সহকারী প্রোগ্রামার নিয়োগ পরীক্ষার প্রশ্ন

সঠিক উত্তর: (b) OpenAI

ChatGPT (চ্যাটজিপিটি) হলো মার্কিন AI প্রতিষ্ঠান OpenAI-এর তৈরি chatbot। এটি OpenAI-এর নিজস্ব GPT (Generative Pre-trained Transformer) মডেলের ওপর চলে।

অন্য অপশনগুলো কেন ভুল

• (a) Microsoft: OpenAI-এর বড় অংশীদার; Microsoft-এর Copilot সেবা OpenAI-এর মডেলে চলে, কিন্তু ChatGPT Microsoft তৈরি করেনি।
• (c) Google: Google-এর নিজস্ব AI chatbot হলো Gemini (আগের নাম Bard), ChatGPT নয়।
• (d) Meta: Meta আলাদা প্রতিষ্ঠান, যার নিজস্ব AI মডেল আছে; ChatGPT-র সঙ্গে এর নির্মাতা হিসেবে কোনো সম্পর্ক নেই।

টিপ: ChatGPT → OpenAI; Copilot → Microsoft (OpenAI-এর মডেলে); Gemini → Google।

উৎস: What is GPT (generative pretrained transformer)? (IBM)

Correct answer: (c) Testing Requirements

A Software Requirements Specification (SRS) states what the software must do and under which constraints. SWEBOK describes functional requirements as the observable behaviours the software must provide, and treats test specifications and results as separate documents from the SRS. How the software will be tested (test plans, test cases, test procedures) is written in test documentation derived from the SRS, not in the SRS itself.

Why the other options are wrong

• (a) Functional Requirements: the core of every SRS: the features and behaviour the system must provide.
• (b) Non-Functional Requirements: quality and constraint requirements (performance, security, usability, reliability) are a standard SRS section.
• (d) Interface Requirements: user, hardware, software and communication interfaces are also specified in the SRS.

Exam note: an SRS should state each requirement so that it can be verified, but the test plan and test cases themselves are separate deliverables.

Source: SWEBOK Guide V4.0a (IEEE Computer Society)

Correct answer: (d) Generative Pre-trained Transformer.

GPT stands for Generative Pre-trained Transformer, a family of large language models (LLMs) developed by OpenAI:

• Generative: it produces new content, such as text.
• Pre-trained: it is first trained on a very large body of text before being adapted to specific tasks.
• Transformer: it uses the transformer neural-network architecture.

Why the other options are wrong

• (a) General Pre-trained Transformer.: The "G" is Generative, not General; this is the usual trap.
• (b) Generative Program Technology.: "P" is Pre-trained and "T" is Transformer; GPT is a model, not a "program technology".
• (c) Generative Process Technology.: Also an invented expansion; the "P" and "T" are wrong.

Tip: GPT = Generative Pre-trained Transformer (the model behind ChatGPT).

Source: What is GPT (generative pretrained transformer)? (IBM)

Correct answer: (b) Implemented

An interface declares a contract (method signatures, constants, and since Java 8 default and static methods). It does nothing on its own until a class writes class C implements I and supplies a body for each abstract method. That is why interfaces are described as meant to be implemented. Oracle's tutorial notes that an interface cannot be instantiated: it can only be implemented by classes or extended by other interfaces.

Why the other options are wrong

• (a) Extended: Only another interface can extend an interface, and that just builds a larger contract which still needs a class to implement it. Extension is possible but is not the purpose.
• (c) Overridden: Methods are overridden; an interface as a whole is not. A class implements the interface's abstract methods.
• (d) Used by creating object: new I() is a compile error, because an interface cannot be instantiated. You create an object of a class that implements it.

Tip: class extends class; class implements interface; interface extends interface.

Source: Defining an Interface / Interfaces - The Java Tutorials (Oracle)

Correct answer: (c) FORTRAN

FORTRAN (FORmula TRANslation) was developed at IBM by a team led by John Backus and debuted in 1957 for the IBM 704. It let scientists write formula-like statements instead of machine code, and its compiler produced code efficient enough to win over sceptics. The Computer History Museum describes it as the first high-level programming language and the first high-quality optimising compiler.

Why the other options are wrong

• (a) C: created at Bell Labs in the early 1970s, about 15 years after FORTRAN.
• (b) COBOL: the business-oriented language specified in 1959 and released around 1960, two to three years after FORTRAN.
• (d) C++: an extension of C from the 1980s, the latest of the four.

Tip: First high-level language = FORTRAN (1957, science); then COBOL (business), then C, then C++.

