Techniques & Methods

A

This approach enhances the resilience of machine learning models by exposing them to custom-crafted challenging examples. The idea is to make the model learn more effective and reliable ways to interpret data.

Techniques & Methods

Alignment

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This refers to the process of fine-tuning an LLM to conform to certain ethical guidelines. It aims to mitigate issues arising from uncurated training data, such as biased or harmful outputs."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">This refers to the process of fine-tuning an LLM to conform to certain ethical guidelines. It aims to mitigate issues arising from uncurated training data, such as biased or harmful outputs.

Techniques & Methods

Attention

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In the context of GPT's transformer architecture, attention refers to the model's ability to understand the relational context between words in a given text sequence. This technique was originally outlined in the paper \"Attention Is All You Need.\""}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">In the context of GPT's transformer architecture, attention refers to the model's ability to understand the relational context between words in a given text sequence. This technique was originally outlined in the paper "Attention Is All You Need."

Techniques & Methods

Attention Mechanism

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This technique, employed in deep learning models, assigns varying degrees of significance or \"attention\" to different words or features during data processing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This technique, employed in deep learning models, assigns varying degrees of significance or "attention" to different words or features during data processing.

Techniques & Methods

Autoregression

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In language models, this refers to the dependency of future token predictions on the tokens previously generated by the model, not just the initial input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In language models, this refers to the dependency of future token predictions on the tokens previously generated by the model, not just the initial input.

B

Techniques & Methods

Backpropagation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This algorithm is used to train neural networks by optimizing the weights of the network in relation to the gradient of a loss function. It is essential for accurately fine-tuning the model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This algorithm is used to train neural networks by optimizing the weights of the network in relation to the gradient of a loss function. It is essential for accurately fine-tuning the model.

Techniques & Methods

Backward Chaining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A reasoning approach where the algorithm starts from a desired outcome and works its way backward to identify the data or conditions that could lead to that outcome."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A reasoning approach where the algorithm starts from a desired outcome and works its way backward to identify the data or conditions that could lead to that outcome.

Techniques & Methods

Bandit Optimization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A real-time decision-making strategy in machine learning that balances the act of exploring new options with exploiting known effective options, all while having limited information."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A real-time decision-making strategy in machine learning that balances the act of exploring new options with exploiting known effective options, all while having limited information.

Techniques & Methods

Beam Search

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing (NLP), this is an algorithm that sifts through sequences of tokens to find the most probable sequence. It does so by keeping track of a fixed number of the best sequences at each step."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing (NLP), this is an algorithm that sifts through sequences of tokens to find the most probable sequence. It does so by keeping track of a fixed number of the best sequences at each step.

C

Techniques & Methods

Chain-of-Thought

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A prompting technique designed to enhance the reasoning abilities of Large Language Models (LLMs). The technique breaks down complex tasks or questions into simpler steps for better understanding, and may even involve additional data gathering through web searches. A more advanced version, known as Tree-of-Thought, allows for logical branching and backtracking."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A prompting technique designed to enhance the reasoning abilities of Large Language Models (LLMs). The technique breaks down complex tasks or questions into simpler steps for better understanding, and may even involve additional data gathering through web searches. A more advanced version, known as Tree-of-Thought, allows for logical branching and backtracking.

Techniques & Methods

Completion

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The generated output from a Large Language Model (LLM) in response to a given prompt. This term originates from the way prompts are usually structured, requiring the model to complete a sentence or thought."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The generated output from a Large Language Model (LLM) in response to a given prompt. This term originates from the way prompts are usually structured, requiring the model to complete a sentence or thought.

Techniques & Methods

Coreference Resolution

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The task of identifying and harmonizing different expressions in a text that refer to the same entity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The task of identifying and harmonizing different expressions in a text that refer to the same entity.

D

Techniques & Methods

Data Augmentation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of increasing the volume of training data by adding variations or creating synthetic data based on existing examples."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of increasing the volume of training data by adding variations or creating synthetic data based on existing examples.

