We suggest that foundation models are general purpose solutions similar to general purpose programmable microprocessors, where fine-tuning and prompt-engineering are analogous to coding for microprocessors. Evaluating general purpose solutions is not like hypothesis testing. We want to know how well the machine will perform on an unknown program with unknown inputs for unknown users with unknown budgets and unknown utility functions. This paper is based on an invited talk by John Mashey, “Lessons from SPEC,” at an ACL-2021 workshop on benchmarking. Mashey started by describing Standard Performance Evaluation Corporation (SPEC), a benchmark that has had more impact than benchmarks in our field because SPEC addresses an import commercial question: which CPU should I buy? In addition, SPEC can be interpreted to show that CPUs are 50,000 faster than they were 40 years ago. It is remarkable that we can make such statements without specifying the program, users, task, dataset, etc. It would be desirable to make quantitative statements about improvements of general purpose foundation models over years/decades without specifying tasks, datasets, use cases, etc.

The previous Emerging Trends article (Church et al., 2021. Natural Language Engineering27(5), 631–645.) introduced deep nets to poets. Poets is an imperfect metaphor, intended as a gesture toward inclusion. The future for deep nets will benefit by reaching out to a broad audience of potential users, including people with little or no programming skills, and little interest in training models. That paper focused on inference, the use of pre-trained models, as is, without fine-tuning. The goal of this paper is to make fine-tuning more accessible to a broader audience. Since fine-tuning is more challenging than inference, the examples in this paper will require modest programming skills, as well as access to a GPU. Fine-tuning starts with a general purpose base (foundation) model and uses a small training set of labeled data to produce a model for a specific downstream application. There are many examples of fine-tuning in natural language processing (question answering (SQuAD) and GLUE benchmark), as well as vision and speech.

Water drift and tidal rise make the use of bonding mechanisms beneficial for small benthopelagic or interstitial marine animals. Chemical adhesives for attachment are very common in molluscs; however, only a few cephalopods have glue producing organs. The family Idiosepiidae is characterized by an epithelial adhesive organ (AO) located on the posterior part of the dorsal mantle area. Previous morphological and histological studies described three non-glandular cell types (basal, interstitial and fusiform cells) and three glandular cell types (goblet, columnar and granular cells) containing protein and carbohydrate components. However, these studies provide different information about the nomenclature and characteristics of the cell types. The present ultrastructural analyses and a 3D reconstruction of the AO of Idiosepius pygmaeus and Idiosepius biserialis therefore serve to investigate the cell distribution, the fine structure of the cells and possible interactions between the cells.We found that basal cells form a continuous cell layer along the basal membrane, overlapped by the other epithelial cells. Embedded in microvilli-covered interstitial cells the glandular cells are more or less evenly distributed within the AO. Goblet and granular cells are solitary glandular cells without conspicuous morphological characteristics, whereas the columnar cells are arranged in dense aggregations of 5–15 cells. Each columnar cell is enclosed by a narrow supporting interstitial cell which contains dense longitudinal filament strands. The secretory process of the cells in the aggregation is synchronized. Each columnar cell aggregate bears approximately two ciliated sensory fusiform cells. The fusiform cells are connected to a neuronal network, aligned along the epithelium base.The results suggest that the bonding system is affected by two secretory cell types (granular and columnar cells). Both are similar in content, synthesis and secretory process but columnar cells are embedded in a particular cell environment. It is unclear in what way this arrangement is associated with the function of the AO. The neurons in several parts of the AO point to a neuronal control of the bonding mechanism. Comparisons with the AO cells of other cephalopods provide no indications for a morphological relationship between the adhesive systems.

Les réseaux d'assainissement subissent de nombreuses dégradations d'ordres structurelles. Les maîtres d'ouvrages estiment à 10 % du linéaire au moins, la longueur des collecteurs dont l'état de dégradation structurelle exige des travaux. Cet article concerne la restructuration d'ouvrages d'assainissement par lamelles composites. Afin d'appliquer cette technique de renforcement, une analyse par éléments finis est nécessaire pour connaître le mécanisme de ruine et donc les zones à renforcer. Une campagne expérimentale a permis de caractériser les différents matériaux (béton, colle et lamelles de carbone) et d'obtenir leurs lois de comportement. Une comparaison entre ovoïde témoin et ovoïde renforcé par lamelle de carbone pourra être effectuée.