False Outputs Research Articles

Detecting hallucinations in large language models using semantic entropy

Large language model (LLM) systems, such as ChatGPT1 or Gemini2, can show impressive reasoning and question-answering capabilities but often ‘hallucinate’ false outputs and unsubstantiated answers3,4. Answering unreliably or without the necessary information prevents adoption in diverse fields, with problems including fabrication of legal precedents5 or untrue facts in news articles6 and even posing a risk to human life in medical domains such as radiology7. Encouraging truthfulness through supervision or reinforcement has been only partially successful8. Researchers need a general method for detecting hallucinations in LLMs that works even with new and unseen questions to which humans might not know the answer. Here we develop new methods grounded in statistics, proposing entropy-based uncertainty estimators for LLMs to detect a subset of hallucinations—confabulations—which are arbitrary and incorrect generations. Our method addresses the fact that one idea can be expressed in many ways by computing uncertainty at the level of meaning rather than specific sequences of words. Our method works across datasets and tasks without a priori knowledge of the task, requires no task-specific data and robustly generalizes to new tasks not seen before. By detecting when a prompt is likely to produce a confabulation, our method helps users understand when they must take extra care with LLMs and opens up new possibilities for using LLMs that are otherwise prevented by their unreliability.

Societal impacts of chatbot and mitigation strategies for negative impacts: A large-scale qualitative survey of ChatGPT users

Discussing ChatGPT's societal impact often focuses on specific domains or groups, typically involving a limited number of participants. In contrast, this study investigates the impact of this artificial intelligence (AI) on society as a whole by conducting a qualitative online survey with 1298 daily users from Hinterland of China across various industries. Identified key benefits include enhanced efficiency and production, improved communication, understand, and cooperation, increased social equalization and harmony, and bolstered public mental health. On the downside, concerns arose about (un)employment issues, biased or false outputs, potential for unethical applications, and user dependency and its negative consequences. To address these concerns, participants suggested use restrictions and management, promoting independent and critical thinking/practice, enhancing user literacy, increasing individual irreplaceability, technological improvement and optimization, and safeguarding against AI threats to human society. The above results elucidate the impact of ChatGPT on various social domains and potential solutions for its negative effects. These findings augment existing research on AI user models from a societal impact perspective, providing a foundation for proposing AI user models that consider the social implications of AI. These outcomes also aim to assist policymakers, researchers, and designers in better understanding the societal impact and user concerns regarding AI, thereby informing the design and enhancement of products, services, or policies.

Fake News Detection Using Enhanced BERT

Since there are so many users on social media, who are not qualified to report news, fake news has become a major problem in recent years. Therefore, it is crucial to identify and restrict the dissemination of false information. Numerous deep learning models that make use of natural language processing have yielded excellent results in the detection of fake news. bidirectional encoder representations from transformers (BERT), based on transfer learning, is one of the most advanced models. In this work, the researchers have compared the earlier studies that employed baseline models versus the research articles where the researchers used a pretrained model BERT for the detection of fake news. The literature analysis revealed that utilizing pretrained algorithms is more effective at identifying fake news because it takes less time to train them and yields better results. Based on the results noted in this article, the researchers have advised the utilization of pretrained models that have already been taught to take advantage of transfer learning, which shortens training time and enables the use of large datasets, as well as a reputable model that performs well in terms of precision, recall, as well as the minimum number of false positive and false negative outputs. As a result, the researchers created an improved BERT model, while considering fine-tuning it to meet the demands of the fake news identification assignment. To obtain the most accurate representation of the input text, the final layer of this model is also unfrozen and trained on news texts. The dataset used in the study included 23 502 articles of fake news and 21 417 items of actual news. This dataset was downloaded from the Kaggle website. The results of this study demonstrated that the proposed model showed a better performance compared with other models, and achieved 99.96% and 99.96% in terms of accuracy and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F$</tex-math> </inline-formula> 1 score, respectively.

Advancing radiology practice and research: harnessing the potential of large language models amidst imperfections.