Techniques & Methods

Data Mining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The activity of examining datasets to unearth novel patterns, which may enhance model performance."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The activity of examining datasets to unearth novel patterns, which may enhance model performance.

Techniques & Methods

Decoding Rules

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Guidelines that govern the process of text generation in a language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Guidelines that govern the process of text generation in a language model.

Techniques & Methods

Dependency Parsing

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing, this is the act of dissecting a sentence to discern the grammatical connections between its various elements, like subjects, objects, and modifiers."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing, this is the act of dissecting a sentence to discern the grammatical connections between its various elements, like subjects, objects, and modifiers.

Techniques & Methods

Distributed Training

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The technique of spreading the training of an AI model across multiple computing devices, enabling quicker processing and greater data capacity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The technique of spreading the training of an AI model across multiple computing devices, enabling quicker processing and greater data capacity.

E

Techniques & Methods

Entity Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This involves tagging unstructured text with labels that identify specific entities like people, organizations, or locations, making the text machine-readable."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This involves tagging unstructured text with labels that identify specific entities like people, organizations, or locations, making the text machine-readable.

Techniques & Methods

Entity Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A broad term describing the techniques used to add structure to data, usually via annotation, so it becomes interpretable by machine learning algorithms."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A broad term describing the techniques used to add structure to data, usually via annotation, so it becomes interpretable by machine learning algorithms.

Techniques & Methods

Evaluation Metrics

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Quantifiable standards like perplexity, accuracy, F1 score, or BLEU score, used to gauge the effectiveness of a language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Quantifiable standards like perplexity, accuracy, F1 score, or BLEU score, used to gauge the effectiveness of a language model.

Techniques & Methods

Extractive Summarization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This entails generating a summary by selecting key sentences or points from the original text. ChatGPT can perform this function when summarizing large volumes of text."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This entails generating a summary by selecting key sentences or points from the original text. ChatGPT can perform this function when summarizing large volumes of text.

F

Techniques & Methods

Feature Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This involves isolating essential pieces of information from a dataset for easier processing and analysis, a step ChatGPT utilizes to better understand prompts and formulate responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This involves isolating essential pieces of information from a dataset for easier processing and analysis, a step ChatGPT utilizes to better understand prompts and formulate responses.

Techniques & Methods

Few-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The ability of a model to understand a task quickly with very few training examples."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The ability of a model to understand a task quickly with very few training examples.

Techniques & Methods

Fine-Grained Control

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This refers to the capability of a model to produce text with specific characteristics, such as a particular tone or style, by altering its internal settings."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This refers to the capability of a model to produce text with specific characteristics, such as a particular tone or style, by altering its internal settings.

Techniques & Methods

Fine Tuning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The process of adapting a pre-trained model to more specific tasks by training it further on a specialized dataset. This is generally more efficient than starting training from scratch."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The process of adapting a pre-trained model to more specific tasks by training it further on a specialized dataset. This is generally more efficient than starting training from scratch.

Techniques & Methods

Forward Chaining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a reasoning technique where the AI starts with available data to form possible solutions, assessing various hypotheses to identify those most relevant to the issue at hand."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a reasoning technique where the AI starts with available data to form possible solutions, assessing various hypotheses to identify those most relevant to the issue at hand.

G

Techniques & Methods

Generation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Refers to the act of creating new text using a language model. This can be automated via the model's own token predictions or guided by human input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Refers to the act of creating new text using a language model. This can be automated via the model's own token predictions or guided by human input.

Techniques & Methods

Greedy Algorithms

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These algorithms, often used in NLP, aim to find the most probable sequence of tokens by selecting the token with the highest likelihood at each step."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These algorithms, often used in NLP, aim to find the most probable sequence of tokens by selecting the token with the highest likelihood at each step.

