Large Language Models (LLMs) [27], including Generative Pre-trained Transformer (GPT) [6] and Bidirectional Encoder Representations from Transformers (BERT) [13], have achieved state-of-theart performance on a wide range of Natural Language Processing (NLP) tasks. LLMs are gaining popularity and receiving increasing attention for their significant applications in knowledge reasoning [12, 52, 57]. ChatGPT is one of the LLMs applications, and probably the application, in the limelight. ChatGPT was used for collating literature and writing professional papers in fields like law [9], and medical education [30, 16]. OpenAI announced GPT-4 in March 2023 that can pass some of the bar exams to AP Biology [39]. These successful stories demonstrate that people have already gained experience in using LLMs, for their performance in handling complex content due to their massive training datasets and model capacity to process and learn from data, enabling their potential for complex tasks that require domain expert knowledge [38]. Given this, as researchers in the field of safety-critical systems, we pose a question: Can safety analysis make use of LLMs?

Adjusting the latency, power, and accuracy of natural language understanding models is a desirable objective of an efficient architecture. This paper proposes an efficient Transformer architecture that adjusts the inference computational cost adaptively with a desired inference latency speedup. In fine-tuning phase, the proposed method detects less important hidden sequence elements (word-vectors) and eliminates them in each encoder layer using a proposed Attention Context Contribution (ACC) metric. After the fine-tuning phase, with the novel offline-tuning property, the inference latency of the model can be adjusted in a wide range of inference speedup selections without any further training. The proposed method is applied to the BERT-base and GPT-2 models for evaluation. Extensive experiments show that most of the word-vectors in higher Transformer layers have less contribution to the subsequent layers; hence, they can be eliminated to improve the inference latency. Experimental results on extensive sentiment analysis, classification, text generation tasks and regression benchmarks like GLUE showed that the method is effective in various datasets with minimal impact on global context. The proposed method mathematically and experimentally improves the inference latency of BERT-base and GPT-2 by up to 4.8 and 3.72 times with less than 0.75% accuracy drop and passable perplexity on average. The suggested approach posits that in Large Language Models (LLMs), although the complete network is necessary for training, it can be truncated during the fine-tuning phase.

Purpose: To assess Generative Pre-trained Transformer version 4's (GPT-4) ability to classify articles containing functional evidence relevant to assessments of variant pathogenicity. Results: GPT-4 settings and prompts were trained on a set of 45 articles and genetic variants. A final test set of 72 manually classified articles and genetic variants were then processed using two prompts. The prompts asked GPT-4 to supply all functional evidence present in an article for a variant or indicate that no functional evidence is present. For articles with having functional evidence, a second prompt asked GPT-4 to classify the evidence into pathogenic, benign, intermediate, and inconclusive categories. The first prompt identified articles with variant-level functional evidence with 87% sensitivity and 89% positive predictive value (PPV). Five of 26 articles with no functional data were indicated as having functional evidence by GPT-4. For variants with functional assays present as determined by both manual review and GPT-4, the sensitivity and PPV of GPT-4 prompt concordance was: Pathogenic (92% sensitive and 73% PPV), Intermediate or Inconclusive (67% sensitive and 93% PPV), Benign (100% sensitive and 73% PPV). Conclusion: The GPT-4 prompts detected the presence or absence of a functional assay with high sensitivity and PPV, and articles with unambiguous evidence supporting a benign or pathogenic classification with high sensitivity and reasonable PPV. Our prompts detected papers with intermediate or inconclusive evidence with lower sensitivity but high PPV. Our results support that GPT-4 may be useful in variant classification workflows by enabling prioritization of articles for review that are likely to have functional evidence supporting or refuting pathogenicity, but not that GPT-4 is capable of fully automating the genetics literature review component of variant classification.

Large decoder-only language models (LMs) can be largely improved in terms of perplexity by retrieval (e.g., RETRO), but its impact on text generation quality and downstream task accuracy is unclear. Thus, it is still an open question: shall we pretrain large autoregressive LMs with retrieval? To answer it, we perform a comprehensive study on a scalable pre-trained retrieval-augmented LM (i.e., RETRO) compared with standard GPT and retrieval-augmented GPT incorporated at fine-tuning or inference stages. We first provide the recipe to reproduce RETRO up to 9.5B parameters while retrieving a text corpus with 330B tokens. Based on that, we have the following novel findings: i) RETRO outperforms GPT on text generation with much less degeneration (i.e., repetition), moderately higher factual accuracy, and slightly lower toxicity with a nontoxic retrieval database. ii) On the LM Evaluation Harness benchmark, RETRO largely outperforms GPT on knowledge-intensive tasks, but is on par with GPT on other tasks. Furthermore, we introduce a simple variant of the model, RETRO++, which largely improves open-domain QA results of original RETRO (e.g., EM score +8.6 on Natural Question) and significantly outperforms retrieval-augmented GPT in both fine-tuning and zero-shot evaluation settings. Our findings highlight the promising direction of pretraining autoregressive LMs with retrieval as future foundation models. We release our code and model at: https://github.com/NVIDIA/Megatron-LM/blob/main/tools/retro/README.md

