Large language models have recently been shown to attain reasonable zero-shot generalization on a diverse set of tasks (Brown et al., 2020). It has been hypothesized that this is a consequence of implicit multitask learning in language model training . Can zero-shot generalization instead be directly induced by explicit multitask learning? To test this question at scale, we develop a system for easily mapping general natural language tasks into a human-readable prompted form. We convert a large set of supervised datasets, each with multiple prompts using varying natural language. These prompted datasets allow for benchmarking the ability of a model to perform completely unseen tasks specified in natural language. We fine-tune a pretrained encoder-decoder model on this multitask mixture covering a wide variety of tasks. The model attains strong zero-shot performance on several standard datasets, often outperforming models up to 16× its size. Further, our approach attains strong performance on a subset of tasks from the BIG-Bench benchmark, outperforming models up to 6× its size. All prompts and trained models are available at github.com/bigscience-workshop/promptsource/ and huggingface.co/bigscience/T0pp.

No abstract

An efficient method has been developed for the enantioselective synthesis of the aflatoxin system with multiple stereocenters via a sequence of organocatalytic Michael-acetalization-reduction-Nef reactions that proceed with high enantioselectivities (90-99% ee). The one-pot reaction sequence provides a facile entry to the aflatoxin system, including dihydroaflatoxin D, which includes a formal total synthesis of aflatoxin B. The first total synthesis of (-)- and (+)-microminutinin was also achieved via this protocol.

Accumulation of cytotoxic lipofuscin bisretinoids may contribute to atrophic age-related macular degeneration (AMD) pathogenesis. Retinal bisretinoid synthesis depends on the influx of serum all-trans-retinol (1) delivered via a tertiary retinol binding protein 4 (RBP4)–transthyretin (TTR)–retinol complex. We previously identified selective RBP4 antagonists that dissociate circulating RBP4–TTR–retinol complexes, reduce serum RBP4 levels, and inhibit bisretinoid synthesis in models of enhanced retinal lipofuscinogenesis. However, the release of TTR by selective RBP4 antagonists may be associated with TTR tetramer destabilization and, potentially, TTR amyloid formation. We describe herein the identification of bispecific RBP4 antagonist–TTR tetramer kinetic stabilizers. Standout analogue (±)-44 possesses suitable potency for both targets, significantly lowers mouse plasma RBP4 levels, and prevents TTR aggregation in a gel-based assay. This new class of bispecific compounds may be especially important as a therapy for dry AMD patients who have another common age-related comorbidity, senile systemic amyloidosis, a nongenetic disease associated with wild-type TTR misfolding.

A cascade organocatalysis has been developed for the enantioselective synthesis of a highly functionalized hexahydrophenanthrene-2-carbaldehyde containing five contiguous stereogenic centers with high diastereoselectivity and high enantioselectivity (>99% ee). The one-pot method comprises a cascade of organocatalytic Michael-Michael-Michael-aldol reactions of 2-methyl-1,5-dinitro-3-((E)-2-nitrovinyl)benzene and α,β-unsaturated aldehydes (e.g., cinnamaldehyde). The structure and absolute configuration of a product were confirmed by X-ray analysis of an appropriate derivative.