Large language models (LLMs) are transforming the field of natural language processing (NLP). These models offer opportunities for radiologists to make a meaningful impact in their field. NLP is a part of artificial intelligence (AI) that uses computer algorithms to study and understand text data. Recent advances in NLP include the Attention mechanism and the Transformer architecture. Transformer-based LLMs, such as GPT-4 and Gemini, are trained on massive amounts of data and generate human-like text. They are ideal for analysing large text data in academic research and clinical practice in radiology. Despite their promise, LLMs have limitations, including their dependency on the diversity and quality of their training data and the potential for false outputs. Albeit these limitations, the use of LLMs in radiology holds promise and is gaining momentum. By embracing the potential of LLMs, radiologists can gain valuable insights and improve the efficiency of their work. This can ultimately lead to improved patient care.

A compound convolutional neural network used to process small perturbation

Input samples formed by deliberately adding subtle interference to the data set may cause the model to give false outputs with high confidence. To solve this problem, this paper proposes a deep learning method. Different interference sources can be considered to be independent of each other. Therefore, different networks are trained based on densenet for different interference sources. Then, the data set added to the combined interference item is sent to each neural network, and the output of each network is used as the input of the neural network in the next stage. The next stage of the neural network is used for multi-objective classification based on the output of the neural network. The output of the neural network is disturbed by adding different features to the input data. Because the depth network of each type of interference is independent, when introducing a new interference term, only the depth network and full connection layer of the new interference term need to be trained, rather than the whole network, which greatly simplifies the training process when introducing a new interference source.

Photo-gated and self-powered three-dimensional DNA motors with boosted biostability for exceptionally precise and efficient tracing of intracellular survivin mRNA

Benefiting from the outstanding signal amplification effect and the admirable construction flexibility, the currently proposed DNA motors (particularly DNA walkers) based biosensing concepts have provided a forceful fluorescence imaging tool for intracellular detection. Even so, this promising sensing means is not only subject to poor controllability and prone to produce false signals but also requires exogenous powering forces owing to the common employment of DNAzyme. In response to these challenges, we are herein motivated to present some meaningful solving strategies. For one thing, the surfaces of gold nanoparticles are conducted with a photo-gated walking behavior by introducing a photocleave mode, under which the light-switchable DNA walkers are capable of being selectively activated via an external ultraviolet source to faultlessly prevent the sensing frame from being pre-initiated during cellular uptake and intracellular delivery. For another, the intracellular biothiols are consumed by MnO2 nanosheets to effectively avoid the competitions to Au–S bonds to eliminate potential false outputs and also self-supply sufficient cofactors (Mn2+) to actualize a self-powered operation pattern as well as facilitate the endocytosis process. Following these breakthroughs, a favorable analysis performance towards a model tumor biomarker (survivin mRNA) is endowed with the newly raised biosensor, whose sensitivity is low to pM level with a sound specificity for identifying single base mismatching. Moreover, the significantly improved autonomous three-dimensional DNA walkers can be used to determine and dynamically trace the targets in live cancer cells with an exceptional precise and efficient manner, commendably impelling the sensing ability of DNA motors in biological specimens.

Depth Information-Based Automatic Annotation of Early Esophageal Cancers in Gastroscopic Images Using Deep Learning Techniques

The early diagnoses of esophageal cancer are of great significance in the clinic because they are critical for reducing mortality. At present, the diagnoses are mainly performed by artificial detection and annotations based on gastroscopic images. However, these procedures are very challenging to clinicians due to the large variability in the appearance of early cancer lesions. To reduce the subjectivity and fatigue in manual annotations and to improve the efficiency of diagnoses, computer-aided annotation methods are highly required. In this work, we proposed a novel method that utilized deep learning (DL) techniques to realize the automatic annotation of early esophageal cancer (EEC) lesions in gastroscopic images. The depth map of gastroscopic images was initially extracted by a DL network. Then, this additional depth information was fused with the original RGB gastroscopic images, which were then sent to another DL network to obtain precise annotations of EEC regions. In total, 4231 gastroscopic images of 732 patients were used to build and validate the proposed method. A total of 3190 of those images were EEC images, and the remaining 1041 were non-EEC images. The experimental results show that the combination of depth information and RGB information improved the annotation performance. The final EEC detection rate and mean Dice Similarity Coefficient (DSC) of our method were 97.54% and 74.43%, respectively. Compared with other state-of-the-art DL-based methods, the proposed method showed better annotation performances and fewer false positive outputs. Therefore, our method offers a good prospect in aiding the clinical diagnoses of EEC.