Note: earlier designs existed on paper or in limited use (e.g. Plankalkül, designed in the 1940s but not implemented then), but FORTRAN is the first successful, widely used one and the expected exam answer.

Sources: History of FORTRAN and FORTRAN II – Computer History Museum; Fortran – IBM History

Correct answer: (c) 3

Each node of a doubly linked list must be able to move both forward and backward, so it needs at least:

1. data: the value stored in the node
2. prev: a pointer to the previous node
3. next: a pointer to the next node

That makes 3 fields.

Why the other options are wrong

• (a) 1: Data alone gives no links at all, so there would be no list.
• (b) 2: Data plus one link is a singly linked list node, which can move in only one direction.
• (d) 4: A fourth field (for example a key alongside the data) is optional and not part of the minimum.

Tip: Minimum fields: singly linked list = 2 (data, next) and doubly linked list = 3 (prev, data, next).

Source: Problem Session 2 - 6.006 Introduction to Algorithms, MIT OpenCourseWare (Spring 2020)

Correct answer: (c) broadcast addressing

IPv6 defines three kinds of address: unicast, anycast and multicast. RFC 4291 says plainly: "There are no broadcast addresses in IPv6". Jobs that IPv4 did with broadcast are done with multicast groups instead, for example the all-nodes link-local group ff02::1, so only hosts that have joined a group need to process the packet.

Why the other options are wrong

• (a) unicast addressing: Supported; a unicast address identifies a single interface.
• (b) multicast addressing: Supported, and it takes over the role of broadcast; the packet goes to every interface in the group.
• (d) anycast addressing: Supported; the packet goes to the "nearest" of several interfaces sharing the address, as measured by routing.

Tip: IPv4: unicast, broadcast, multicast. IPv6: unicast, multicast, anycast (no broadcast).

Source: RFC 4291: IP Version 6 Addressing Architecture

Correct answer: (b) Graph

In a non-linear data structure an element can be connected to several others, so the elements do not form a single line. A graph is the clearest case: a vertex can have edges to any number of other vertices, and there is no single "next" element. Trees are the other standard non-linear structure.

Why the other options are wrong

• (a) Array: elements sit one after another by index, a single sequence, so it is linear.
• (c) Queue: elements are kept in arrival order (FIFO), one behind another, so it is linear.
• (d) Linked list: each node points to one next node, forming a chain, so it is linear.

Tip: Linear = array, stack, queue, linked list. Non-linear = tree, graph.

Source: Lecture Plan: Data Structures (CSC 13102), IIT (ISM) Dhanbad

Correct answer: (a) Tunneling

Tunneling means wrapping a whole packet of one protocol inside a packet of another, so it can cross a network that speaks only the outer protocol. The entry point adds the outer header (encapsulation), and the exit point removes it (decapsulation). It works in both directions: IPv6 packets are tunnelled over IPv4 networks (configured tunnels, RFC 4213), and IPv4 packets are tunnelled over IPv6-only networks, as in Dual-Stack Lite, which combines IPv4-in-IPv6 tunnels with NAT.

Why the other options are wrong

• (b) hashing: Computes a fixed-length digest of data for integrity or lookup; it does not carry packets.
• (c) NAT: Rewrites addresses in the existing header instead of wrapping the packet; even NAT64 translates between IPv6 and IPv4 headers, which is translation, not encapsulation.
• (d) Traversing: A general word (as in "NAT traversal"), not the name of an IPv4/IPv6 transition mechanism.

Tip: The three IPv4–IPv6 transition approaches: dual stack, tunneling (encapsulation) and translation (NAT64).

Sources: RFC 4213: Basic Transition Mechanisms for IPv6 Hosts and Routers; RFC 6333: Dual-Stack Lite Broadband Deployments Following IPv4 Exhaustion

Correct answer: (d) Structure reference

"Structure reference" is the name of the member-access (dot) operator a.b, which selects member b of object a. C++ does not allow . to be overloaded; the other operators that cannot be overloaded are :: (scope resolution), .* (member access through a pointer to member) and ?: (conditional). Note that C has no operator overloading at all, so the question is really about C++.

Do not confuse it with "structure dereference" a->b, which can be overloaded (smart pointers do it).

Why the other options are wrong

• (a) Bitwise right shift assignment: >>=, like every compound-assignment operator, can be overloaded, e.g. T& operator>>=(int n);.
• (b) Address of: Unary & can be overloaded, which is why std::addressof exists to get an object's real address.
• (c) Indirection: Unary * is routinely overloaded by iterators and smart pointers, e.g. T& operator*();.

Tip: The ones you can never overload: :: . .* ?:.

Sources: Operator overloading - cppreference.com; Operators in C and C++ - Wikipedia