Which of the following companies developed the GPT family of models?
OpenAI
Which embedding models are in GPT? Multiple answers.
text-embedding-3-small
text-embedding-3-large
text-embedding-ada-002
Models that can generate and edit images, given a natural language prompt.
GPT Image 1
DALL E 3
Which model is a fast, affordable small model for focused tasks
GPT4o-mini
GPT-o3 is a well-rounded and powerful model across domains. It sets a new
standard for math, science, coding, and visual reasoning tasks. It also excels at
technical writing and instruction-following.
How many context window and max output tokens could this GPTo4 do?
200,000 context window
100,000 max output tokens
Claude 3 is developed by which company?
Anthropic
Gemini models are part of which company's AI suite?
Google
Which of the following LLMs is open-source?
LLaMA 3
Which of the following LLMs is known to be optimized for reasoning tasks as of
mid-2024?
GPT-4 Turbo
Claude 3 Opus
What is the maximum context length of GPT-4 Turbo?
128K tokens
What is the maximum context length for Claude 3 Opus?
200K+ tokens
What models has Gemini recently released ?
gemini-2.5-pro
gemini-2.5-flash
Gemini 1.5 models support a context window of up to:
1M tokens
Which model has been shown to retain strong performance over long contexts (e.g.,
>500K tokens)?
Gemini 1.5
LLaMA 3 is currently available in which model sizes?
8B and 70B
Which of these models is best known for strong logical and mathematical reasoning?
GPT-4 Turbo
Claude 3 Opus
Gemini 1.5 Pro
Which of these models supports image generation natively (via its own tools)?
GPT-4 Turbo (via DALL·E)
What is the name of OpenAI's model for image generation?
DALL·E
What LLM can generate images natively inside Google's AI suite?
Imagen
Which of the following models includes multimodal capabilities (text + vision)?
All of the above
What is one key advantage of Claude 3 models?
Human-like, non-robotic tone
What differentiates GPT-4 Turbo from GPT-4?
It's cheaper and faster
Why is Gemini 1.5 notable in terms of memory?
Has 1 million token context
What is a key strength of open-source models like Mistral or LLaMA?
Customizability and privacy
Which model is known for fast and accurate summarization of long documents?
Both Gemini 1.5 and Claude 3
Which LLM is considered best for legal or scientific summarization in 2024?
GPT-4 Turbo
Claude 3 Opus
Which model allows users to upload and analyze PDFs directly in ChatGPT?
GPT-4 Turbo
Claude 3's tone is often described as:
Friendly and helpful
Which model is best for low-latency and on-device use?
LLaMA 3
If you want a customizable model that runs locally for enterprise use, your best bet
is:
Mistral or LLaMA
What is a common disadvantage of GPT-4 Turbo?
Expensive in API use
A major limitation of Claude 3 models is:
No access to web search
What is a drawback of open-source models like LLaMA and Mistral?
Lower performance compared to proprietary models
Gemini models sometimes face criticism for:
Slow inference speed
Which of the following is a drawback of GPT-3.5?
Weak in multi-step reasoning
Which of the following LLMs uses a mixture-of-experts (MoE) architecture in some
versions?
Gemini 1.5
Which tokenizer do GPT-4 models use?
tiktoken
Which statement best describes Claude 3 Opus's behavior in ambiguous prompt
scenarios?
Attempts clarification or reflects uncertainty
What is a key architectural feature that allows Gemini 1.5 to handle ultra-long
contexts (>1M tokens)?
Sliding window self-attention
What is the key difference between LLaMA 2 and LLaMA 3 in terms of training
optimization?
LLaMA 3 includes more alignment tuning
Which model currently scores highest on complex reasoning benchmarks like
MMLU and GSM8K (as of mid-2024)?
Claude 3 Opus
Which of the following tasks is Claude 3 less optimized for compared to GPT-4
Turbo?
Programming
What is the likely reason GPT-4 Turbo has cheaper inference costs than GPT-4?
Based on a dense, efficient MoE-style architecture
Which model is most efficient at answering factual queries over long documents
(100K+ tokens)?
