In sequence classification tasks that require processing pairs of sentences, the [SEP] token acts as a dedicated separator. This special symbol explicitly delineates the boundary between two distinct sentences when they are combined into a single input sequence.

Google

A simple method for designing a self-supervised classification task to train an encoder is Next Sentence Prediction (NSP), as presented in the original BERT paper. This approach is built on the assumption that a good text encoder should effectively capture the relationship between two sentences. To model this, NSP uses the output of encoding two consecutive sentences, $$\mathrm{Sent}_{A}$$ and $$\mathrm{Sent}_{B}$$, to determine whether $$\mathrm{Sent}_{B}$$ is indeed the next sentence following $$\mathrm{Sent}_{A}$$. For example, if $$\mathrm{Sent}_{A}$$ is 'It is raining .' and $$\mathrm{Sent}_{B}$$ is 'I need an umbrella .', the model is tasked with recognizing this sequential relationship.

Next Sentence Prediction (NSP)

Reference of Foundations of Large Language Models Course

In sentence-pair tasks such as Next Sentence Prediction (NSP), two sentences are combined into a single input sequence. A special start token, `[CLS]`, is prepended, while a `[SEP]` token separates the two sentences. For example, a training sample formatted as `[CLS] It is raining . [SEP] I need an umbrella . [SEP]` is classified with the label `IsNext`, indicating that the second sentence logically follows the first.

Example of Next Sentence Prediction (NSP) Input Formatting

To create training data for the Next Sentence Prediction (NSP) task, pairs of sentences (SentA and SentB) are sampled. Positive examples are generated by taking two consecutive sentences from a text corpus. Negative examples are created by pairing a sentence with another sentence randomly selected from the corpus. This process effectively transforms the NSP task into a binary classification problem.

Training Data Generation for Next Sentence Prediction

The Next Sentence Prediction (NSP) task is typically not the sole objective during model pre-training. Instead, it is often employed as an additional, or auxiliary, training loss. This means the model is trained to optimize for the NSP objective simultaneously with other objectives, such as Masked Language Modeling, to learn a more comprehensive understanding of language.

Next Sentence Prediction as an Auxiliary Training Objective

A key criticism of the Next Sentence Prediction (NSP) task is that it may not be sufficiently challenging. This relative simplicity can encourage the model to learn to make predictions based on superficial evidence rather than developing a deeper semantic understanding of the relationship between sentences.

Limitation of Next Sentence Prediction: Reliance on Superficial Cues

An example of an input with two unrelated sentences for a Next Sentence Prediction (NSP) task is `[CLS] The cat sleeps on the windowsill . [SEP] Apples grow on trees . [SEP]`. Because the second sentence does not logically follow the first, this training sample is classified with the label `NotNext`. This serves as a negative example to train the classification system.

Example of an Unrelated Sentence Pair for NSP

As proposed in the original paper by Devlin et al. (2019), the standard BERT model is a Transformer encoder pre-trained with a dual-task objective. This training process involves simultaneously learning from two tasks: Masked Language Modeling (MLM) and Next Sentence Prediction (NSP). The total training loss is calculated as the sum of the individual losses from these two objectives.

Training Objective of the Standard BERT Model

Analyze the pre-training strategy described in the case study below. Explain the underlying reasoning for why this specific classification task would be beneficial for the model's intended final application.

Pre-training Strategy for a Question-Answering Model

Some pre-training objectives, such as predicting the relationship between two simple, consecutive sentences, may not be sufficiently challenging. This can inadvertently train the model to rely on superficial patterns or 'easy evidence' for its predictions, rather than fostering a deeper, more robust understanding of language.

Potential for Learning Superficial Cues in Simple Prediction Tasks

A language model is pre-trained on a large corpus of text using a specific objective: for any given pair of sentences, the model must predict whether the second sentence is the one that actually follows the first in the source document. Which of the following best describes the primary type of understanding this training method is intended to instill in the model?

A language model is pre-trained exclusively on a task where it learns to predict if one sentence immediately follows another in a large text corpus. While the model achieves high accuracy on this pre-training task, it struggles when fine-tuned for tasks requiring nuanced logical inference between sentences. Which of the following statements provides the most insightful critique of the pre-training task, explaining this performance gap?

Your team is building an internal model that must ...

Your team is pre-training a text model for an inte...

Your team is pre-training an internal LLM for a co...

Your team is pre-training an internal LLM to suppo...

