Prompt engineering

753: 742: 731: 56:?", a command such as "write a poem about leaves falling", or a longer statement including context, instructions, and conversation history. Prompt engineering may involve phrasing a query, specifying a style, providing relevant context or assigning a role to the AI such as "Act as a native French speaker". A prompt may include a few examples for a model to learn from, such as asking the model to complete "maison → house, chat → cat, chien →" (the expected response being 5207: 4295: 541:

retrieval, the LLM generates an output that incorporates information from both the query and the retrieved documents. This method is particularly beneficial for handling proprietary or dynamic information that was not included in the initial training or fine-tuning phases of the model. RAG is also notable for its use of "few-shot" learning, where the model uses a small number of examples, often automatically retrieved from a database, to inform its outputs.

5187: 3988: 510: 550: 456: 815:(TTV) generation is an emerging technology enabling the creation of videos directly from textual descriptions. This field holds potential for transforming video production, animation, and storytelling. By utilizing the power of artificial intelligence, TTV allows users to bypass traditional video editing tools and translate their ideas into moving images. 3040:

Madaan, Aman; Tandon, Niket; Gupta, Prakhar; Hallinan, Skyler; Gao, Luyu; Wiegreffe, Sarah; Alon, Uri; Dziri, Nouha; Prabhumoye, Shrimai; Yang, Yiming; Gupta, Shashank; Prasad Majumder, Bodhisattwa; Hermann, Katherine; Welleck, Sean; Yazdanbakhsh, Amir (2023-03-01). "Self-Refine: Iterative Refinement

1893:

model (such as an LLM) which was trained to follow human-given instructions to follow instructions provided by a malicious user. This stands in contrast to the intended operation of instruction-following systems, wherein the ML model is intended only to follow trusted instructions (prompts) provided

540:

In response to a query, a document retriever selects the most relevant documents. This relevance is typically determined by first encoding both the query and the documents into vectors, then identifying documents whose vectors are closest in Euclidean distance to the query vector. Following document

436:

But if one cannot access such scores (such as when one is accessing the model through a restrictive API), uncertainty can still be estimated and incorporated into the model output. One simple method is to prompt the model to use words to estimate uncertainty. Another is to prompt the model to refuse

205:

For example, given the question "Q: The cafeteria had 23 apples. If they used 20 to make lunch and bought 6 more, how many apples do they have?", a CoT prompt might induce the LLM to answer "A: The cafeteria had 23 apples originally. They used 20 to make lunch. So they had 23 - 20 = 3. They bought 6

2498:

Bach, Stephen H.; Sanh, Victor; Yong, Zheng-Xin; Webson, Albert; Raffel, Colin; Nayak, Nihal V.; Sharma, Abheesht; Kim, Taewoon; M Saiful Bari; Fevry, Thibault; Alyafeai, Zaid; Dey, Manan; Santilli, Andrea; Sun, Zhiqing; Ben-David, Srulik; Xu, Canwen; Chhablani, Gunjan; Wang, Han; Jason Alan Fries;

697:

documentation encourages short, descriptive prompts: instead of "Show me a picture of lots of blooming California poppies, make them bright, vibrant orange, and draw them in an illustrated style with colored pencils", an effective prompt might be "Bright orange California poppies drawn with colored

2796:

Chung, Hyung Won; Hou, Le; Longpre, Shayne; Zoph, Barret; Tay, Yi; Fedus, William; Li, Yunxuan; Wang, Xuezhi; Dehghani, Mostafa; Brahma, Siddhartha; Webson, Albert; Gu, Shixiang Shane; Dai, Zhuyun; Suzgun, Mirac; Chen, Xinyun; Chowdhery, Aakanksha; Castro-Ros, Alex; Pellat, Marie; Robinson, Kevin;

340:

Self-refine prompts the LLM to solve the problem, then prompts the LLM to critique its solution, then prompts the LLM to solve the problem again in view of the problem, solution, and critique. This process is repeated until stopped, either by running out of tokens, time, or by the LLM outputting a

2291:

Wei, Jason; Tay, Yi; Bommasani, Rishi; Raffel, Colin; Zoph, Barret; Borgeaud, Sebastian; Yogatama, Dani; Bosma, Maarten; Zhou, Denny; Metzler, Donald; Chi, Ed H.; Hashimoto, Tatsunori; Vinyals, Oriol; Liang, Percy; Dean, Jeff; Fedus, William (31 August 2022). "Emergent Abilities of Large Language

2252:

276:

Input: There are a set of bricks. The yellow brick C is on top of the brick E. The yellow brick D is on top of the brick A. The yellow brick E is on top of the brick D. The white brick A is on top of the brick B. For the brick B, the color is white. Now we have to get a specific brick. The bricks

557:

GraphRAG, coined by Microsoft Research, extends RAG such that instead of relying solely on vector similarity (as in most RAG approaches), GraphRAG uses the LLM-generated knowledge graph. This graph allows the model to connect disparate pieces of information, synthesize insights, and holistically

71:

or a text-to-audio model, a typical prompt is a description of a desired output such as "a high-quality photo of an astronaut riding a horse" or "Lo-fi slow BPM electro chill with organic samples". Prompting a text-to-image model may involve adding, removing, emphasizing and re-ordering words to

383:

prompting is similar to tree-of-thought. The model is prompted to answer a question with an explanation. The model is then prompted to explain parts of the explanation, and so on. Inconsistent explanation trees are pruned or discarded. This improves performance on complex commonsense reasoning.

2265:

In prompting, a pre-trained language model is given a prompt (e.g. a natural language instruction) of a task and completes the response without any further training or gradient updates to its parameters... The ability to perform a task via few-shot prompting is emergent when a model has random

561:

Researchers have demonstrated GraphRAG's effectiveness using datasets like the Violent Incident Information from News Articles (VIINA). By combining LLM-generated knowledge graphs with graph machine learning, GraphRAG substantially improves both the comprehensiveness and diversity of generated

269:

Chain-of-Symbol prompting in conjunction with CoT prompting assists LLMs with its difficulty of spatial reasoning in text. In other words, using arbitrary symbols such as ' / ' assist the LLM to interpret spacing in text. This assists in reasoning and increases the performance of the LLM.

616:. Questions nearest to the centroids of each cluster are selected. An LLM does zero-shot CoT on each question. The resulting CoT examples are added to the dataset. When prompted with a new question, CoT examples to the nearest questions can be retrieved and added to the prompt. 2797:

Valter, Dasha; Narang, Sharan; Mishra, Gaurav; Yu, Adams; Zhao, Vincent; Huang, Yanping; Dai, Andrew; Yu, Hongkun; Petrov, Slav; Chi, Ed H.; Dean, Jeff; Devlin, Jacob; Roberts, Adam; Zhou, Denny; Le, Quoc V.; Wei, Jason (2022). "Scaling Instruction-Finetuned Language Models".

140:

tasks (using 62 datasets, as each task can have multiple datasets). The model showed good performance on new tasks, surpassing models trained directly on just performing one task (without pretraining). To solve a task, T0 is given the task in a structured prompt, for example

132:

could be cast as a question answering problem over a context. In addition, they trained a first single, joint, multi-task model that would answer any task-related question like "What is the sentiment" or "Translate this sentence to German" or "Who is the president?"

2499:

Al-shaibani, Maged S.; Sharma, Shanya; Thakker, Urmish; Almubarak, Khalid; Tang, Xiangru; Radev, Dragomir; Mike Tian-Jian Jiang; Rush, Alexander M. (2022). "PromptSource: An Integrated Development Environment and Repository for Natural Language Prompts".

2948:

Zhou, Denny; Schärli, Nathanael; Hou, Le; Wei, Jason; Scales, Nathan; Wang, Xuezhi; Schuurmans, Dale; Cui, Claire; Bousquet, Olivier; Le, Quoc; Chi, Ed (2022-05-01). "Least-to-Most Prompting Enables Complex Reasoning in Large Language Models".

186:(LLMs) to solve a problem as a series of intermediate steps before giving a final answer. Chain-of-thought prompting improves reasoning ability by inducing the model to answer a multi-step problem with steps of reasoning that mimic a 2370:

By the time you type a query into ChatGPT, the network should be fixed; unlike humans, it should not continue to learn. So it came as a surprise that LLMs do, in fact, learn from their users' prompts—an ability known as in-context

107:

for each specific task, which are not temporary, what has been learnt during in-context learning is of a temporary nature. It does not carry the temporary contexts or biases, except the ones already present in the (pre)training

3755:

Kirillov, Alexander; Mintun, Eric; Ravi, Nikhila; Mao, Hanzi; Rolland, Chloe; Gustafson, Laura; Xiao, Tete; Whitehead, Spencer; Berg, Alexander C.; Lo, Wan-Yen; Dollár, Piotr; Girshick, Ross (2023-04-01). "Segment Anything".

528:

and answer formulation by a Large Language Model (LLM). The initial phase utilizes dense embeddings to retrieve documents. This retrieval can be based on a variety of database formats depending on the use case, such as a

3179: 3431:

Guo, Qingyan; Wang, Rui; Guo, Junliang; Li, Bei; Song, Kaitao; Tan, Xu; Liu, Guoqing; Bian, Jiang; Yang, Yujiu (2023). "Connecting Large Language Models with Evolutionary Algorithms Yields Powerful Prompt Optimizers".

1720:

An earlier result uses the same idea of gradient descent search, but is designed for masked language models like BERT, and searches only over token sequences, rather than numerical vectors. Formally, it searches for

1083: 1019: 955: 1132: 323:

performs several chain-of-thought rollouts, then selects the most commonly reached conclusion out of all the rollouts. If the rollouts disagree by a lot, a human can be queried for the correct chain of thought.

3730:

Gal, Rinon; Alaluf, Yuval; Atzmon, Yuval; Patashnik, Or; Bermano, Amit H.; Chechik, Gal; Cohen-Or, Daniel (2022). "An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion".

288:

first prompts the model to generate relevant facts for completing the prompt, then proceed to complete the prompt. The completion quality is usually higher, as the model can be conditioned on relevant facts.

3230:

Lewis, Patrick; Perez, Ethan; Piktus, Aleksandra; Petroni, Fabio; Karpukhin, Vladimir; Goyal, Naman; Küttler, Heinrich; Lewis, Mike; Yih, Wen-tau; Rocktäschel, Tim; Riedel, Sebastian; Kiela, Douwe (2020).

3843:

In this work, we explore "prompt tuning," a simple yet effective mechanism for learning "soft prompts"...Unlike the discrete text prompts used by GPT-3, soft prompts are learned through back-propagation

2973:

Wang, Xuezhi; Wei, Jason; Schuurmans, Dale; Le, Quoc; Chi, Ed; Narang, Sharan; Chowdhery, Aakanksha; Zhou, Denny (2022-03-01). "Self-Consistency Improves Chain of Thought Reasoning in Language Models".

569:

for question answering using text-to-query generation. These techniques can be combined to perform search across both unstructured and structured data, providing expanded context and improved ranking.

277:

must now be grabbed from top to bottom, and if the lower brick is to be grabbed, the upper brick must be removed first. How to get brick D? B/A/D/E/C C/E E/D D Output: So we get the result as C, E, D.

