AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles (e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities, and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.

BackgroundMycoparasitism, a lifestyle where one fungus is parasitic on another fungus, has special relevance when the prey is a plant pathogen, providing a strategy for biological control of pests for plant protection. Probably, the most studied biocontrol agents are species of the genus Hypocrea/Trichoderma.ResultsHere we report an analysis of the genome sequences of the two biocontrol species Trichoderma atroviride (teleomorph Hypocrea atroviridis) and Trichoderma virens (formerly Gliocladium virens, teleomorph Hypocrea virens), and a comparison with Trichoderma reesei (teleomorph Hypocrea jecorina). These three Trichoderma species display a remarkable conservation of gene order (78 to 96%), and a lack of active mobile elements probably due to repeat-induced point mutation. Several gene families are expanded in the two mycoparasitic species relative to T. reesei or other ascomycetes, and are overrepresented in non-syntenic genome regions. A phylogenetic analysis shows that T. reesei and T. virens are derived relative to T. atroviride. The mycoparasitism-specific genes thus arose in a common Trichoderma ancestor but were subsequently lost in T. reesei.ConclusionsThe data offer a better understanding of mycoparasitism, and thus enforce the development of improved biocontrol strains for efficient and environmentally friendly protection of plants.

Using an in vitro randomization and functional selection procedure, we have identified three novel classes of exonic splicing enhancers (ESEs) recognized by human SF2/ASF, SRp40, and SRp55, respectively. These ESEs are functional in splicing and are highly specific. For SF2/ASF and SRp55, in most cases, only the cognate SR protein can efficiently recognize an ESE and activate splicing. In contrast, the SRp40-selected ESEs can function with either SRp40 or SRp55, but not with SF2/ASF. UV cross-linking/competition and immunoprecipitation experiments showed that SR proteins recognize their cognate ESEs in nuclear extract by direct and specific binding. A motif search algorithm was used to derive consensus sequences for ESEs recognized by these SR proteins. Each SR protein yielded a distinct 5-to 7-nucleotide degenerate consensus. These three consensus sequences occur at higher frequencies in exons than in introns and may thus help define exon-intron boundaries. They occur in clusters within regions corresponding to naturally occurring, mapped ESEs. We conclude that a remarkably diverse set of sequences can function as ESEs. The degeneracy of these motifs is consistent with the fact that exonic enhancers evolved within extremely diverse protein coding sequences and are recognized by a small number of SR proteins that bind RNA with limited sequence specificity. Pre-mRNA splicing consists of two trans-esterification reactions, which occur in a large RNA-protein complex termed the spliceosome. This high-fidelity process requires precise recognition of the intron-exon borders by the spliceosome. The poorly conserved metazoan splice sites and branch site do not provide sufficient information for this recognition. Additional intron and exon sequences are often necessary for efficient and/or accurate splicing of many higher eukaryotic pre-mRNAs. The positive exon cis-acting elements, known as exonic splicing enhancers (ESEs), are often, though not always, found in a purine-rich context. A well-studied example is the ESE in the alternative exon M2 of the mouse IgM gene. This 73-nucleotide ESE is essential for splicing of the preceding intron between exons M1 and M2. The M2 ESE can also stimulate splicing of a heterologous regulated intron of the Drosophila doublesex (dsx) gene. Enhancer activity in the context of the IgM pre-mRNA could also be obtained by insertion of certain natural or synthetic purine-rich sequences in place of the natural ESE. However, deletion of the purine-rich sequences within the M2 ESE did not abolish its activity com-

Vascular endothelial growth factor (VEGF) is an important mediator of the intense angiogenesis which is characteristic of glioblastoma. While genetic manipulation of VEGF/VEGF receptor expression has previously been shown to inhibit glioblastoma growth, to date, no study has examined the efficacy of pharmacologic blockade of VEGF activity as a means to inhibit intracranial growth of human glioblastoma. Using intraperitoneal administration of a neutralizing anti-VEGF antibody, we demonstrate that inhibition of VEGF significantly prolongs survival in athymic rats inoculated in the basal ganglia with G55 human glioblastoma cells. Systemic anti-VEGF inhibition causes decreased tumor vascularity as well as a marked increase in tumor cell apoptosis in intracranial tumors. Although intracranial glioblastoma tumors grow more slowly as a consequence of anti-VEGF treatment, the histologic pattern of growth suggests that these tumors adapt to inhibition of angiogenesis by increased infiltration and cooption of the host vasculature.