You lead an internal LLM initiative for a company with two high-value use cases: (1) a customer-support assistant that must generate policy-compliant, fluent responses (long-form text generation), and (2) an enterprise search/reranking system that must accurately judge whether two pieces of text belong together (e.g., a ticket description and a proposed resolution, or a question and a candidate answer). You have a fixed pre-training budget and must choose ONE primary pre-training objective and optionally ONE auxiliary objective from the following families: masked token reconstruction (masked language modeling), sentence-pair coherence classification (next sentence prediction), left-to-right next-token prediction (causal language modeling), reconstructing clean text from a corrupted input using an encoder-decoder denoising objective, and predicting tokens under a random permutation order (permuted language modeling).

Write a recommendation memo that (a) selects your primary objective and optional auxiliary objective, (b) explains how the information flow/conditioning structure of your chosen objectives will shape what the model learns for BOTH use cases, and (c) explicitly discusses at least two tradeoffs you are accepting (e.g., bidirectional understanding vs generation behavior, exposure to [MASK]/corruption artifacts vs realistic inference-time inputs, learning inter-sentence relations vs relying on superficial cues, or permutation-based context vs strict left-to-right factorization). Your answer should make clear why your chosen combination is better than at least one plausible alternative combination for this company scenario.

Selecting a Pre-training Objective Mix for a Corporate LLM

You lead an internal team building an LLM-based assistant for a regulated enterprise. After pre-training and light fine-tuning, you observe three recurring failures in pilot use:

1) In long-form drafting, the model often contradicts itself across paragraphs and sometimes “jumps” to content that would only make sense if it had seen later text.
2) In document review, the model is strong at filling in missing words inside a sentence, but it is unreliable at deciding whether a proposed follow-up sentence is a coherent continuation of the previous one (e.g., it misses topic shifts and non sequiturs).
3) In reconstruction-style tasks (e.g., restoring a partially corrupted policy excerpt), the model sometimes produces fluent text that does not faithfully match the original wording, even when the corruption is minor.

Assume you can choose among these pre-training objectives (alone or in combination): masked token prediction, next-sentence relationship classification, left-to-right next-token prediction, denoising reconstruction from corrupted inputs using an encoder-decoder, and permuted-order token prediction.

Write a recommendation memo that (a) identifies the most likely objective-level causes of the three failures, and (b) proposes a revised objective mix (and why) that would reduce all three issues. Your answer must explicitly connect each proposed change to how information is (or is not) allowed to flow during training (bidirectional context vs strictly past context vs permuted context) and to the difference between predicting a few tokens vs reconstructing an entire sequence vs classifying sentence-pair coherence.

Diagnosing Pre-training Objective Mismatch from Product Failures

You lead an internal team pre-training a language model for two production uses: (1) a contract-review assistant that must accurately fill in missing clauses inside long documents (insertion and span infilling), and (2) a customer-support agent that must generate responses token-by-token with low latency. Your training data is a mix of clean policy documents and noisy OCR’d PDFs where words are frequently dropped, duplicated, or out of order. Compute budget allows only ONE primary pre-training objective (you may add at most one lightweight auxiliary loss).

Write a recommendation memo that:
- Selects the primary objective from: masked language modeling, causal language modeling, denoising autoencoder reconstruction, or permuted language modeling.
- Justifies the choice by explicitly analyzing how the objective’s information flow (bidirectional vs left-to-right vs permuted) interacts with (a) the OCR noise pattern and (b) the two deployment requirements (infilling vs streaming generation).
- Proposes one auxiliary objective (either next sentence prediction or none) and argues for/against it, including at least one concrete risk of relying on superficial cues.

Your answer should make clear tradeoffs (what you gain and what you give up) and explain why your chosen combination is the best fit for this scenario.

Choosing a Pre-training Objective Under Data Constraints and Deployment Needs

You lead an internal team building an enterprise writing assistant that must (1) generate long, coherent policy drafts from a short prompt, (2) accurately fill in missing clauses inside existing documents during redlining, and (3) decide whether a proposed paragraph logically follows the previous paragraph in a policy (to flag non-sequiturs). You have budget to pre-train ONE base model from scratch and can choose ONE primary training objective; you may optionally add ONE auxiliary objective if you can justify why it complements the primary objective without undermining it. The candidate objectives are: causal language modeling, masked language modeling, denoising autoencoder reconstruction (corrupt input then reconstruct), permuted language modeling, and next sentence prediction.