3107:

Jung, Jaehun; Qin, Lianhui; Welleck, Sean; Brahman, Faeze; Bhagavatula, Chandra; Le Bras, Ronan; Choi, Yejin (2022). "Maieutic Prompting: Logically Consistent Reasoning with Recursive Explanations".

598:

Each of the generated instructions is used to prompt the target LLM, followed by each of the inputs. The log-probabilities of the outputs are computed and added. This is the score of the instruction.

1827: 1460: 3083:

Yao, Shunyu; Yu, Dian; Zhao, Jeffrey; Shafran, Izhak; Griffiths, Thomas L.; Cao, Yuan; Narasimhan, Karthik (2023-05-17). "Tree of Thoughts: Deliberate Problem Solving with Large Language Models".

2383:

Johannes von Oswald; Niklasson, Eyvind; Randazzo, Ettore; Sacramento, João; Mordvintsev, Alexander; Zhmoginov, Andrey; Vladymyrov, Max (2022). "Transformers learn in-context by gradient descent".

595:

Prompting LLM is presented with example input-output pairs, and asked to generate instructions that could have caused a model following the instructions to generate the outputs, given the inputs.

1164:

to prompt-specific parameters: in prefix-tuning, they are parameters associated with the prompt tokens at each layer; in prompt tuning, they are merely the soft tokens added to the vocabulary.

3779:

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)

3143:

The directional stimulus serves as hints or cues for each input query to guide LLMs toward the desired output, such as keywords that the desired summary should include for summarization.

306:

prompts a model to first list the sub-problems to a problem, then solve them in sequence, such that later sub-problems can be solved with the help of answers to previous sub-problems.

421:

By default, the output of language models may not contain estimates of uncertainty. The model may output text that appears confident, though the underlying token predictions have low

332:

Complexity-based prompting performs several CoT rollouts, then select the rollouts with the longest chains of thought, then select the most commonly reached conclusion out of those.

1343: 3474:

Zhou, Yongchao; Ioan Muresanu, Andrei; Han, Ziwen; Paster, Keiran; Pitis, Silviu; Chan, Harris; Ba, Jimmy (2022-11-01). "Large Language Models Are Human-Level Prompt Engineers".

3743:

Using only 3-5 images of a user-provided concept, like an object or a style, we learn to represent it through new "words" in the embedding space of a frozen text-to-image model.

3194:

If you explicitly indicate in your prompt that you want the generative AI to emit a certainty or uncertainty qualification then you will almost certainly get such an indication.

261:

Chain-of-thought prompting is just one of many prompt-engineering techniques. Various other techniques have been proposed. At least 29 distinct techniques have been published.

3347:

Fernando, Chrisantha; Banarse, Dylan; Michalewski, Henryk; Osindero, Simon; Rocktäschel, Tim (2023). "Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution".

4027: 1224: 30:

This article is about means of interacting (command prompts) with an artificial intelligence system. For general computer command line interfaces and commando entries, see

1856: 1715: 1663: 1158: 5081: 1085:

be the token embeddings of the input and output respectively. During training, the tunable embeddings, input, and output tokens are concatenated into a single sequence

1634: 2147:

Brown, Tom; Mann, Benjamin; Ryder, Nick; Subbiah, Melanie; Kaplan, Jared D.; Dhariwal, Prafulla; Neelakantan, Arvind (2020). "Language models are few-shot learners".

1544: 1598: 1571: 364:

generalizes chain-of-thought by prompting the model to generate one or more "possible next steps", and then running the model on each of the possible next steps by

858:

based on a set of example images. This embedding vector acts as a "pseudo-word" which can be included in a prompt to express the content or style of the examples.

433:

likelihood scores in their token predictions, and so the model output uncertainty can be directly estimated by reading out the token prediction likelihood scores.

3389:

Pryzant, Reid; Iter, Dan; Li, Jerry; Lee, Yin Tat; Zhu, Chenguang; Zeng, Michael (2023). "Automatic Prompt Optimization with "Gradient Descent" and Beam Search".

2961:...least-to-most prompting. The key idea in this strategy is to break down a complex problem into a series of simpler subproblems and then solve them in sequence. 752: 741: 413:

Article: {article} Q: Write a short summary of the article in 2-4 sentences that accurately incorporates the provided keywords. Keywords: {keywords} A:

1683: 1284: 1264: 1244: 3131:

Li, Zekun; Peng, Baolin; He, Pengcheng; Galley, Michel; Gao, Jianfeng; Yan, Xifeng (2023). "Guiding Large Language Models via Directional Stimulus Prompting".

792:

Text-to-image models do not natively understand negation. The prompt "a party with no cake" is likely to produce an image including a cake. As an alternative,

3588: 2432:

Garg, Shivam; Tsipras, Dimitris; Liang, Percy; Valiant, Gregory (2022). "What Can Transformers Learn In-Context? A Case Study of Simple Function Classes".

2126:

Garg, Shivam; Tsipras, Dimitris; Liang, Percy; Valiant, Gregory (2022). "What Can Transformers Learn In-Context? A Case Study of Simple Function Classes".

4329: 2325: 2456:

McCann, Bryan; Shirish, Nitish; Xiong, Caiming; Socher, Richard (2018). "The Natural Language Decathlon: Multitask Learning as Question Answering".

4923: 878:

by prompting. As an alternative to text prompts, Segment Anything can accept bounding boxes, segmentation masks, and foreground/background points.

714: 2628: 2599: 1024: 960: 896: 2324:

Wei, Jason; Wang, Xuezhi; Schuurmans, Dale; Bosma, Maarten; Ichter, Brian; Xia, Fei; Chi, Ed H.; Le, Quoc V.; Zhou, Denny (31 October 2022).

1088: 3306:

A Benchmark to Understand the Role of Knowledge Graphs on Large Language Model's Accuracy for Question Answering on Enterprise SQL Databases

3264: 2444:

Training a model to perform in-context learning can be viewed as an instance of the more general learning-to-learn or meta-learning paradigm

2177: 2749:

Kojima, Takeshi; Shixiang Shane Gu; Reid, Machel; Matsuo, Yutaka; Iwasawa, Yusuke (2022). "Large Language Models are Zero-Shot Reasoners".

534: 5271: 4298: 3517: 347:

I have some code. Give one suggestion to improve readability. Don't fix the code, just give a suggestion. Code: {code} Suggestion:

2995:

Diao, Shizhe; Wang, Pengcheng; Lin, Yong; Zhang, Tong (2023-02-01). "Active Prompting with Chain-of-Thought for Large Language Models".

2080: 1989:

We demonstrate language models can perform down-stream tasks in a zero-shot setting – without any parameter or architecture modification

1981: 625: 4097: 3016:

Fu, Yao; Peng, Hao; Sabharwal, Ashish; Clark, Peter; Khot, Tushar (2022-10-01). "Complexity-Based Prompting for Multi-Step Reasoning".

2048:"Basic prompt: 'Write a poem about leaves falling.' Better prompt: 'Write a poem in the style of Edgar Allan Poe about leaves falling.' 3180:"Latest Prompt Engineering Technique Aims To Get Certainty And Uncertainty Of Generative AI Directly On The Table And Out In The Open" 2771: 2214:

Next, I gave a more complicated prompt to attempt to throw MusicGen for a loop: "Lo-fi slow BPM electro chill with organic samples."

4107: 3939: 3799:

In this paper, we propose prefix-tuning, a lightweight alternative to fine-tuning... Prefix-tuning draws inspiration from prompting

3495:

Zhang, Zhuosheng; Zhang, Aston; Li, Mu; Smola, Alex (2022-10-01). "Automatic Chain of Thought Prompting in Large Language Models".

2721: 608:

CoT examples can be generated by LLM themselves. In "auto-CoT", a library of questions are converted to vectors by a model such as

213:

prompting technique. However, simply appending the words "Let's think step-by-step", has also proven effective, which makes CoT a

4439: 4034: 2898:

Liu, Jiacheng; Liu, Alisa; Lu, Ximing; Welleck, Sean; West, Peter; Le Bras, Ronan; Choi, Yejin; Hajishirzi, Hannaneh (May 2022).

2615:'Chain-of-thought prompting allows us to describe multistep problems as a series of intermediate steps,' Google CEO Sundar Pichai 93: 85: 3232: 1928: 4322: 701:

Word order affects the output of a text-to-image prompt. Words closer to the start of a prompt may be emphasized more heavily.

217:

prompting technique. This allows for better scaling as a user no longer needs to formulate many specific CoT Q&A examples.

100:

in downstream scaling laws occur such that its efficacy increases at a different rate in larger models than in smaller models.

713:

has been used in Stable Diffusion and Midjourney prompts to generate images in the distinctive style of Polish digital artist

1724: 1348: 3855:

Sun, Simeng; Liu, Yang; Iter, Dan; Zhu, Chenguang; Iyyer, Mohit (2023). "How Does In-Context Learning Help Prompt Tuning?".

3280:

Edge, Darren; Trinh, Ha; Cheng, Newman; Bradley, Joshua; Chao, Alex; Mody, Apurva; Truitt, Steven; Larson, Jonathan (2024),

5251: 5112: 730: 152:

A repository for prompts reported that over 2,000 public prompts for around 170 datasets were available in February 2022.

5213: 4764: 4501: 2395:

Thus we show how trained Transformers become mesa-optimizers i.e. learn models by gradient descent in their forward pass

4072: 4043: 243:. It is possible to fine-tune models on CoT reasoning datasets to enhance this capability further and stimulate better 17: 2420:

Mesa-Optimization is the situation that occurs when a learned model (such as a neural network) is itself an optimizer.

5025: 4652: 4459: 4315: 2202: 496: 709:

Some text-to-image models are capable of imitating the style of particular artists by name. For example, the phrase

4980: 3707:"Understanding OpenAI's Sora: A Revolutionary Leap | PromptSora: Discover Prompts and Videos for Sora from Open AI" 2669: 117: 3811:

Lester, Brian; Al-Rfou, Rami; Constant, Noah (2021). "The Power of Scale for Parameter-Efficient Prompt Tuning".

3206: 2654: 2352: 4003: 2819: 2695: 2566: 886:

In "prefix-tuning", "prompt tuning" or "soft prompting", floating-point-valued vectors are searched directly by

5256: 5167: 5107: 4705: 4092: 4067: 2904:

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

2521: 521: 229: 113: 1959: 4700: 4389: 3706: 2061: 1903: 2844:

Sahoo, Pranab; Singh, Ayush Kumar; Saha, Sriparna; Jain, Vinija; Mondal, Samrat; Chadha, Aman (2024-02-05),

84:, defined as a model's ability to temporarily learn from prompts. The ability for in-context learning is an 5142: 4496: 4449: 4444: 2547: 1289: 650:

images. Text-to-image models typically do not understand grammar and sentence structure in the same way as

3681: 471:. In particular, it dives into the technical vector implementation before positioning the overall concept. 5193: 4489: 4415: 137: 129: 104: 353:

Code: {code} Let's use this suggestion to improve the code. Suggestion: {suggestion} New Code:

4817: 4752: 4353: 4216: 2477:

Sanh, Victor; et al. (2021). "Multitask Prompted Training Enables Zero-Shot Task Generalization".