Microslit on a chip: A simplified filter to capture circulating tumor cells enlarged with microbeads.

Microchips are widely used to separate circulating tumor cells (CTCs) from whole blood by virtues of sophisticated manipulation for microparticles. Here, we present a chip with an 8 μm high and 27.9 mm wide slit to capture cancer cells bound to 3 μm beads. Apart from a higher purity and recovery rate, the slit design allows for simplified fabrication, easy cell imaging, less clogging, lower chamber pressure and, therefore, higher throughput. The beads were conjugated with anti-epithelial cell adhesion molecules (anti-EpCAM) to selectively bind to breast cancer cells (MCF-7) used to spike the whole blood. The diameter of the cell-bead construct was in average 23.1 μm, making them separable from other cells in the blood. As a result, the cancer cells were separated from 5 mL of whole blood with a purity of 52.0% and a recovery rate of 91.1%, and also we confirmed that the device can be applicable to clinical samples of human breast cancer patients. The simple design with microslit, by eliminating any high-aspect ratio features, is expected to reduce possible defects on the chip and, therefore, more suitable for mass production without false separation outputs.

Annotating Early Esophageal Cancers Based on Two Saliency Levels of Gastroscopic Images

Early diagnoses of esophageal cancer can greatly improve the survival rate of patients. At present, the lesion annotation of early esophageal cancers (EEC) in gastroscopic images is generally performed by medical personnel in a clinic. To reduce the effect of subjectivity and fatigue in manual annotation, computer-aided annotation is required. However, automated annotation of EEC lesions using images is a challenging task owing to the fine-grained variability in the appearance of EEC lesions. This study modifies the traditional EEC annotation framework and utilizes visual salient information to develop a two saliency levels-based lesion annotation (TSL-BLA) for EEC annotations on gastroscopic images. Unlike existing methods, the proposed framework has a strong ability of constraining false positive outputs. What is more, TSL-BLA is also placed an additional emphasis on the annotation of small EEC lesions. A total of 871 gastroscopic images from 231 patients were used to validate TSL-BLA. 365 of those images contain 434 EEC lesions and 506 images do not contain any lesions. 101 small lesion regions are extracted from the 434 lesions to further validate the performance of TSL-BLA. The experimental results show that the mean detection rate and Dice similarity coefficients of TSL-BLA were 97.24 and 75.15%, respectively. Compared with other state-of-the-art methods, TSL-BLA shows better performance. Moreover, it shows strong superiority when annotating small EEC lesions. It also produces fewer false positive outputs and has a fast running speed. Therefore, The proposed method has good application prospects in aiding clinical EEC diagnoses.

The Impact of Employee Sycophantic Behaviour on Organisation Environment: A Conceptual Study of Hospitality Sector in India

In today competitive environment, sycophantic behaviour played in every organisation significant role and the organisation environment also influenced by this behaviour. If we go into deep discussion, we realize as well that this behaviour is behaviour to get competitive advantage for personal growth. It is also observed that, performance of organisation depends upon the surrounding as read in many HRM books that “Good environment leads to good performance of any companies”. Service industry required team spirit for increasing performance level and for that behaviour is generally being adopted by the employees to please their bosses. Some questions arise like: 1) what does sycophancy means? 2) How does it influence the environment? And does have positive impact on employee Performance? To do Oiling and battering to their bosses to get promotion and good place in the industry is known as sycophancy. According Ekong and Essien, the sycophantic groups are those who involve more in intrigues, lying about others to cause confusion-official liar and those who pretend to be the adviser to the political leader. But their false praise performance and outputs may vary. The above matter can be seen in private sector as well in public or government sector also at some extent. But if we talk about government sector, Jhatial found in his study that 80% of respondents in government sector agreed that connection or sifarish, references, sycophancy and cronyism play significant role in HRM decision making process whereas only 20% of respondents found that it have done on merit based decision.

