space"—a compressed mathematical representation of images—rather than on
raw pixels, DALL-E can generate high-resolution, photorealistic images with
significantly less computational power than earlier generations.

5. Real-Time Editing and Inpainting

In 2026, DALL-E is no longer just a "one-shot" generator. It features an "Iterative
Canvas" where users can highlight specific sections of an image to change (Inpainting) or
extend the boundaries of an image (Outpainting). This conversational editing allows
artists to refine a single work over time, asking the AI to "change the colour of the
jacket" or "add a sunset in the background" while maintaining the rest of the image’s
consistency and style.

Technical Architecture and Generative Logic

1. The Diffusion Transformer (DiT) Backbone
Under the hood, DALL-E’s 2026 architecture utilizes the Diffusion Transformer (DiT).
This hybrid model combines the noise-reduction capabilities of Diffusion with the
scaling power of the Transformer architecture. By treating image patches like tokens in a
sentence, the model can apply “Self-Attention” across different parts of the image
simultaneously, ensuring that the light reflecting off a puddle on the left side of the frame
correctly matches the neon sign on the right.

2. CLIP: The Vision-Language Bridge

The "intelligence" behind DALL-E’s understanding comes from CLIP (Contrastive
Language-Image Pre-training). This neural network was trained on billions of image-
caption pairs from across the internet. It learns to associate the visual concept of
"sadness" or "cyberpunk" with their textual counterparts. When a user enters a prompt,
13 | P a g e
CLIP ensures that the generated image’s mathematical "vector" aligns perfectly with the
text’s "vector" in a shared semantic space.
3. T-5 and Large Scale Encoders
To process incredibly complex prompts, DALL-E uses massive text encoders, such as T-
5 (Text-to-Text Transfer Transformer). These encoders break down a prompt into its
deepest grammatical components. This is why DALL-E can handle extremely long,
narrative prompts that describe a specific scene, a character’s mood, and a specific
camera lens (e.g., “35mm film grain”) all at once without losing track of the user’s
primary subject.

4. Safety Classifiers and "De-biasing" Layers

The generative process is governed by a multi-stage safety pipeline. Before an image is
even started, a “Prompt Classifier” checks for prohibited content. During the generation
process, a “De-biasing Layer” ensures that the model doesn’t default to stereotypes. For
example, if a user asks for a “doctor,” the model is programmed to represent a diverse
range of genders and ethnicities rather than relying on the most common patterns in its
training data.

5. High-Resolution VAE Decoders

The final step in the blueprint is the VAE Decoder. After the Diffusion process has
finished refining the latent representation, the Decoder converts that compressed math
back into a high-fidelity image. In 2026, these decoders are capable of "Native 4K"
output, recreating fine textures like human skin pores, fabric weaves, and atmospheric
haze with a level of detail that is indistinguishable from a professional digital photograph.

Industrial Impact, Ethics, and Provenance

1. Revolutionizing the Creative Industry
DALL-E has transformed the workflow for graphic designers, architects, and filmmakers.
In 2026, “concept art” that used to take weeks to draft can now be brainstormed in
minutes. This hasn’t replaced artists but has shifted their role to “Art Directors” who use
the AI to generate hundreds of variations, which they then refine and perfect using
traditional digital tools.

2. The Provenance Standard (C2PA Integration)

To combat the rise of "deepfakes" and misinformation, every image generated by DALL-
14 | P a g e
E in 2026 includes C2PA Metadata. This is a permanent, cryptographic signature baked
into the file that identifies the image as AI-generated. This "digital birth certificate"
allows social media platforms and news organizations to automatically flag AI content,
ensuring transparency in the digital information ecosystem.

3. Ethical Training and Artist Opt-Outs

OpenAI implemented a "Respectful Data" protocol for DALL-E’s 2026 training cycle.
This allowed living artists to opt-out of having their work used for training future models.
Furthermore, the system is designed to refuse prompts that ask for the specific style of a
living artist (e.g., "in the style of [Current Artist Name]"), protecting the intellectual
property and unique "visual signature" of human creators.