834: 553:

GraphRAG with a knowledge graph combining access patterns for unstructured, structured and mixed data.

404:

includes a hint or cue, such as desired keywords, to guide a language model toward the desired output.

53: 5246: 5218: 5076: 4715: 4546: 4369: 1886: 97: 3777:

Li, Xiang Lisa; Liang, Percy (2021). "Prefix-Tuning: Optimizing Continuous Prompts for Generation".

3460: 3417: 3375: 1170: 228:, CoT prompting significantly aided the model, allowing it to perform comparably with task-specific 5261: 5117: 4374: 4102: 3876:

Shin, Taylor; Razeghi, Yasaman; Logan IV, Robert L.; Wallace, Eric; Singh, Sameer (November 2020).

1942:

Prompt engineering is the process of structuring words that can be interpreted and understood by a

474: 430: 3613: 1832: 1691: 1639: 1141: 190:. It allows large language models to overcome difficulties with some reasoning tasks that require 5162: 5147: 4800: 4795: 4695: 4563: 4344: 4150: 4145: 4112: 577:

Large language models (LLM) themselves can be used to compose prompts for large language models.

478: 2655:"Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance" 5241: 5122: 4882: 4601: 4596: 4170: 4020: 2846:

A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications

1946:

model. Think of it as the language you need to speak in order to tell an AI model what to draw.

854:

For text-to-image models, "Textual inversion" performs an optimization process to create a new

800:

appear in the resulting image. A common approach is to include generic undesired terms such as

237: 31: 3882:

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

3328:

Singh, Chandan; Morris, John; Aneja, Jyoti; Rush, Alexander; Gao, Jianfeng (October 4, 2022).

1980:

Radford, Alec; Wu, Jeffrey; Child, Rewon; Luan, David; Amodei, Dario; Sutskever, Ilya (2019).

1603: 646:

were released to the public. These models take text prompts as input and use them to generate

5152: 5137: 5102: 4790: 4690: 4558: 3447: 3404: 3362: 2105: 1465: 651: 609: 240: 199: 183: 5020: 3964: 604:

Repeat until some stopping criteria is reached, then output the highest-scored instructions.

295:

Generate some knowledge about the concepts in the input. Input: {question} Knowledge:

5266: 5172: 5127: 4573: 4518: 4364: 4359: 4077: 3878:"AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts" 1877: 1576: 1549: 365: 89: 49: 2871:

Hu, Hanxu; Lu, Hongyuan; Zhang, Huajian; Song, Yun-Ze; Lam, Wai; Zhang, Yue (2023-10-03),

662:

A text-to-image prompt commonly includes a description of the subject of the art (such as

8: 4747: 4725: 4474: 4469: 4427: 4379: 4208: 4180: 3914: 2357: 2266:

performance until a certain scale, after which performance increases to well-above random

812: 466: 422: 166:

In 2023 several text-to-text and text-to-image prompt databases were publicly available.

146: 40:

is the process of structuring an instruction that can be interpreted and understood by a

3993: 3657: 209:

As originally proposed, each CoT prompt included a few Q&A examples. This made it a

136:

In 2021, researchers fine-tuned one generatively pretrained model (T0) on performing 12

5132: 4710: 4117: 3895: 3856: 3834: 3816: 3790: 3757: 3732: 3496: 3475: 3433: 3390: 3348: 3309: 3285: 3244: 3156: 3132: 3108: 3084: 3063: 3042: 3017: 2996: 2975: 2950: 2927: 2907: 2876: 2849: 2798: 2750: 2500: 2478: 2457: 2433: 2407: 2384: 2331: 2293: 2254: 2156: 2127: 1668: 1269: 1249: 1229: 875: 525: 3890: 3830: 3813:

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing

846:

Some approaches augment or replace natural language text prompts with non-text input.

601:

The highest-scored instructions are given to the prompting LLM for further variations.

5198: 5186: 4990: 4642: 4513: 4506: 3899: 3838: 3794: 2931: 2574: 191: 3786: 2922: 2178:"This horse-riding astronaut is a milestone on AI's long road towards understanding" 1286:

is the rest of the model. In prefix-tuning, one provide a set of input-output pairs

4943: 4933: 4740: 4534: 4484: 4479: 4422: 4410: 4198: 3885: 3826: 3782: 2917: 1890: 1867: 887: 822: 762: 639: 613: 244: 233: 187: 45: 3330:"Explaining Patterns in Data with Language Models via Interpretable Autoprompting" 2330:. Advances in Neural Information Processing Systems (NeurIPS 2022). Vol. 35. 5056: 5000: 4822: 4464: 4384: 4262: 3566: 1161: 871: 566: 530: 380: 112:, from one conversation to the other. This result of "mesa-optimization" within 48:

text describing the task that an AI should perform: a prompt for a text-to-text

5030: 4995: 4985: 4810: 4568: 4394: 4122: 4082: 3877: 867: 855: 514: 225: 3614:"Stable Diffusion 2.0 and the Importance of Negative Prompts for Good Results" 2382: 2226: 143:

If {{premise}} is true, is it also true that {{hypothesis}}? ||| {{entailed}}.

5235: 4975: 4955: 4872: 4551: 4062: 2899: 2578: 2278: 2081:"How to Prime and Prompt ChatGPT for More Reliable Contract Drafting Support" 1873: 1135: 631: 68: 41: 592:

There are two LLMs. One is the target LLM, and another is the prompting LLM.

5061: 4892: 4307: 4267: 3546: 1968:

Prompt engineering is the art of communicating with a generative AI model.

437:

to answer in a standardized way if the input does not satisfy conditions.

128:

In 2018, researchers first proposed that all previously separate tasks in

5157: 4928: 4837: 4832: 4454: 4432: 4057: 3884:. Online: Association for Computational Linguistics. pp. 4222–4235. 3282:

From Local to Global: A Graph RAG Approach to Query-Focused Summarization

3062:

Long, Jieyi (2023-05-15). "Large Language Model Guided Tree-of-Thought".

2906:. Dublin, Ireland: Association for Computational Linguistics: 3154–3169. 1078:{\displaystyle \mathbf {Y} =\{\mathbf {y_{1}} ,\dots ,\mathbf {y_{n}} \}} 1014:{\displaystyle \mathbf {X} =\{\mathbf {x_{1}} ,\dots ,\mathbf {x_{m}} \}} 950:{\displaystyle \mathbf {E} =\{\mathbf {e_{1}} ,\dots ,\mathbf {e_{k}} \}} 761:

Demonstration of the effect of negative prompts on images generated with

585: 369: 3658:"Introducing Make-A-Video: An AI system that generates videos from text" 3589:"This Artist Is Dominating AI-Generated Art and He's Not Happy About It" 837:– As yet unreleased, Sora purportedly can produce high-resolution videos 5051: 5010: 5005: 4918: 4827: 4735: 4647: 4627: 4282: 4193: 2696:"Google's Chain of Thought Prompting Can Boost Today's Best Algorithms" 2227:"How to Write AI Photoshoot Prompts: A Guide for Better Product Photos" 1127:{\displaystyle {\text{concat}}(\mathbf {E} ;\mathbf {X} ;\mathbf {Y} )} 694: 643: 312:

Input: Q: {question} A: Let's break down this problem: 1.

195: 2277:

Caballero, Ethan; Gupta, Kshitij; Rish, Irina; Krueger, David (2022).

2033: 5046: 5015: 4913: 4757: 4720: 4657: 4611: 4606: 4591: 4257: 2722:"Amazon's Alexa scientists demonstrate bigger AI isn't always better" 2652: 2327:

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

831:

Make-a-Video – Focuses on creating detailed and diverse video outputs

558:

understand summarized semantic concepts over large data collections.

2281:. International Conference on Learning Representations (ICLR), 2023. 825:– Offers a user-friendly interface and supports various video styles 4948: 4780: 3940:"What's Old Is New Again: GPT-3 Prompt Injection Attack Affects AI" 3861: 3821: 3762: 3737: 3633: 3501: 3480: 3438: 3395: 3353: 3329: 3314: 3290: 3249: 3161: 3137: 3113: 3089: 3068: 3047: 3022: 3001: 2980: 2955: 2912: 2881: 2873:

Chain-of-Symbol Prompting Elicits Planning in Large Language Models

2854: 2803: 2772:"LLMs have not learned our language — we're trying to learn theirs" 2755: 2505: 2483: 2462: 2438: 2389: 2336: 2298: 2259: 2161: 2132: 2062:"The ultimate guide to prompt engineering your GPT-3.5-Turbo model" 796:

allow a user to indicate, in a separate prompt, which terms should

635: 109: 72:

achieve a desired subject, style, layout, lighting, and aesthetic.

3346: 3233:"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks" 2001: 5071: 4908: 4862: 4785: 4685: 4680: 4632: 4277: 4232: 2748: 1717:

is preappended to the hidden states in every layer of the model.

626:

Artificial intelligence art § Prompt engineering and sharing

4012: 5086: 5066: 4938: 4730: 4272: 4188: 3183: 1167:

More formally, this is prompt tuning. Let an LLM be written as

881: 647: 160: 2629:"Google's Latest AI Model Can Be Taught How to Solve Problems" 2600:"Google's Latest AI Model Can Be Taught How to Solve Problems" 4887: 4867: 4857: 4852: 4847: 4842: 4805: 4637: 4165: 4160: 4155: 3518:"Dall-E2 VS Stable Diffusion: Same Prompt, Different Results" 509: 426: 1973: 572: 4877: 4237: 3266:

GraphRAG: Unlocking LLM discovery on narrative private data

1927:

Diab, Mohamad; Herrera, Julian; Chernow, Bob (2022-10-28).

957:

be a set of soft prompt tokens (tunable embeddings), while

549: 221: 3875: 3473: 1688:

For prefix tuning, it is similar, but the "prefix vector"

3039: 2900:"Generated Knowledge Prompting for Commonsense Reasoning" 3965:"GPT-3 'prompt injection' attack causes bot bad manners" 3729: 2893: 2891: 2522:"Language Models Perform Reasoning via Chain of Thought" 2455: 3229: 2431: 2125: 1822:{\displaystyle \arg \max _{\tilde {X}}\sum _{i}\log Pr} 1455:{\displaystyle \arg \max _{\tilde {Z}}\sum _{i}\log Pr} 828:

Lumiere – Designed for high-resolution video generation

654:, and require a different set of prompting techniques. 544: 206:

more apples, so they have 3 + 6 = 9. The answer is 9."

2670:"Harnessing the power of GPT-3 in scientific research" 2323: 2245: 1858:

is ranges over token sequences of a specified length.

3810: 3754: 3279: 3106: 2888: 2567:"Get the Best From ChatGPT With These Golden Prompts" 2290: 2251: 2146: 1982:"Language Models are Unsupervised Multitask Learners" 1835: 1727: 1694: 1671: 1642: 1606: 1579: 1552: 1468: 1351: 1292: 1272: 1252: 1232: 1173: 1144: 1091: 1027: 963: 899: 517:

and Large Language Model (LLM) for answer formulation

3327: 3304:

Sequeda, Juan; Allemang, Dean; Jacob, Bryon (2023),

3303: 3015: 2653:

Sharan Narang and Aakanksha Chowdhery (2022-04-04).