H

Techniques & Methods

Hallucination

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This occurs when a generative AI model produces content that deviates from factual or logical expectations. For example, generating an animal image that combines disparate features from different species. Such outcomes are usually considered errors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This occurs when a generative AI model produces content that deviates from factual or logical expectations. For example, generating an animal image that combines disparate features from different species. Such outcomes are usually considered errors.

Techniques & Methods

Heuristics

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are general guidelines or tactics employed for decision-making or problem-solving. ChatGPT uses heuristics to understand input and generate appropriate output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are general guidelines or tactics employed for decision-making or problem-solving. ChatGPT uses heuristics to understand input and generate appropriate output.

I

Techniques & Methods

Inference

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of utilizing a trained model to make predictions or decisions based on new, unseen data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of utilizing a trained model to make predictions or decisions based on new, unseen data.

Techniques & Methods

Information Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is the technique of pulling out structured information from unstructured sources like text or images. It is a process ChatGPT employs to understand user input and formulate relevant responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is the technique of pulling out structured information from unstructured sources like text or images. It is a process ChatGPT employs to understand user input and formulate relevant responses.

J

Techniques & Methods

Joint Probability

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is the likelihood of multiple events happening simultaneously. ChatGPT uses joint probability calculations to understand and respond to user input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is the likelihood of multiple events happening simultaneously. ChatGPT uses joint probability calculations to understand and respond to user input.

K

Techniques & Methods

Knowledge Representation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of formally structuring knowledge so that it is machine-readable. ChatGPT employs this to better understand and respond to queries."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of formally structuring knowledge so that it is machine-readable. ChatGPT employs this to better understand and respond to queries.

L

Techniques & Methods

Linguistic Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of tagging sentence data sets with particular subjects, often for purposes like sentiment analysis or other NLP tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of tagging sentence data sets with particular subjects, often for purposes like sentiment analysis or other NLP tasks.

Techniques & Methods

Low Rank Adaption (LoRA)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A fine-tuning technique that is both memory-efficient and cost-effective, allowing for quicker model training."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A fine-tuning technique that is both memory-efficient and cost-effective, allowing for quicker model training.

M

Techniques & Methods

Markov Decision Process

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A mathematical structure for modeling decision-making scenarios where the result of each decision relies on the current state. This framework is employed by ChatGPT for interpreting inputs and producing fitting responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A mathematical structure for modeling decision-making scenarios where the result of each decision relies on the current state. This framework is employed by ChatGPT for interpreting inputs and producing fitting responses.

Techniques & Methods

Masked Language Modeling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A training technique where selected tokens in an input sequence are obscured, requiring the model to predict them based on the remaining contextual information."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A training technique where selected tokens in an input sequence are obscured, requiring the model to predict them based on the remaining contextual information.

Techniques & Methods

Multitask Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The technique of training a single model on various tasks at once, enhancing its performance and shared learning across different tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The technique of training a single model on various tasks at once, enhancing its performance and shared learning across different tasks.

N

Techniques & Methods

Named Entity Recognition (NER)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An NLP task that involves identifying and categorizing named entities in text, like names of people, places, or organizations."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An NLP task that involves identifying and categorizing named entities in text, like names of people, places, or organizations.

O

Techniques & Methods

Offline Reinforcement Learning (RL)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for training AI models using a pre-collected dataset, without needing real-time environmental interactions."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for training AI models using a pre-collected dataset, without needing real-time environmental interactions.

Techniques & Methods

One-Shot / Few-Shot

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Terminology for prompts that either ask a straightforward question (one-shot) or require multiple examples for improved performance (few-shot)."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Terminology for prompts that either ask a straightforward question (one-shot) or require multiple examples for improved performance (few-shot).

Techniques & Methods

One-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The capability of a model to understand a new task from just a single example during its training phase."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The capability of a model to understand a new task from just a single example during its training phase.

Techniques & Methods

Online Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A machine learning approach where the model gets trained incrementally as new data becomes available."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A machine learning approach where the model gets trained incrementally as new data becomes available.