Case constraints: Your training corpus is mostly internal policies and emails; at inference time the assistant must support both free-form generation and in-place editing. Early prototypes show two failure modes you must address: (A) generated drafts are locally fluent but drift off-topic after ~2 pages, and (B) in-place clause completion is often grammatically correct but contradicts a constraint stated later in the same paragraph.

Which objective would you choose as the PRIMARY objective, and which (if any) would you add as an AUXILIARY objective? In your answer, explicitly connect your choice to how information is (or is not) available to the model during training (bidirectional vs. left-to-right vs. permuted/denoising), and explain how your choice mitigates BOTH failure modes (A) and (B) while still supporting requirement (3) about paragraph-to-paragraph coherence.

Pre-training Objective Choice for a Multi-Modal Enterprise Writing Assistant

You lead an internal team pre-training an LLM for a corporate knowledge assistant that must (1) generate long, coherent incident summaries and (2) support sentence-pair tasks such as “Does this policy clause entail that control requirement?” During a red-team exercise, the model shows two issues: (a) when asked to fill a missing word in the middle of a sentence, it often produces plausible but inconsistent completions, and (b) when generating multi-paragraph summaries, it sometimes contradicts earlier sentences. Your training pipeline used an encoder-only Transformer with a masked-token objective plus a binary classifier trained to predict whether Sentence B follows Sentence A; you did not train any left-to-right next-token objective. A colleague proposes switching to an encoder-decoder denoising objective (reconstruct clean text from corrupted text) and adding a permuted prediction objective; another colleague proposes replacing everything with a pure causal (left-to-right) language modeling objective.

As the decision-maker, which proposal (or combination) would you choose to address BOTH issues with the least mismatch to the two product requirements, and why? In your answer, explicitly connect how the information available at training time (bidirectional masking vs sentence-pair classification vs left-to-right generation vs reconstruction-from-noise vs permuted prediction order) would causally affect (a) mid-sentence infilling quality and (b) long-form generation consistency.

Root-Cause Analysis of Pre-training Objective Leakage and Coherence Failures

You lead model strategy for an internal enterprise assistant used by legal and compliance teams. The assistant must (1) generate long, coherent policy drafts and email responses, (2) answer questions that require understanding relationships across two adjacent paragraphs (e.g., “Does the exception in paragraph 2 apply to the rule in paragraph 1?”), and (3) be robust to noisy inputs from OCR and copy/paste (missing words, duplicated phrases, and occasional sentence reordering). You have budget to pre-train ONE base model from scratch and can choose ONE primary pre-training objective from the following families: causal language modeling (left-to-right next-token prediction), masked language modeling (predict masked tokens using both left and right context), denoising autoencoder reconstruction (reconstruct clean text from a corrupted version), permuted language modeling (predict tokens in a random permutation order), and optionally add Next Sentence Prediction (NSP) as an auxiliary loss.

Case study task: Choose the single primary objective you would use and decide whether you would include NSP as an auxiliary loss. Justify your choices by explicitly explaining the tradeoffs among (a) generation quality for long outputs, (b) bidirectional understanding within a span, (c) modeling cross-sentence/paragraph relationships, and (d) robustness to the specified noise patterns. Your answer should make clear why at least two of the non-chosen objectives are less suitable given these constraints.

Selecting a Pre-training Objective for a Regulated Enterprise Assistant

In the Next Sentence Prediction (NSP) task, the relationship between two sentences is evaluated by classifying each training sample into a binary label set, consisting of $$\mathrm{IsNext}$$ and $$\mathrm{NotNext}$$. The model processes the input sequence and utilizes the encoder's output vector for the initial `[CLS]` token. This vector, denoted as $$\mathbf{h}_{\mathrm{cls}}$$ (or $$\mathbf{h}_0$$), serves as the sequence representation. A classifier is then constructed on top of $$\mathbf{h}_{\mathrm{cls}}$$ to predict the correct sequence label.

Binary Classification System for Next Sentence Prediction

A classifier can be constructed on top of the sequence representation vector, denoted by $${}\mathbf{h}_{\mathrm{cls}}$$ (or $${}\mathbf{h}_0$$), which corresponds to the encoder's output for the initial $${}[\mathrm{CLS}]$$ token. Using this framework, one can compute the conditional probability of a specific label $${}c$$ based on the representation, mathematically expressed as $${}\Pr(c|\mathbf{h}_{\mathrm{cls}})$$. While many loss functions are available for such classification problems, maximum likelihood training often involves defining a specific loss, such as $${}\mathrm{Loss}_{\mathrm{NSP}}$$ for Next Sentence Prediction tasks.

Learn Before

Related