2476: 1960:"A developer's guide to prompt engineering and LLMs" 513:

Two-phase process of document retrieval using dense

416: 3035: 3033: 2972: 2795: 2744: 2742: 1636:, then prepend the vector with the "prefix vector" 849: 2843: 2497: 1979: 1926: 1850: 1821: 1709: 1677: 1657: 1628: 1592: 1565: 1538: 1454: 1337: 1278: 1258: 1238: 1218: 1152: 1134:, and fed to the large language models (LLM). The 1126: 1077: 1013: 949: 3494: 3237:Advances in Neural Information Processing Systems 3055: 2947: 2203:"Meta open sources an AI-powered music generator" 2149:Advances in Neural Information Processing Systems 2034:"How to write an effective GPT-3 or GPT-4 prompt" 565:Earlier work showed the effectiveness of using a 445: 396: 5233: 3388: 3082: 3030: 2994: 2943: 2941: 2820:"Better Language Models Without Massive Compute" 2739: 2027: 2025: 1958:Ziegler, Albert; Berryman, John (17 July 2023). 1735: 1359: 280: 3207:"How Each Index Works - LlamaIndex 🦙 v0.10.17" 3155:OpenAI (2023-03-27). "GPT-4 Technical Report". 3102: 3100: 2897: 1957: 3130: 1345:, and then use gradient descent to search for 440: 253:Q: {question} A: Let's think step by step. 4323: 4028: 3854: 3682:"Video generation models as world simulators" 3567:"Stable Diffusion prompt: a definitive guide" 2938: 2870: 2078: 2059: 2022: 890:, to maximize the log-likelihood on outputs. 327: 4337: 3430: 3097: 2548:"How to Turn Your Chatbot Into a Life Coach" 1326: 1293: 1072: 1036: 1008: 972: 944: 908: 882:Using gradient descent to search for prompts 870:'s AI research released Segment Anything, a 477:. There might be a discussion about this on 315: 3126: 3124: 2319: 2317: 2315: 2313: 2311: 2309: 182:(CoT) prompting is a technique that allows 4330: 4316: 4035: 4021: 3962: 3541: 3539: 3537: 562:answers for global sensemaking questions. 298: 27:Structuring text as input to generative AI 3963:Vigliarolo, Brandon (19 September 2022). 3889: 3860: 3820: 3761: 3736: 3611: 3500: 3479: 3437: 3394: 3352: 3313: 3289: 3248: 3160: 3136: 3112: 3088: 3067: 3046: 3021: 3000: 2979: 2954: 2921: 2911: 2880: 2853: 2802: 2754: 2504: 2482: 2461: 2437: 2388: 2335: 2297: 2258: 2160: 2131: 573:Using language models to generate prompts 497:Learn how and when to remove this message 4539: 3915:"Prompt injection attacks against GPT-3" 3912: 3748: 3586: 3121: 2818:Wei, Jason; Tay, Yi (29 November 2022). 2763: 2306: 2060:Gouws-Stewart, Natasha (June 16, 2023). 2031: 548: 508: 3776: 3534: 2769: 2519: 2200: 1573:, if the model first encodes the input 524:(RAG) is a two-phase process involving 372:, or some other method of tree search. 145:is the prompt used for making T0 solve 14: 5234: 3243:. Curran Associates, Inc.: 9459–9474. 3154: 2817: 2693: 2176:Heaven, Will Douglas (April 6, 2022). 2175: 1922: 1920: 1918: 75: 4311: 4016: 3913:Willison, Simon (12 September 2022). 3515: 3177: 2866: 2864: 2626: 2597: 1338:{\displaystyle \{(X^{i},Y^{i})\}_{i}} 1266:is the token-to-vector function, and 804:in the negative prompt for an image. 375: 194:and multiple steps to solve, such as 5168:Generative adversarial network (GAN) 3937: 3061: 2667: 2564: 2545: 2353:"How AI Knows Things No One Told It" 2079:Greenberg, J., Laura (31 May 2023). 1546:is the log-likelihood of outputting 1246:is a sequence of linguistic tokens, 545:Graph retrieval-augmented generation 449: 159:prompting technique was proposed by 2017:what is the fermat's little theorem 1915: 1861: 841: 720: 425:scores. Large language models like 393:Q: {question} A: False, because 256: 232:models on several tasks, achieving 174: 92:. In-context learning itself is an 24: 5272:Generative artificial intelligence 4073:Generative pre-trained transformer 2861: 2350: 861: 390:Q: {question} A: True, because 356: 25: 5283: 4042: 2032:Robinson, Reid (August 3, 2023). 657: 417:Prompting to disclose uncertainty 236:results at the time on the GSM8K 80:Prompt engineering is enabled by 5206: 5205: 5185: 4294: 4293: 3986: 3704: 3587:Heikkilä, Melissa (2022-09-16). 2520:Wei, Jason; Zhou (11 May 2022). 1146: 1117: 1109: 1101: 1066: 1062: 1045: 1041: 1029: 1002: 998: 981: 977: 965: 938: 934: 917: 913: 901: 850:Textual inversion and embeddings 807: 751: 740: 729: 704: 619: 454: 410:Article: {article} Keywords: 94:emergent property of model scale 52:can be a query such as "what is 3956: 3931: 3906: 3891:10.18653/v1/2020.emnlp-main.346 3869: 3848: 3831:10.18653/V1/2021.EMNLP-MAIN.243 3804: 3770: 3723: 3698: 3674: 3650: 3626: 3605: 3580: 3559: 3516:Monge, Jim Clyde (2022-08-25). 3509: 3488: 3467: 3424: 3382: 3340: 3321: 3297: 3273: 3257: 3223: 3199: 3171: 3148: 3076: 3009: 2988: 2966: 2837: 2811: 2789: 2770:Dickson, Ben (30 August 2022). 2719: 2713: 2687: 2661: 2646: 2620: 2591: 2558: 2539: 2513: 2491: 2470: 2449: 2425: 2400: 2376: 2344: 2284: 2271: 2219: 2194: 2169: 2140: 666:), the desired medium (such as 265:Chain-of-Symbol (CoS) Prompting 169: 5118:Recurrent neural network (RNN) 5108:Differentiable neural computer 4068:Generative adversarial network 2668:Dang, Ekta (8 February 2023). 2119: 2098: 2072: 2053: 1994: 1951: 1929:"Stable Diffusion Prompt Book" 1842: 1816: 1797: 1787: 1773: 1744: 1701: 1649: 1623: 1610: 1533: 1530: 1517: 1505: 1495: 1481: 1449: 1446: 1433: 1421: 1411: 1397: 1368: 1322: 1296: 1219:{\displaystyle LLM(X)=F(E(X))} 1213: 1210: 1204: 1198: 1189: 1183: 1121: 1097: 711:in the style of Greg Rutkowski 588:over prompts for another LLM: 522:Retrieval-augmented generation 446:Retrieval-augmented generation 402:Directional-stimulus prompting 397:Directional-stimulus prompting 335: 13: 1: 5163:Variational autoencoder (VAE) 5123:Long short-term memory (LSTM) 4390:Computational learning theory 3787:10.18653/V1/2021.ACL-LONG.353 2923:10.18653/v1/2022.acl-long.225 2694:Montti, Roger (13 May 2022). 2565:Chen, Brian X. (2023-05-25). 2546:Chen, Brian X. (2023-06-23). 1909: 1904:Social engineering (security) 783:: "round stones, round rocks" 286:Generated knowledge prompting 281:Generated knowledge prompting 5143:Convolutional neural network 2279:"Broken Neural Scaling Laws" 2201:Wiggers, Kyle (2023-06-12). 1894:by the ML model's operator. 1851:{\displaystyle {\tilde {X}}} 1710:{\displaystyle {\tilde {Z}}} 1658:{\displaystyle {\tilde {Z}}} 1153:{\displaystyle \mathbf {Y} } 103:In contrast to training and 7: 5252:Natural language processing 5138:Multilayer perceptron (MLP) 3938:Papp, Donald (2022-09-17). 3634:"Lumiere - Google Research" 3178:Eliot, Lance (2023-08-18). 1897: 612:. The question vectors are 441:Automatic prompt generation 10: 5288: 5214:Artificial neural networks 5128:Gated recurrent unit (GRU) 4354:Differentiable programming 1887:computer security exploits 1871: 1865: 1160:tokens; the gradients are 623: 584:algorithm uses one LLM to 537:, or keyword table index. 328:Complexity-based prompting 123: 67:When communicating with a 29: 5181: 5095: 5039: 4968: 4901: 4773: 4673: 4666: 4620: 4584: 4547:Artificial neural network 4527: 4403: 4370:Automatic differentiation 4343: 4291: 4250: 4225: 4207: 4179: 4138: 4131: 4050: 3638:Lumiere - Google Research 1889:carried out by getting a 802:ugly, boring, bad anatomy 582:automatic prompt engineer 362:Tree-of-thought prompting 321:Self-consistency decoding 316:Self-consistency decoding 4375:Neuromorphic engineering 4338:Differentiable computing 4103:Self-supervised learning 3612:Max Woolf (2022-11-28). 1629:{\displaystyle E(X^{i})} 120:or "learning to learn". 5148:Residual neural network 4564:Artificial Intelligence 4113:Variational autoencoder 1885:is a family of related 1539:{\displaystyle \log Pr} 874:model that can perform 304:Least-to-most prompting 299:Least-to-most prompting 54:Fermat's little theorem 3815:. pp. 3045–3059. 