Microsoft Copilot just rolled out a new feature that creates songs via text prompt, thanks to a partnership with AI-based music creation platform Suno. Microsoft says it gives users the ability to craft personalized songs, "regardless of musical background." Suno has long been working with various algorithms to create an AI that can whip up entire songs and it looks like the company has struck gold. Music creation now made easier with @Suno_ai_ integration in Copilot. When you access this tool, all you have to do is enter a prompt and let the algorithm do the rest.

Here's what 2023 taught me: The year started with Big Tech going all in on generative AI. The runaway success of OpenAI's ChatGPT prompted every major tech company to release its own version. This year might go down in history as the year we saw the most AI launches: Meta's LLaMA 2, Google's Bard chatbot and Gemini, Baidu's Ernie Bot, OpenAI's GPT-4, and a handful of other models, including one from a French open-source challenger, Mistral. But despite the initial hype, we haven't seen any AI applications become an overnight success. Microsoft and Google pitched powerful AI-powered search, but it turned out to be more of a dud than a killer app.

TomTom just announced a "fully integrated, AI-powered conversational automotive assistant" which should start popping up in dashboard infotainment platforms in the near-ish future. The company has issued some bold claims for the AI, saying it'll offer "more sophisticated voice interaction" and allow users to converse naturally to navigate, find stops along a route, control onboard systems, open windows and just about anything else you find yourself doing while driving. The company, best known for GPS platforms, partnered up with Microsoft to develop this AI assistant. Cosmos DB is a multi-model database and Cognitive Services is a set of APIs for use in AI applications, so this should be a capable assistant that draws from the latest advancements. TomTom promises that the voice assistant will integrate into a variety of interfaces offered by major automobile manufacturers, stating that the auto company will retain ownership of its branding.

Prof Michael Wooldridge has been an AI researcher for more than 30 years, and in the year that AI was supercharged by ChatGPT, he is giving the Royal Institution's Christmas lectures on the truth about AI. The Guardian science correspondent Nicola Davis sat down with him to find out how he sees AI evolving, what makes human intelligence unique, and what really keeps him awake at night. Madeleine Finlay hears from them both in this Science Weekly Christmas special.

We present a novel methodology aimed at optimizing the application of frozen large language models (LLMs) for resource-intensive vision-language (VL) pre-training. The current paradigm uses visual features as prompts to guide language models, with a focus on determining the most relevant visual features for corresponding text. Our approach diverges by concentrating on the language component, specifically identifying the optimal prompts to align with visual features. We introduce the Prompt-Transformer (P-Former), a model that predicts these ideal prompts, which is trained exclusively on linguistic data, bypassing the need for image-text pairings. This strategy subtly bifurcates the end-to-end VL training process into an additional, separate stage. Our experiments reveal that our framework significantly enhances the performance of a robust image-to-text baseline (BLIP-2), and effectively narrows the performance gap between models trained with either 4M or 129M image-text pairs. Importantly, our framework is modality-agnostic and flexible in terms of architectural design, as validated by its successful application in a video learning task using varied base modules. The code will be made available at https://github.com/yiren-jian/BLIText.

Spreadsheets are a vital tool for end-user data management. Using large language models for formula authoring assistance in these environments can be difficult, as these models are expensive to train and challenging to deploy due to their size (up to billions of parameters). We present FLAME, a transformer-based model trained exclusively on Excel formulas that leverages domain insights to achieve competitive performance while being substantially smaller (60M parameters) and training on two orders of magnitude less data. We curate a training dataset using sketch deduplication, introduce an Excel-specific formula tokenizer, and use domain-specific versions of masked span prediction and noisy auto-encoding as pre-training objectives. We evaluate FLAME on formula repair, formula completion, and similarity-based formula retrieval. FLAME can outperform much larger models, such as the Davinci (175B) and Cushman (12B) variants of Codex and CodeT5 (220M), in 10 of 14 evaluation settings for the repair and completion tasks. For formula retrieval, FLAME outperforms CodeT5, CodeBERT, and GraphCodeBERT.