Techniques & Methods

Overuse Penalty

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A corrective measure applied in ChatGPT’s decoding algorithm to curb repetitive phrasing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A corrective measure applied in ChatGPT’s decoding algorithm to curb repetitive phrasing.

P

Techniques & Methods

Part-of-Speech Tagging (POS)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An NLP task where each word in a sentence is labeled according to its grammatical function, such as a noun or verb."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An NLP task where each word in a sentence is labeled according to its grammatical function, such as a noun or verb.

Techniques & Methods

Pre-training in AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The initial phase of training a language model on a massive dataset before honing it for specialized tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The initial phase of training a language model on a massive dataset before honing it for specialized tasks.

Techniques & Methods

Prompt

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The textual stimulus presented to a large language model (LLM). Prompts can range from short queries, like \"What's the capital of England?\", to lengthy textual descriptions of problems or tasks. Models often have token limitations, affecting the length of acceptable prompts."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The textual stimulus presented to a large language model (LLM). Prompts can range from short queries, like "What's the capital of England?", to lengthy textual descriptions of problems or tasks. Models often have token limitations, affecting the length of acceptable prompts.

Techniques & Methods

Prompt Engineering

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The art of carefully crafting or selecting the query or statement fed into a machine learning model in order to yield a specific, desired output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The art of carefully crafting or selecting the query or statement fed into a machine learning model in order to yield a specific, desired output.

Techniques & Methods

Prompt Injection

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for influencing or manipulating the behavior of products like chatbots or GitHub Copilot that employ LLM technology. This can reveal internal system settings or undermine alignment training."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for influencing or manipulating the behavior of products like chatbots or GitHub Copilot that employ LLM technology. This can reveal internal system settings or undermine alignment training.

Techniques & Methods

Proximal Policy Optimization (PPO)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An algorithm employed in reinforcement learning to streamline the training process for models."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An algorithm employed in reinforcement learning to streamline the training process for models.

Q

Techniques & Methods

Query

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A request made to a database or other source of information to extract data. ChatGPT utilizes queries to pull pertinent information for generating responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A request made to a database or other source of information to extract data. ChatGPT utilizes queries to pull pertinent information for generating responses.

R

Techniques & Methods

Regularization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A method for curtailing overfitting by incorporating a penalty term into the loss function to limit model complexity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A method for curtailing overfitting by incorporating a penalty term into the loss function to limit model complexity.

Techniques & Methods

Reinforcement Learning from Human Feedback (RLHF)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for aligning models by using human feedback to train a reward model, subsequently fine-tuning a large language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for aligning models by using human feedback to train a reward model, subsequently fine-tuning a large language model.

Techniques & Methods

Response Quality

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An evaluation metric considering how well an AI system replies to prompts, taking into account the relevance, coherence, and accuracy of the response."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An evaluation metric considering how well an AI system replies to prompts, taking into account the relevance, coherence, and accuracy of the response.

Techniques & Methods

Retrieval Augmented Generation (RAG)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of enriching a model's prompt with additional information obtained from web searches or private documents to generate a more informed output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of enriching a model's prompt with additional information obtained from web searches or private documents to generate a more informed output.

S

Techniques & Methods

Scaling Laws

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An observed principle in AI stating that model performance typically improves with increases in data, computational power, and model size."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An observed principle in AI stating that model performance typically improves with increases in data, computational power, and model size.

Techniques & Methods

Self-Attention

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A mechanism within Transformer-based models allowing them to discern relationships among tokens in an input sequence, thereby informing the output generation process."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A mechanism within Transformer-based models allowing them to discern relationships among tokens in an input sequence, thereby informing the output generation process.

Techniques & Methods

Semantic Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of tagging queries or products with relevant markers to improve search engine accuracy."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of tagging queries or products with relevant markers to improve search engine accuracy.