3781:. pp. 4582–4597. 3455:Cite journal requires 3412:Cite journal requires 3370:Cite journal requires 1852: 1823: 1711: 1679: 1659: 1630: 1594: 1567: 1540: 1456: 1339: 1280: 1260: 1240: 1220: 1154: 1138:are computed over the 1128: 1079: 1015: 951: 690:), color and texture. 554: 518: 238:mathematical reasoning 60:), an approach called 32:Command-line interface 5257:Unsupervised learning 5103:Neural Turing machine 4691:Human image synthesis 3593:MIT Technology Review 3041:with Self-Feedback". 2700:Search Engine Journal 2182:MIT Technology Review 2002:"Introducing ChatGPT" 1853: 1824: 1712: 1680: 1660: 1631: 1595: 1593:{\displaystyle X^{i}} 1568: 1566:{\displaystyle Y^{i}} 1541: 1457: 1340: 1281: 1261: 1241: 1221: 1155: 1129: 1080: 1016: 952: 682:), lighting (such as 664:bright orange poppies 652:large language models 552: 512: 200:commonsense reasoning 184:large language models 116:layers, is a form of 90:large language models 5194:Computer programming 5173:Graph neural network 4748:Text-to-video models 4726:Text-to-image models 4574:Large language model 4559:Scientific computing 4365:Statistical manifold 4360:Information geometry 4078:Large language model 2627:McAuliffe, Zachary. 2598:McAuliffe, Zachary. 2106:"GPT Best Practices" 1878:Cross-site scripting 1833: 1725: 1692: 1669: 1640: 1604: 1577: 1550: 1466: 1349: 1290: 1270: 1250: 1230: 1171: 1142: 1089: 1025: 961: 897: 771:: no negative prompt 467:confusing or unclear 429:can have accurately 350:Example refinement: 44:model. A prompt is 4540:In-context learning 4380:Pattern recognition 3969:www.theregister.com 2408:"Mesa-Optimization" 2358:Scientific American 475:clarify the section 224:, a 540B parameter 82:in-context learning 76:In-context learning 5133:Echo state network 5021:Jürgen Schmidhuber 4716:Facial recognition 4711:Speech recognition 4621:Software libraries 4118:Vision transformer 4088:Prompt engineering 4004:Prompt engineering 3705:Team, PromptSora. 3211:docs.llamaindex.ai 2571:The New York Times 2552:The New York Times 1848: 1819: 1760: 1750: 1707: 1675: 1655: 1626: 1590: 1563: 1536: 1452: 1384: 1374: 1335: 1276: 1256: 1236: 1216: 1150: 1124: 1075: 1011: 947: 876:image segmentation 674:), style (such as 555: 526:document retrieval 519: 376:Maieutic prompting 344:Example critique: 38:Prompt engineering 18:Few-shot prompting 5229: 5228: 4991:Stephen Grossberg 4964: 4963: 4305: 4304: 4246: 4245: 3919:simonwillison.net 2824:ai.googleblog.com 2526:ai.googleblog.com 2085:contractnerds.com 1845: 1800: 1751: 1747: 1734: 1704: 1678:{\displaystyle F} 1652: 1508: 1424: 1375: 1371: 1358: 1279:{\displaystyle F} 1259:{\displaystyle E} 1239:{\displaystyle X} 1095: 533:, summary index, 507: 506: 499: 62:few-shot learning 16:(Redirected from 5279: 5247:Machine learning 5219:Machine learning 5209: 5208: 5189: 4944:Action selection 4934:Self-driving car 4741:Stable Diffusion 4706:Speech synthesis 4671: 4670: 4535:Machine learning 4411:Gradient descent 4332: 4325: 4318: 4309: 4308: 4297: 4296: 4199:Stable Diffusion 4136: 4135: 4037: 4030: 4023: 4014: 4013: 3990: 3989: 3979: 3978: 3976: 3975: 3960: 3954: 3953: 3951: 3950: 3935: 3929: 3928: 3926: 3925: 3910: 3904: 3903: 3893: 3873: 3867: 3866: 3864: 3852: 3846: 3845: 3824: 3808: 3802: 3801: 3774: 3768: 3767: 3765: 3752: 3746: 3745: 3740: 3727: 3721: 3720: 3718: 3717: 3702: 3696: 3695: 3693: 3692: 3678: 3672: 3671: 3669: 3668: 3654: 3648: 3647: 3645: 3644: 3630: 3624: 3623: 3621: 3620: 3609: 3603: 3602: 3600: 3599: 3584: 3578: 3577: 3575: 3574: 3563: 3557: 3556: 3554: 3553: 3543: 3532: 3531: 3529: 3528: 3513: 3507: 3506: 3504: 3492: 3486: 3485: 3483: 3471: 3465: 3464: 3458: 3453: 3451: 3443: 3441: 3428: 3422: 3421: 3415: 3410: 3408: 3400: 3398: 3386: 3380: 3379: 3373: 3368: 3366: 3358: 3356: 3344: 3338: 3337: 3325: 3319: 3318: 3317: 3301: 3295: 3294: 3293: 3277: 3271: 3270: 3261: 3255: 3254: 3252: 3227: 3221: 3220: 3218: 3217: 3203: 3197: 3196: 3191: 3190: 3175: 3169: 3166: 3164: 3152: 3146: 3145: 3140: 3128: 3119: 3118: 3116: 3104: 3095: 3094: 3092: 3080: 3074: 3073: 3071: 3059: 3053: 3052: 3050: 3037: 3028: 3027: 3025: 3013: 3007: 3006: 3004: 2992: 2986: 2985: 2983: 2970: 2964: 2963: 2958: 2945: 2936: 2935: 2925: 2915: 2895: 2886: 2885: 2884: 2868: 2859: 2858: 2857: 2841: 2835: 2834: 2832: 2830: 2815: 2809: 2808: 2806: 2793: 2787: 2786: 2784: 2782: 2767: 2761: 2760: 2758: 2746: 2737: 2736: 2734: 2732: 2717: 2711: 2710: 2708: 2706: 2691: 2685: 2684: 2682: 2680: 2665: 2659: 2658: 2650: 2644: 2643: 2641: 2639: 2624: 2618: 2617: 2612: 2610: 2595: 2589: 2588: 2586: 2585: 2562: 2556: 2555: 2543: 2537: 2536: 2534: 2532: 2517: 2511: 2510: 2508: 2495: 2489: 2488: 2486: 2474: 2468: 2467: 2465: 2453: 2447: 2446: 2441: 2429: 2423: 2422: 2417: 2415: 2404: 2398: 2397: 2392: 2380: 2374: 2373: 2367: 2365: 2351:Musser, George. 2348: 2342: 2341: 2339: 2321: 2304: 2303: 2301: 2288: 2282: 2275: 2269: 2268: 2262: 2249: 2243: 2242: 2240: 2238: 2223: 2217: 2216: 2211: 2210: 2198: 2192: 2191: 2189: 2188: 2173: 2167: 2166: 2164: 2144: 2138: 2137: 2135: 2123: 2117: 2116: 2114: 2113: 2102: 2096: 2095: 2093: 2091: 2076: 2070: 2069: 2066:masterofcode.com 2057: 2051: 2050: 2045: 2044: 2029: 2020: 2019: 2014: 2013: 1998: 1992: 1991: 1986: 1977: 1971: 1970: 1955: 1949: 1948: 1939: 1938: 1933: 1924: 1891:machine learning 1883:Prompt injection 1868:Prompt injection 1862:Prompt injection 1857: 1855: 1854: 1849: 1847: 1846: 1838: 1828: 1826: 1825: 1820: 1815: 1814: 1802: 1801: 1793: 1790: 1785: 1784: 1759: 1749: 1748: 1740: 1716: 1714: 1713: 1708: 1706: 1705: 1697: 1684: 1682: 1681: 1676: 1664: 1662: 1661: 1656: 1654: 1653: 1645: 1635: 1633: 1632: 1627: 1622: 1621: 1600:into the vector 1599: 1597: 1596: 1591: 1589: 1588: 1572: 1570: 1569: 1564: 1562: 1561: 1545: 1543: 1542: 1537: 1529: 1528: 1510: 1509: 1501: 1498: 1493: 1492: 1461: 1459: 1458: 1453: 1445: 1444: 1426: 1425: 1417: 1414: 1409: 1408: 1383: 1373: 1372: 1364: 1344: 1342: 1341: 1336: 1334: 1333: 1321: 1320: 1308: 1307: 1285: 1283: 1282: 1277: 1265: 1263: 1262: 1257: 1245: 1243: 1242: 1237: 1225: 1223: 1222: 1217: 1159: 1157: 1156: 1151: 1149: 1133: 1131: 1130: 1125: 1120: 1112: 1104: 1096: 1093: 1084: 1082: 1081: 1076: 1071: 1070: 1069: 1050: 1049: 1048: 1032: 1020: 1018: 1017: 1012: 1007: 1006: 1005: 986: 985: 984: 968: 956: 954: 953: 948: 943: 942: 941: 922: 921: 920: 904: 888:gradient descent 842:Non-text prompts 818:Models include: 794:negative prompts 763:Stable Diffusion 755: 744: 733: 721:Negative prompts 688:crepuscular rays 668:digital painting 640:Stable Diffusion 502: 495: 491: 488: 482: 458: 457: 450: 257:Other techniques 245:interpretability 234:state of the art 220:When applied to 192:logical thinking 188:train of thought 180:Chain-of-thought 175:Chain-of-thought 157:chain-of-thought 144: 86:emergent ability 46:natural language 21: 5287: 5286: 5282: 5281: 5280: 5278: 5277: 5276: 5262:2022 neologisms 5232: 5231: 5230: 5225: 5177: 5091: 5057:Google DeepMind 5035: 5001:Geoffrey Hinton 4960: 4897: 4823:Project Debater 4769: 4667:Implementations 4662: 4616: 4580: 4523: 4465:Backpropagation 4399: 4385:Tensor calculus 4339: 4336: 4306: 4301: 4287: 4263:Google DeepMind 4242: 4221: 4203: 4175: 4127: 4046: 4041: 4011: 4010: 4009: 3991: 3987: 3982: 3973: 3971: 3961: 3957: 3948: 3946: 3936: 3932: 3923: 3921: 3911: 3907: 3874: 3870: 3853: 3849: 3809: 3805: 3775: 3771: 3753: 3749: 3728: 3724: 3715: 3713: 3703: 3699: 3690: 3688: 3680: 3679: 3675: 3666: 3664: 3656: 3655: 3651: 3642: 3640: 