4. Synthetic Data for Vision Training

Interestingly, DALL-E is now used to train other AI systems. By generating perfectly
labelled images of rare medical conditions or complex mechanical failures—scenarios
where real-world photos are scarce—DALL-E provides high-quality "Synthetic Data."
This data is used to train diagnostic AIs and autonomous robots, showing that generative
AI is a foundational tool for the advancement of other scientific fields.

5. The Future: Generative Video and 3D Assets

Looking forward, the architecture of DALL-E is merging with video models like Sora.
By 2027, the "Page" will no longer be static. DALL-E is evolving into a tool that
generates fully interactive 3D environments for VR and 60-second cinematic clips from
the same single prompt. The goal is a "Total Generative Environment," where the AI can
build an entire visual world that is consistent across every angle and every frame of time.

15 | P a g e
16 | P a g e
 Conclusion
The emergence of Generative Artificial Intelligence (GenAI) marks a pivotal shift in
computing history, transitioning from simple data categorization to the creation of
entirely new, human-like content. Through tools like ChatGPT, Google Gemini, and
DALL-E, we have witnessed the democratization of high-tier cognitive and creative
capabilities.
While these technologies offer unprecedented efficiency—enabling rapid prototyping and
personalized education—they also introduce significant challenges, including
"hallucinations," copyright complexities, and the urgent need for digital literacy.
Ultimately, the success of these tools rests on their ability to act as collaborative partners
that enhance human ingenuity rather than replacing it.

Future Outlook
Looking ahead toward the late 2020s, the trajectory of AI is moving toward Artificial
General Intelligence (AGI) and a state of seamless human-AI symbiosis.
• Proactive "Life-OS": Systems like Gemini are evolving into proactive assistants that
anticipate user needs—such as organizing itineraries or suggesting gifts—
reducing the "cognitive load" on humans .
• Total Generative Environments: In the visual domain, the architecture of DALL-E
is expected to merge with video models to create fully interactive 3D
environments for VR, where entire worlds are generated from a single prompt .
• Agentic Independence: ChatGPT and similar models are entering the "Agentic
Era," shifting from passive chatbots to active executors capable of managing
complex, multi-step workflows with minimal human intervention .
• Ethical Infrastructure: Future development will rely heavily on universal
"watermarking" (such as SynthID or C2PA metadata) and "Constitutional AI"
principles to ensure digital authenticity and safety in an AI-saturated world.
As these models continue to evolve, the focus remains on ensuring they are transparent,
controllable, and fundamentally beneficial to all of society.

17 | P a g e
 Bibliography
[Link] Material type Title Author(s) Key Focus
Exploring generative Shoitan, R., et al. Comprehensive
1. Journal artificial intelligence: a (2026) overview of GAI
Article comprehensive guide evolution from
Markov Chains
to ChatGPT and
DALL-E.
Generative AI in Álvarez Ariza, J., Analysis of
2. Research Engineering and et al. (2025) empirical studies
Paper Computing Education: and educational
A Scoping Review practices
involving tools
like Google
Gemini and
ChatGPT
(Álvarez Ariza et
al., 2025).
3. Academic ChatGPT and the Bakagianni, et al. Detailed
Review Future of Generative (2025) examination of
AI: Architecture, Transformer
Limitations, and architecture,
Advancements RLHF, and
technical
limitations like
hallucinations.
4. Scientific Transformation of Various (2026) Documentation
Preprint Large Language of the
Models: A State of Art progression from
GPT-1 to GPT-4o
and Gemini 1.5.
5. Technical Sparsity and Chaudhari, M., Explores the
Paper Superposition in et al. (2025) Mixture of
Mixture of Experts Experts (MoE)
logic used in
models like
Google Gemini
(Chaudhari et al.,
2025).
6. Security Exposing the Villa, et al. Contrast
Report Guardrails: Reverse- (2025) between DALL-E
Engineering Safety 2’s blocklists and
18 | P a g e
 Filters in DALL-E DALL-E 3’s
LLM-based
prompt revision.
7. Impact Study The Impact of Dwivedi, et al. Examines how
Generative AI on (2024); text-to-image
Creative Industries Walkowiak & tools facilitate
Potts (2025) swift concept
visualization in
design.
8. Educational University students' Noroozi, et al. Quantitative
Study perceptions of (2025) study on how
generative AI for students use AI
critical thinking for critical
thinking and
creativity
support.

19 | P a g e