Gemini 1.5
Which LLM shows lower hallucination rate under chain-of-thought prompting?
Claude 3 Opus
If an LLM advertises a 1M-token context window, what challenge is most relevant?
GPU memory pressure during inference
Why might models like Claude or Gemini underperform in structured code output
compared to GPT-4 Turbo?
Less focus on code-related alignment
In terms of encoding efficiency, which tokenizer produces the fewest tokens for
multilingual content?
SentencePiece
Which transformer variant is best suited for long-context efficiency?
Sliding window / chunked attention
What is a key limitation of GPT-style decoder-only transformers for retrieval-based
tasks?
No simultaneous bidirectional attention
Which of the following is a risk of increasing the context window without proper
training adaptation?
Loss of semantic coherence
Why might Claude 3 outperform GPT-4 Turbo on tasks requiring sensitivity to
human intent?
Constitutional AI alignment
What is a potential disadvantage of mixture-of-experts architectures?
Sparse routing complexity and latency
Why are open-source models like Mistral and LLaMA gaining popularity in
enterprise contexts?
On-premise privacy & low latency
What kind of alignment training is unique to Anthropic’s Claude models?
Reinforcement learning from human feedback (RLHF)
Which LLM family is most optimized for integration into real-time collaborative
tools (e.g., Docs, Sheets)?
Gemini
Why is Gemini 1.5 considered particularly strong for long-document Q&A?
Compression-aware attention mechanisms
What is one known problem when using extremely long inputs with models like
Gemini or Claude?
Context dilution — attention is spread too thin
Which of the following tasks remains challenging even for top-tier LLMs in 2024?
Multi-hop factual QA with source grounding
What is one architectural or training reason LLaMA 3 is easier to fine-tune than
GPT-4?
Weights and architecture are public
Which company offers the most complete multimodal API across vision, audio, text,
and image generation?
OpenAI
Gemini’s development strategy aligns most closely with which of the following
goals?
Seamless integration into Google Workspace
What is one core disadvantage of using Claude 3 in real-world customer-facing
agents?
Slower inference latency
Which model has the strongest code interpretation and generation capabilities as of
mid-2024?
GPT-4 Turbo
What trade-off does GPT-4 Turbo likely make to achieve higher speed and lower
cost?
Use of sparse activation (experts) and more batching
GPT-4.0 Turbo differs from GPT-4.0 by:
Faster inference and lower cost per token
PT-4.0's token limit is:
Up to 1 million tokens (with special context-aware mode)
A key innovation in the newest GPT-4.0 update is:
Hierarchical memory mechanism for long conversations
“DeepSeaK” (or DeepSeek) focuses on:
Scientific knowledge retrieval
DeepSeaK’s architectural highlight is:
Mixture-of-experts with sparse activation
Which is the go-to embedding model from OpenAI?
text-embedding-ada-002
Semantic similarity on embeddings is typically calculated via:
Cosine similarity
Increasing dimensions in an embedding (e.g., 1024 → 2048) typically results in:
Higher compute cost and finer semantic distinction
How are embeddings used in retrieval-augmented generation (RAG)?
By selecting context documents via vector matching
Which tuning method improves embedding quality for specific domains?
Contrastive fine-tuning
GPT-4.0 supports how many languages for reasoning tasks?
~100+
Which model was specifically optimized for code-mixed or low-resource languages?
BLOOMZ
Zero-shot translation performance is best achieved by:
GPT-4.0
Multilingual embeddings benefit from:
Shared embedding space across languages
What problem emerges when applying monolingual tokenizers to multilingual
corpora?
Increased token counts and fragmentation
GPT-4.0 is preferable to DeepSeeK for tasks requiring:
Complex reasoning and generative planning
DeepSeaK’s retrieval injection helps reduce:
Hallucinations from unsupported content
How do embeddings & cosine similarity help GPT-4.0 in summarizing user manuals
They surface relevant segments for context before summarizing
For multilingual sentiment analysis, the best approach is to use:
Universal sentence embeddings + lightweight classifier
The trade-off of using massive context windows (e.g., 1M tokens) is:
Higher latency and memory usage