3632: 3631: 3627: 3618: 3616: 3610: 3606: 3597: 3595: 3585: 3581: 3572: 3570: 3565: 3564: 3560: 3551: 3549: 3545: 3544: 3535: 3526: 3524: 3514: 3510: 3493: 3489: 3472: 3468: 3456: 3454: 3445: 3444: 3429: 3425: 3413: 3411: 3402: 3401: 3387: 3383: 3371: 3369: 3360: 3359: 3345: 3341: 3326: 3322: 3302: 3298: 3278: 3274: 3263: 3262: 3258: 3228: 3224: 3215: 3213: 3205: 3204: 3200: 3188: 3186: 3176: 3172: 3153: 3149: 3129: 3122: 3105: 3098: 3081: 3077: 3060: 3056: 3038: 3031: 3014: 3010: 2993: 2989: 2971: 2967: 2946: 2939: 2896: 2889: 2869: 2862: 2842: 2838: 2828: 2826: 2816: 2812: 2794: 2790: 2780: 2778: 2768: 2764: 2747: 2740: 2730: 2728: 2718: 2714: 2704: 2702: 2692: 2688: 2678: 2676: 2666: 2662: 2651: 2647: 2637: 2635: 2625: 2621: 2608: 2606: 2596: 2592: 2583: 2581: 2563: 2559: 2544: 2540: 2530: 2528: 2518: 2514: 2496: 2492: 2475: 2471: 2454: 2450: 2430: 2426: 2413: 2411: 2406: 2405: 2401: 2381: 2377: 2363: 2361: 2349: 2345: 2322: 2307: 2289: 2285: 2276: 2272: 2250: 2246: 2236: 2234: 2233:. June 12, 2023 2225: 2224: 2220: 2208: 2206: 2199: 2195: 2186: 2184: 2174: 2170: 2145: 2141: 2124: 2120: 2111: 2109: 2104: 2103: 2099: 2089: 2087: 2077: 2073: 2058: 2054: 2042: 2040: 2030: 2023: 2011: 2009: 2000: 1999: 1995: 1984: 1978: 1974: 1964:The GitHub Blog 1956: 1952: 1936: 1934: 1931: 1925: 1916: 1912: 1900: 1880: 1870: 1864: 1837: 1836: 1834: 1831: 1830: 1810: 1806: 1792: 1791: 1786: 1780: 1776: 1755: 1739: 1738: 1726: 1723: 1722: 1696: 1695: 1693: 1690: 1689: 1670: 1667: 1666: 1644: 1643: 1641: 1638: 1637: 1617: 1613: 1605: 1602: 1601: 1584: 1580: 1578: 1575: 1574: 1557: 1553: 1551: 1548: 1547: 1524: 1520: 1500: 1499: 1494: 1488: 1484: 1467: 1464: 1463: 1440: 1436: 1416: 1415: 1410: 1404: 1400: 1379: 1363: 1362: 1350: 1347: 1346: 1329: 1325: 1316: 1312: 1303: 1299: 1291: 1288: 1287: 1271: 1268: 1267: 1251: 1248: 1247: 1231: 1228: 1227: 1172: 1169: 1168: 1145: 1143: 1140: 1139: 1116: 1108: 1100: 1092: 1090: 1087: 1086: 1065: 1061: 1060: 1044: 1040: 1039: 1028: 1026: 1023: 1022: 1001: 997: 996: 980: 976: 975: 964: 962: 959: 958: 937: 933: 932: 916: 912: 911: 900: 898: 895: 894: 884: 872:computer vision 864: 862:Image prompting 852: 844: 810: 790: 789: 788: 787: 786: 777:: "green trees" 758: 757: 756: 747: 746: 745: 736: 735: 734: 723: 707: 660: 628: 622: 575: 567:knowledge graph 547: 531:vector database 503: 492: 486: 483: 472: 459: 455: 448: 443: 419: 414: 411: 399: 394: 391: 378: 359: 357:Tree-of-thought 354: 348: 338: 330: 318: 313: 301: 296: 283: 278: 259: 254: 177: 172: 142: 126: 78: 35: 28: 23: 22: 15: 12: 11: 5: 5285: 5275: 5274: 5269: 5264: 5259: 5254: 5249: 5244: 5227: 5226: 5224: 5223: 5222: 5221: 5216: 5203: 5202: 5201: 5196: 5182: 5179: 5178: 5176: 5175: 5170: 5165: 5160: 5155: 5150: 5145: 5140: 5135: 5130: 5125: 5120: 5115: 5110: 5105: 5099: 5097: 5093: 5092: 5090: 5089: 5084: 5079: 5074: 5069: 5064: 5059: 5054: 5049: 5043: 5041: 5037: 5036: 5034: 5033: 5031:Ilya Sutskever 5028: 5023: 5018: 5013: 5008: 5003: 4998: 4996:Demis Hassabis 4993: 4988: 4986:Ian Goodfellow 4983: 4978: 4972: 4970: 4966: 4965: 4962: 4961: 4959: 4958: 4953: 4952: 4951: 4941: 4936: 4931: 4926: 4921: 4916: 4911: 4905: 4903: 4899: 4898: 4896: 4895: 4890: 4885: 4880: 4875: 4870: 4865: 4860: 4855: 4850: 4845: 4840: 4835: 4830: 4825: 4820: 4815: 4814: 4813: 4803: 4798: 4793: 4788: 4783: 4777: 4775: 4771: 4770: 4768: 4767: 4762: 4761: 4760: 4755: 4745: 4744: 4743: 4738: 4733: 4723: 4718: 4713: 4708: 4703: 4698: 4693: 4688: 4683: 4677: 4675: 4668: 4664: 4663: 4661: 4660: 4655: 4650: 4645: 4640: 4635: 4630: 4624: 4622: 4618: 4617: 4615: 4614: 4609: 4604: 4599: 4594: 4588: 4586: 4582: 4581: 4579: 4578: 4577: 4576: 4569:Language model 4566: 4561: 4556: 4555: 4554: 4544: 4543: 4542: 4531: 4529: 4525: 4524: 4522: 4521: 4519:Autoregression 4516: 4511: 4510: 4509: 4499: 4497:Regularization 4494: 4493: 4492: 4487: 4482: 4472: 4467: 4462: 4460:Loss functions 4457: 4452: 4447: 4442: 4437: 4436: 4435: 4425: 4420: 4419: 4418: 4407: 4405: 4401: 4400: 4398: 4397: 4395:Inductive bias 4392: 4387: 4382: 4377: 4372: 4367: 4362: 4357: 4349: 4347: 4341: 4340: 4335: 4334: 4327: 4320: 4312: 4303: 4302: 4292: 4289: 4288: 4286: 4285: 4280: 4275: 4270: 4265: 4260: 4254: 4252: 4248: 4247: 4244: 4243: 4241: 4240: 4235: 4229: 4227: 4223: 4222: 4220: 4219: 4213: 4211: 4205: 4204: 4202: 4201: 4196: 4191: 4185: 4183: 4177: 4176: 4174: 4173: 4168: 4163: 4158: 4153: 4148: 4142: 4140: 4133: 4129: 4128: 4126: 4125: 4123:Word embedding 4120: 4115: 4110: 4105: 4100: 4095: 4090: 4085: 4083:Neural network 4080: 4075: 4070: 4065: 4060: 4054: 4052: 4048: 4047: 4040: 4039: 4032: 4025: 4017: 3992: 3985: 3984: 3983: 3981: 3980: 3955: 3930: 3905: 3868: 3847: 3803: 3769: 3747: 3722: 3697: 3673: 3649: 3625: 3604: 3579: 3558: 3533: 3508: 3487: 3466: 3457:|journal= 3423: 3414:|journal= 3381: 3372:|journal= 3339: 3320: 3296: 3272: 3256: 3222: 3198: 3170: 3147: 3120: 3096: 3075: 3054: 3029: 3008: 2987: 2965: 2937: 2887: 2860: 2836: 2810: 2788: 2762: 2738: 2720:Ray, Tiernan. 2712: 2686: 2660: 2645: 2619: 2590: 2557: 2538: 2512: 2490: 2469: 2448: 2424: 2399: 2375: 2343: 2305: 2283: 2270: 2244: 2218: 2193: 2168: 2139: 2118: 2097: 2071: 2052: 2021: 1993: 1972: 1950: 1913: 1911: 1908: 1907: 1906: 1899: 1896: 1866:Main article: 1863: 1860: 1844: 1841: 1818: 1813: 1809: 1805: 1799: 1796: 1789: 1783: 1779: 1775: 1772: 1769: 1766: 1763: 1758: 1754: 1746: 1743: 1737: 1733: 1730: 1703: 1700: 1674: 1651: 1648: 1625: 1620: 1616: 1612: 1609: 1587: 1583: 1560: 1556: 1535: 1532: 1527: 1523: 1519: 1516: 1513: 1507: 1504: 1497: 1491: 1487: 1483: 1480: 1477: 1474: 1471: 1451: 1448: 1443: 1439: 1435: 1432: 1429: 1423: 1420: 1413: 1407: 1403: 1399: 1396: 1393: 1390: 1387: 1382: 1378: 1370: 1367: 1361: 1357: 1354: 1332: 1328: 1324: 1319: 1315: 1311: 1306: 1302: 1298: 1295: 1275: 1255: 1235: 1215: 1212: 1209: 1206: 1203: 1200: 1197: 1194: 1191: 1188: 1185: 1182: 1179: 1176: 1162:backpropagated 1148: 1123: 1119: 1115: 1111: 1107: 1103: 1099: 1074: 1068: 1064: 1059: 1056: 1053: 1047: 1043: 1038: 1035: 1031: 1010: 1004: 1000: 995: 992: 989: 983: 979: 974: 971: 967: 946: 940: 936: 931: 928: 925: 919: 915: 910: 907: 903: 893:Formally, let 883: 880: 863: 860: 856:word embedding 851: 848: 843: 840: 839: 838: 832: 829: 826: 809: 806: 785: 784: 778: 772: 765: 760: 759: 750: 749: 748: 739: 738: 737: 728: 727: 726: 725: 724: 722: 719: 715:Greg Rutkowski 706: 703: 676:hyperrealistic 659: 658:Prompt formats 656: 621: 618: 606: 605: 602: 599: 596: 593: 574: 571: 546: 543: 505: 504: 462: 460: 453: 447: 444: 442: 439: 418: 415: 412: 409: 398: 395: 392: 389: 377: 374: 358: 355: 352: 346: 341:"stop" token. 