Techniques & Methods

Semantic Similarity

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A metric evaluating the degree of meaning similarity between two text snippets, usually in a vector space."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A metric evaluating the degree of meaning similarity between two text snippets, usually in a vector space.

Techniques & Methods

Sequence Generation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of using a language model to produce a chain of tokens, which could form a sentence or paragraph, based on learned patterns."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of using a language model to produce a chain of tokens, which could form a sentence or paragraph, based on learned patterns.

Techniques & Methods

Supervised Fine-Tuning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The preliminary stage of fine-tuning in which human AI trainers supply dialogues involving both the user and the AI to the model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The preliminary stage of fine-tuning in which human AI trainers supply dialogues involving both the user and the AI to the model.

Techniques & Methods

System Prompt

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A built-in prompt that accompanies user input to set the behavior of a chatbot or other LLM applications. Instances of system prompt leaks have offered insights into how these tools are configured."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A built-in prompt that accompanies user input to set the behavior of a chatbot or other LLM applications. Instances of system prompt leaks have offered insights into how these tools are configured.

T

Techniques & Methods

Text Classification

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The categorization of textual content based on its semantic content. ChatGPT utilizes text classification to understand user queries and generate suitable responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The categorization of textual content based on its semantic content. ChatGPT utilizes text classification to understand user queries and generate suitable responses.

Techniques & Methods

Topic Modeling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A method for identifying prevalent themes across a set of textual documents by analyzing word co-occurrence patterns."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A method for identifying prevalent themes across a set of textual documents by analyzing word co-occurrence patterns.

Techniques & Methods

Training

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The process of exposing a neural network to vast datasets to adjust its parameters and improve its output. Large model training involves enormous datasets and can take significant time."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The process of exposing a neural network to vast datasets to adjust its parameters and improve its output. Large model training involves enormous datasets and can take significant time.

Techniques & Methods

Transfer Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique involving the training of a machine on a related task to enhance its performance on its primary task."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique involving the training of a machine on a related task to enhance its performance on its primary task.

U

Techniques & Methods

Upstream Sampling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A fine-tuning technique used in ChatGPT that generates multiple responses which are then ranked to select the best one."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A fine-tuning technique used in ChatGPT that generates multiple responses which are then ranked to select the best one.

V

Techniques & Methods

Validation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A process in machine learning where a model is assessed using a subset of data it has not seen during training to check its learning efficacy."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A process in machine learning where a model is assessed using a subset of data it has not seen during training to check its learning efficacy.

Techniques & Methods

Variation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing, a variation is also known as a query or utterance and works alongside intents to facilitate human-like conversation. A variation is a specific phrasing that someone might use to accomplish a certain action or intent. For example, if the intent is to \"pay by credit card,\" a variation could be, \"Can I use my card to pay?\""}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing, a variation is also known as a query or utterance and works alongside intents to facilitate human-like conversation. A variation is a specific phrasing that someone might use to accomplish a certain action or intent. For example, if the intent is to "pay by credit card," a variation could be, "Can I use my card to pay?"

Techniques & Methods

Vector Representation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a method for converting words into a numerical format that computers can understand and process. Algorithms for vector representation are employed by ChatGPT to comprehend the semantics of words and produce suitable replies."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a method for converting words into a numerical format that computers can understand and process. Algorithms for vector representation are employed by ChatGPT to comprehend the semantics of words and produce suitable replies.

W

Techniques & Methods

Word Embedding

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Similar to vector representation, word embedding is a technique for converting words into a format that computers can analyze. ChatGPT uses algorithms for word embedding to understand the semantics of text and provide fitting responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Similar to vector representation, word embedding is a technique for converting words into a format that computers can analyze. ChatGPT uses algorithms for word embedding to understand the semantics of text and provide fitting responses.

Z

Techniques & Methods

Zero-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This denotes an AI model's capacity to comprehend and respond appropriately to tasks it was not explicitly trained on."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This denotes an AI model's capacity to comprehend and respond appropriately to tasks it was not explicitly trained on.