A VELOUR PLEBISCITE TRACING FOR HARDWARE AND SOFTWARE MISCUE REMEDIAL SYSTEMS

Referending algorithms are used to adjudicate among the consequences of surplus modules in miscue-remedial systems. Not exactly, popular allies entail a request-precise ’Disciple Point’ value to be précised, whereas biased standard allies are incapable to generate a compassionate productivity while refusal concurrence exists between the allies inputs. The disciple is tentatively appraised from the summit of vision security and accessibility, and contrast with the imperfect popular disciple in a Triple Modular Redundancy prearranged outline. We show that the Velour disciple gives extra exact outputs (advanced accessibility) than the wrong popular disciple with petite and hefty slip-ups, fewer false outputs (superior security) than the imperfect popular disciple in the occurrence of tiny slip-ups, and a smaller amount kind outputs than the inaccurate popular choose. The proportion of the kind outputs of the bulk disciples that are fruitfully hold by the Velour fanatic (ensuing in accurate outputs) is more than the proportion of those that are disastrously determined by the Velour fanatic (ensuing in erroneous outputs).

Fuzzy Three-valued Switching Functions Using Choquet Integral

Fuzzy three-valued switching functions have true, false and unknown inputs and outputs values. The functions are calculated by Choquet integral whose fuzzy measure values are assigned from normal three-valued switching functions. The functions are extension of normal three-valued functions and satisfy the monotonicity with respect to vagueness.

Le projet isis: une interface intégrée de sécurité

In order to meet safety, reliability and availability constraints for passenger transport systems, the manufacturers of railway control-command systems must develop specific computer architectures for signalling, switch control and automatic driving applications. Typically, these systems consist of a core (with one or two central processing units) and peripherals via which the computer communicates with external devices. In the case of some applications, for example signal boxes, the architecture contains a very large number of fail-safe True or False outputs. These functionalities are currently performed by conventional technology that uses discrete components or hybrid circuitry.

Deep submarine gas vents in the aeolian offshore

Consistent results, concerning the detection of deep submarine hydrothermal vents, were obtained during two cruises (1991 and 1996) offshore from the Aeolian Islands, by CTD profiling from sea surface down to seafloor, and water-sampling casts. In 1991 an echo sounder showed a wide plume at a depth of about 800 m, within which water samples displayed anomalies in He and NH a content, suggesting also the presence of a water-vapour phase. The latter, in 1996, was remarkably observed as a horizontally diffusing plume at about 350 m. Near-plume casts were characterised by high CO2 and CH4 and low O2 concentrations in seawater, disturbed light transmission profiles, and false bottom outputs appearing at ∼300–350 m down to the seafloor from the rosette-mounted altimeter. No significant temperature/salinity anomalies were noted during either events. These preliminary results show the presence of deep hydrothermal activity, over an area where, one century ago, the occurrence of submarine eruptions was detected.

A Biologically Supported Error-Correcting Learning Rule.

We show that a form of synaptic plasticity recently discovered in slices of the rat visual cortex (Artola et al. 1990) can support an error-correcting learning rule. The rule increases weights when both pre- and postsynaptic units are highly active, and decreases them when pre-synaptic activity is high and postsynaptic activation is less than the threshold for weight increment but greater than a lower threshold. We show that this rule corrects false positive outputs in feedforward associative memory, that in an appropriate opponent-unit architecture it corrects misses, and that it performs better than the optimal Hebbian learning rule reported by Willshaw and Dayan (1990).

Static hazards in combinational threshold logic networks†

The possibility of transitory false outputs in conventional digital logic circuits is well known, such output ‘ spikes ’ being the result of different propagation times through the logic network from inputs to output. The usual solution to such ‘ Static hazards ’ is also well known, being the incorporation of additional gates in the system to cover such input transitions. This paper shows that the application of threshold logic gates to logic synthesis has attractions in very easily eliminating such hazards, in many cases without the need for any additional covering gates in the network.

Secure Information Transmission Over Microwave and Telephone Channels

Communications systems are often used for transmission of control signals where the possibility of false outputs must be minimized. To minimize erroneous system outputs, special design considerations are involved.