337: 334: 329: 326: 317: 314: 311: 300: 297: 294: 282: 279: 275: 258: 255: 252: 226:language model 176: 173: 171: 168: 125: 122: 77: 74: 50:language model 26: 9: 6: 4: 3: 2: 5284: 5273: 5270: 5268: 5265: 5263: 5260: 5258: 5255: 5253: 5250: 5248: 5245: 5243: 5242:Deep learning 5240: 5239: 5237: 5220: 5217: 5215: 5212: 5211: 5204: 5200: 5197: 5195: 5192: 5191: 5188: 5184: 5183: 5180: 5174: 5171: 5169: 5166: 5164: 5161: 5159: 5156: 5154: 5151: 5149: 5146: 5144: 5141: 5139: 5136: 5134: 5131: 5129: 5126: 5124: 5121: 5119: 5116: 5114: 5111: 5109: 5106: 5104: 5101: 5100: 5098: 5096:Architectures 5094: 5088: 5085: 5083: 5080: 5078: 5075: 5073: 5070: 5068: 5065: 5063: 5060: 5058: 5055: 5053: 5050: 5048: 5045: 5044: 5042: 5040:Organizations 5038: 5032: 5029: 5027: 5024: 5022: 5019: 5017: 5014: 5012: 5009: 5007: 5004: 5002: 4999: 4997: 4994: 4992: 4989: 4987: 4984: 4982: 4979: 4977: 4976:Yoshua Bengio 4974: 4973: 4971: 4967: 4957: 4956:Robot control 4954: 4950: 4947: 4946: 4945: 4942: 4940: 4937: 4935: 4932: 4930: 4927: 4925: 4922: 4920: 4917: 4915: 4912: 4910: 4907: 4906: 4904: 4900: 4894: 4891: 4889: 4886: 4884: 4881: 4879: 4876: 4874: 4873:Chinchilla AI 4871: 4869: 4866: 4864: 4861: 4859: 4856: 4854: 4851: 4849: 4846: 4844: 4841: 4839: 4836: 4834: 4831: 4829: 4826: 4824: 4821: 4819: 4816: 4812: 4809: 4808: 4807: 4804: 4802: 4799: 4797: 4794: 4792: 4789: 4787: 4784: 4782: 4779: 4778: 4776: 4772: 4766: 4763: 4759: 4756: 4754: 4751: 4750: 4749: 4746: 4742: 4739: 4737: 4734: 4732: 4729: 4728: 4727: 4724: 4722: 4719: 4717: 4714: 4712: 4709: 4707: 4704: 4702: 4699: 4697: 4694: 4692: 4689: 4687: 4684: 4682: 4679: 4678: 4676: 4672: 4669: 4665: 4659: 4656: 4654: 4651: 4649: 4646: 4644: 4641: 4639: 4636: 4634: 4631: 4629: 4626: 4625: 4623: 4619: 4613: 4610: 4608: 4605: 4603: 4600: 4598: 4595: 4593: 4590: 4589: 4587: 4583: 4575: 4572: 4571: 4570: 4567: 4565: 4562: 4560: 4557: 4553: 4552:Deep learning 4550: 4549: 4548: 4545: 4541: 4538: 4537: 4536: 4533: 4532: 4530: 4526: 4520: 4517: 4515: 4512: 4508: 4505: 4504: 4503: 4500: 4498: 4495: 4491: 4488: 4486: 4483: 4481: 4478: 4477: 4476: 4473: 4471: 4468: 4466: 4463: 4461: 4458: 4456: 4453: 4451: 4448: 4446: 4443: 4441: 4440:Hallucination 4438: 4434: 4431: 4430: 4429: 4426: 4424: 4421: 4417: 4414: 4413: 4412: 4409: 4408: 4406: 4402: 4396: 4393: 4391: 4388: 4386: 4383: 4381: 4378: 4376: 4373: 4371: 4368: 4366: 4363: 4361: 4358: 4356: 4355: 4351: 4350: 4348: 4346: 4342: 4333: 4328: 4326: 4321: 4319: 4314: 4313: 4310: 4300: 4290: 4284: 4281: 4279: 4276: 4274: 4271: 4269: 4266: 4264: 4261: 4259: 4256: 4255: 4253: 4249: 4239: 4236: 4234: 4231: 4230: 4228: 4224: 4218: 4215: 4214: 4212: 4210: 4206: 4200: 4197: 4195: 4192: 4190: 4187: 4186: 4184: 4182: 4178: 4172: 4169: 4167: 4164: 4162: 4159: 4157: 4154: 4152: 4149: 4147: 4144: 4143: 4141: 4137: 4134: 4130: 4124: 4121: 4119: 4116: 4114: 4111: 4109: 4106: 4104: 4101: 4099: 4096: 4094: 4091: 4089: 4086: 4084: 4081: 4079: 4076: 4074: 4071: 4069: 4066: 4064: 4063:Deep learning 4061: 4059: 4056: 4055: 4053: 4049: 4045: 4044:Generative AI 4038: 4033: 4031: 4026: 4024: 4019: 4018: 4015: 4007: 4006: 4005: 3999: 3995: 3970: 3966: 3959: 3945: 3941: 3934: 3920: 3916: 3909: 3901: 3897: 3892: 3887: 3883: 3879: 3872: 3863: 3858: 3851: 3844: 3840: 3836: 3832: 3828: 3823: 3818: 3814: 3807: 3800: 3796: 3792: 3788: 3784: 3780: 3773: 3764: 3759: 3751: 3744: 3739: 3734: 3726: 3712: 3708: 3701: 3687: 3683: 3677: 3663: 3659: 3653: 3639: 3635: 3629: 3615: 3608: 3594: 3590: 3583: 3568: 3562: 3548: 3542: 3540: 3538: 3523: 3519: 3512: 3503: 3498: 3491: 3482: 3477: 3470: 3462: 3449: 3440: 3435: 3427: 3419: 3406: 3397: 3392: 3385: 3377: 3364: 3355: 3350: 3343: 3335: 3331: 3324: 3316: 3311: 3307: 3300: 3292: 3287: 3283: 3276: 3268: 3267: 3260: 3251: 3246: 3242: 3238: 3234: 3226: 3212: 3208: 3202: 3195: 3185: 3181: 3174: 3168: 3163: 3158: 3151: 3144: 3139: 3134: 3127: 3125: 3115: 3110: 3103: 3101: 3091: 3086: 3079: 3070: 3065: 3058: 3049: 3044: 3036: 3034: 3024: 3019: 3012: 3003: 2998: 2991: 2982: 2977: 2969: 2962: 2957: 2952: 2944: 2942: 2933: 2929: 2924: 2919: 2914: 2909: 2905: 2901: 2894: 2892: 2883: 2878: 2874: 2867: 2865: 2856: 2851: 2847: 2840: 2825: 2821: 2814: 2805: 2800: 2792: 2777: 2773: 2766: 2757: 2752: 2745: 2743: 2727: 2723: 2716: 2701: 2697: 2690: 2675: 2671: 2664: 2656: 2649: 2634: 2630: 2623: 2616: 2605: 2601: 2594: 2580: 2576: 2572: 2568: 2561: 2553: 2549: 2542: 2527: 2523: 2516: 2507: 2502: 2494: 2485: 2480: 2473: 2464: 2459: 2452: 2445: 2440: 2435: 2428: 2421: 2410:. 31 May 2019 2409: 2403: 2396: 2391: 2386: 2379: 2372: 2360: 2359: 2354: 2347: 2338: 2333: 2329: 2328: 2320: 2318: 2316: 2314: 2312: 2310: 2300: 2295: 2287: 2280: 2274: 2267: 2261: 2256: 2248: 2232: 2228: 2222: 2215: 2204: 2197: 2183: 2179: 2172: 2163: 2158: 2155:: 1877–1901. 2154: 2150: 2143: 2134: 2129: 2122: 2107: 2101: 2086: 2082: 2075: 2067: 2063: 2056: 2049: 2039: 2035: 2028: 2026: 2018: 2007: 2003: 1997: 1990: 1983: 1976: 1969: 1965: 1961: 1954: 1947: 1945: 1944:text-to-image 1930: 1923: 1921: 1919: 1914: 1905: 1902: 1901: 1895: 1892: 1888: 1884: 1879: 1875: 1874:SQL injection 1869: 1859: 1839: 1811: 1807: 1803: 1794: 1781: 1777: 1770: 1767: 1764: 1761: 1756: 1752: 1741: 1731: 1728: 1718: 1698: 1686: 1672: 1665:, then apply 1646: 1618: 1614: 1607: 1585: 1581: 1558: 1554: 1525: 1521: 1514: 1511: 1502: 1489: 1485: 1478: 1475: 1472: 1469: 1441: 1437: 1430: 1427: 1418: 1405: 1401: 1394: 1391: 1388: 1385: 1380: 1376: 1365: 1355: 1352: 1330: 1317: 1313: 1309: 1304: 1300: 1273: 1253: 1233: 1207: 1201: 1195: 1192: 1186: 1180: 1177: 1174: 1165: 1163: 1137: 1113: 1105: 1057: 1054: 1051: 1033: 993: 990: 987: 969: 929: 926: 923: 905: 891: 889: 879: 877: 873: 869: 859: 857: 847: 836: 835:OpenAI's Sora 833: 830: 827: 824: 821: 820: 819: 816: 814: 813:Text-to-video 808:Text-to-video 805: 803: 799: 795: 782: 779: 776: 773: 770: 767: 766: 764: 754: 743: 732: 718: 716: 712: 705:Artist styles 702: 699: 696: 691: 689: 685: 681: 677: 673: 669: 665: 655: 653: 649: 645: 641: 637: 633: 632:text-to-image 627: 620:Text-to-image 617: 615: 611: 603: 600: 597: 594: 591: 590: 589: 587: 583: 578: 570: 568: 563: 559: 551: 542: 538: 536: 532: 527: 523: 516: 511: 501: 498: 490: 480: 479:the talk page 476: 470: 468: 463:This section 461: 452: 451: 438: 434: 432: 428: 424: 408: 405: 403: 388: 385: 382: 373: 371: 367: 366:breadth-first 363: 351: 345: 342: 333: 325: 322: 310: 307: 305: 293: 290: 287: 274: 271: 267: 266: 262: 251: 248: 246: 242: 239: 235: 231: 227: 223: 218: 216: 212: 207: 203: 201: 197: 193: 189: 185: 181: 167: 164: 163:researchers. 162: 158: 153: 150: 148: 139: 134: 131: 121: 119: 118:meta-learning 115: 111: 106: 101: 99: 95: 91: 87: 83: 73: 70: 69:text-to-image 65: 63: 59: 55: 51: 47: 43: 42:generative AI 39: 33: 19: 5062:Hugging Face 5026:David Silver 4674:Audio–visual 4528:Applications 4507:Augmentation 4352: 4268:Hugging Face 4087: 4002: 4001: 4000:profile for 3997: 3972:. Retrieved 3968: 3958: 3947:. Retrieved 3943: 3933: 3922:. Retrieved 3918: 3908: 3881: 3871: 3850: 3842: 3812: 3806: 3798: 3778: 3772: 3750: 3742: 3725: 3714:. Retrieved 3710: 3700: 3689:. Retrieved 3685: 3676: 3665:. Retrieved 3661: 3652: 3641:. Retrieved 3637: 3628: 3617:. Retrieved 3607: 3596:. Retrieved 3592: 3582: 3571:. Retrieved 3569:. 2023-05-14 3561: 3550:. Retrieved 3525:. Retrieved 3522:MLearning.ai 3521: 3511: 3490: 3469: 3448:cite journal 3426: 3405:cite journal 3384: 3363:cite journal 3342: 3333: 3323: 3305: 3299: 3281: 3275: 3265: 3259: 3240: 3236: 3225: 3214:. Retrieved 3210: 3201: 3193: 3187:. Retrieved 3173: 3167: 3150: 3142: 3078: 3057: 3011: 2990: 2968: 2960: 2903: 2872: 2845: 2839: 2827:. Retrieved 2823: 2813: 2791: 2779:. Retrieved 2775: 2765: 2729:. Retrieved 2725: 2715: 2703:. Retrieved 2699: 2689: 2677:. Retrieved 2673: 2663: 2648: 2636:. Retrieved 2632: 2622: 2614: 2607:. Retrieved 2603: 2593: 2582:. Retrieved 2570: 2560: 2551: 2541: 2529:. Retrieved 2525: 2515: 2493: 2472: 2451: 2443: 2427: 2419: 2412:. Retrieved 2402: 2394: 2378: 2369: 2362:. Retrieved 2356: 2346: 2326: 2286: 2273: 2264: 2247: 2235:. Retrieved 2230: 2221: 2213: 2207:. Retrieved 2205:. TechCrunch 2196: 2185:. Retrieved 2181: 2171: 2152: 2148: 2142: 2121: 2110:. Retrieved 2100: 2088:. Retrieved 2084: 2074: 2065: 2055: 2047: 2041:. Retrieved 2037: 2016: 2010:. Retrieved 2008:. 2022-11-30 2005: 1996: 1988: 1975: 1967: 1963: 1953: 1943: 1941: 1935:. Retrieved 1882: 1881: 1719: 1687: 1462:. In words, 1166: 892: 885: 865: 853: 845: 823:Runway Gen-2 817: 811: 801: 797: 793: 791: 780: 774: 768: 710: 708: 700: 692: 687: 684:rim lighting 683: 679: 675: 671: 667: 663: 661: 634:models like 629: 607: 581: 579: 576: 564: 560: 556: 539: 520: 493: 484: 473:Please help 464: 435: 420: 406: 401: 400: 386: 379: 361: 360: 349: 343: 339: 331: 320: 319: 308: 303: 302: 291: 285: 284: 272: 268: 264: 263: 260: 249: 219: 214: 210: 208: 204: 179: 178: 170:Text-to-text 165: 156: 155:In 2022 the 154: 151: 135: 127: 102: 81: 79: 66: 61: 57: 37: 36: 5267:Linguistics 5210:Categories 5158:Autoencoder 5113:Transformer 4981:Alex Graves 4929:OpenAI Five 4833:IBM Watsonx 4455:Convolution 4433:Overfitting 4108:Transformer 4058:Autoencoder 3662:ai.meta.com 2776:VentureBeat 2674:VentureBeat 2006:OpenAI Blog 672:photography 586:beam search 336:Self-refine 202:questions. 114:transformer 105:fine-tuning 5236:Categories 5199:Technology 5052:EleutherAI 5011:Fei-Fei Li 5006:Yann LeCun 4919:Q-learning 4902:Decisional 4828:IBM Watson 4736:Midjourney 4628:TensorFlow 4475:Activation 4428:Regression 4423:Clustering 4283:Mistral AI 4194:Midjourney 3974:2023-02-09 3949:2023-02-09 3924:2023-02-09 3862:2302.11521 3822:2104.08691 3763:2304.02643 3738:2208.01618 3716:2024-02-25 3711:PromptSora 3691:2024-02-25 3686:openai.com 3667:2024-02-25 3643:2024-02-25 3619:2023-08-14 3598:2023-08-14 3573:2023-08-14 3552:2023-08-14 3527:2022-08-31 3502:2210.03493 3481:2211.01910 3439:2309.08532 3396:2305.03495 3354:2309.16797 3315:2311.07509 3291:2404.16130 3250:2005.11401 3216:2024-04-08 3189:2024-08-31 3162:2303.08774 3138:2302.11520 3114:2205.11822 3090:2305.10601 3069:2305.08291 3048:2303.17651 3023:2210.00720 3002:2302.12246 2981:2203.11171 2956:2205.10625 2913:2110.08387 2882:2305.10276 2855:2402.07927 2804:2210.11416 2756:2205.11916 2584:2023-08-16 2506:2202.01279 2484:2110.08207 2463:1806.08730 2439:2208.01066 2390:2212.07677 2337:2201.11903 2299:2206.07682 2260:2206.07682 2209:2023-08-15 2187:2023-08-14 2162:2005.14165 2133:2208.01066 2112:2023-08-16 2043:2023-08-14 2012:2023-08-16 1987:. OpenAI. 1937:2023-08-07 1910:References 1872:See also: 698:pencils". 695:Midjourney 644:Midjourney 624:See also: 535:tree index 515:embeddings 469:to readers 431:calibrated 423:likelihood 230:fine-tuned 196:arithmetic 147:entailment 96:, meaning 5082:MIT CSAIL 5047:Anthropic 5016:Andrew Ng 4914:AlphaZero 4758:VideoPoet 4721:AlphaFold 4658:MindSpore 4612:SpiNNaker 4607:Memristor 4514:Diffusion 4490:Rectifier 4470:Batchnorm 4450:Attention 4445:Adversary 4258:Anthropic 4251:Companies 3900:226222232 3839:233296808 3795:230433941 3547:"Prompts" 2932:239016123 2579:0362-4331 2371:learning. 2292:Models". 2253:Models". 1843:~ 1804:∗ 1798:~ 1765:⁡ 1753:∑ 1745:~ 1732:⁡ 1702:~ 1650:~ 1512:∗ 1506:~ 1473:⁡ 1428:∗ 1422:~ 1389:⁡ 1377:∑ 1369:~ 1356:⁡ 1055:… 991:… 927:… 866:In 2023, 630:In 2022, 614:clustered 487:July 2024 407:Example: 387:Example: 309:Example: 292:Example: 273:Example: 250:Example: 241:benchmark 215:zero-shot 5190:Portals 4949:Auto-GPT 4781:Word2vec 4585:Hardware 4502:Datasets 4404:Concepts 4299:Category 4051:Concepts 3944:Hackaday 2829:10 March 2781:10 March 2731:10 March 2705:10 March 2679:10 March 2638:10 March 2609:10 March 2531:10 March 2237:June 12, 2231:claid.ai 2108:. OpenAI 1898:See also 1226:, where 636:DALL-E 2 381:Maieutic 211:few-shot 5072:Meta AI 4909:AlphaGo 4893:PanGu-Σ 4863:ChatGPT 4838:Granite 4786:Seq2seq 4765:Whisper 4686:WaveNet 4681:AlexNet 4653:Flux.jl 4633:PyTorch 4485:Sigmoid 4480:Softmax 4345:General 4278:Meta AI 4233:Suno AI 3994:Scholia 2090:24 July 680:pop-art 465:may be 124:History 110:dataset 5087:Huawei 5067:OpenAI 4969:People 4939:MuZero 4801:Gemini 4796:Claude 4731:DALL-E 4643:Theano 4273:OpenAI 4226:Musics 4209:Videos 4189:DALL-E 4181:Images 4151:Gemini 4146:Claude 4132:Models 3996:has a 3898: 3837: 3793: 3269:, 2024 3184:Forbes 2930: 2577: 2414:17 May 2364:17 May 2038:Zapier 1829:where 1136:losses 1094:concat 781:Bottom 775:Centre 648:AI art 642:, and 161:Google 98:breaks 5153:Mamba 4924:SARSA 4888:LLaMA 4883:BLOOM 4868:GPT-J 4858:GPT-4 4853:GPT-3 4848:GPT-2 4843:GPT-1 4806:LaMDA 4638:Keras 4171:LLaMA 4166:GPT-4 4161:GPT-3 4156:GPT-2 3998:topic 3896:S2CID 3857:arXiv 3835:S2CID 3817:arXiv 3791:S2CID 3758:arXiv 3733:arXiv 3497:arXiv 3476:arXiv 3434:arXiv 3391:arXiv 3349:arXiv 3334:arXiv 3310:arXiv 3286:arXiv 3245:arXiv 3157:arXiv 3133:arXiv 3109:arXiv 3085:arXiv 3064:arXiv 3043:arXiv 3018:arXiv 2997:arXiv 2976:arXiv 2951:arXiv 2928:S2CID 2908:arXiv 2877:arXiv 2850:arXiv 2799:arXiv 2751:arXiv 2726:ZDNET 2501:arXiv 2479:arXiv 2458:arXiv 2434:arXiv 2385:arXiv 2332:arXiv 2294:arXiv 2255:arXiv 2157:arXiv 2128:arXiv 1985:(PDF) 1932:(PDF) 427:GPT-4 5077:Mila 4878:PaLM 4811:Bard 4791:BERT 4774:Text 4753:Sora 4238:Udio 4217:Sora 4139:Text 4098:RLHF 3461:help 3418:help 3376:help 2831:2023 2783:2023 2733:2023 2707:2023 2681:2023 2640:2023 2633:CNET 2611:2023 2604:CNET 2575:ISSN 2533:2023 2416:2023 2366:2023 2239:2023 2092:2023 1876:and 1021:and 868:Meta 693:The 610:BERT 580:The 370:beam 222:PaLM 4818:NMT 4701:OCR 4696:HWR 4648:JAX 4602:VPU 4597:TPU 4592:IPU 4416:SGD 4093:RAG 3886:doi 3827:doi 3783:doi 2918:doi 1762:log 1736:max 1729:arg 1470:log 1386:log 1360:max 1353:arg 798:not 769:Top 686:or 678:or 670:or 198:or 138:NLP 130:NLP 88:of 58:dog 5238:: 3967:. 3942:. 3917:. 3894:. 3880:. 3841:. 3833:. 3825:. 3797:. 3789:. 3741:. 3709:. 3684:. 3660:. 3636:. 3591:. 3536:^ 3520:. 3452:: 3450:}} 3446:{{ 3409:: 3407:}} 3403:{{ 3367:: 3365:}} 3361:{{ 3332:. 3308:, 3284:, 3241:33 3239:. 3235:. 3209:. 3192:. 3182:. 3141:. 3123:^ 3099:^ 3032:^ 2959:. 2940:^ 2926:. 2916:. 2902:. 2890:^ 2875:, 2863:^ 2848:, 2822:. 2774:. 2741:^ 2724:. 2698:. 2672:. 2631:. 2613:. 2602:. 2573:. 2569:. 2550:. 2524:. 2442:. 2418:. 2393:. 2368:. 2355:. 2308:^ 2263:. 2229:. 2212:. 2180:. 2153:33 2151:. 2083:. 2064:. 2046:. 2036:. 2024:^ 2015:. 2004:. 1966:. 1962:. 1940:. 1917:^ 1685:. 717:. 638:, 368:, 247:. 149:. 64:. 4331:e 4324:t 4317:v 4036:e 4029:t 4022:v 4008:. 3977:. 3952:. 3927:. 3902:. 3888:: 3865:. 3859:: 3829:: 3819:: 3785:: 3766:. 3760:: 3735:: 3719:. 3694:. 3670:. 3646:. 3622:. 3601:. 3576:. 3555:. 3530:. 3505:. 3499:: 3484:. 3478:: 3463:) 3459:( 3442:. 3436:: 3420:) 3416:( 3399:. 3393:: 3378:) 3374:( 3357:. 3351:: 3336:. 3312:: 3288:: 3253:. 3247:: 3219:. 3165:. 3159:: 3135:: 3117:. 3111:: 3093:. 3087:: 3072:. 3066:: 3051:. 3045:: 3026:. 3020:: 3005:. 2999:: 2984:. 2978:: 2953:: 2934:. 2920:: 2910:: 2879:: 2852:: 2833:. 2807:. 2801:: 2785:. 2759:. 2753:: 2735:. 2709:. 2683:. 2657:. 2642:. 2587:. 2554:. 2535:. 2509:. 2503:: 2487:. 2481:: 2466:. 2460:: 2436:: 2387:: 2340:. 2334:: 2302:. 2296:: 2257:: 2241:. 2190:. 2165:. 2159:: 2136:. 2130:: 2115:. 2094:. 2068:. 1840:X 1817:] 1812:i 1808:X 1795:X 1788:| 1782:i 1778:Y 1774:[ 1771:r 1768:P 1757:i 1742:X 1699:Z 1673:F 1647:Z 1624:) 1619:i 1615:X 1611:( 1608:E 1586:i 1582:X 1559:i 1555:Y 1534:] 1531:) 1526:i 1522:X 1518:( 1515:E 1503:Z 1496:| 1490:i 1486:Y 1482:[ 1479:r 1476:P 1450:] 1447:) 1442:i 1438:X 1434:( 1431:E 1419:Z 1412:| 1406:i 1402:Y 1398:[ 1395:r 1392:P 1381:i 1366:Z 1331:i 1327:} 1323:) 1318:i 1314:Y 1310:, 1305:i 1301:X 1297:( 1294:{ 1274:F 1254:E 1234:X 1214:) 1211:) 1208:X 1205:( 1202:E 1199:( 1196:F 1193:= 1190:) 1187:X 1184:( 1181:M 1178:L 1175:L 1147:Y 1122:) 1118:Y 1114:; 1110:X 1106:; 1102:E 1098:( 1073:} 1067:n 1063:y 1058:, 1052:, 1046:1 1042:y 1037:{ 1034:= 1030:Y 1009:} 1003:m 999:x 994:, 988:, 982:1 978:x 973:{ 970:= 966:X 945:} 939:k 935:e 930:, 924:, 918:1 914:e 909:{ 906:= 902:E 500:) 494:( 489:) 485:( 481:. 34:. 20:)

Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.