Search for: All records

Changes in global mean sea level (GMSL) during the late Cenozoic remain uncertain. We use a reconstruction of changes in δ¹⁸O of seawater to reconstruct GMSL since 4.5 million years ago (Ma) that accounts for temperature-driven changes in the δ¹⁸O of global ice sheets. Between 4.5 and 3 Ma, sea level highstands remained up to 20 m above present whereas the first lowstands below present suggest onset of Northern Hemisphere glaciation at 4 Ma. Intensification of global glaciation occurred from 3 Ma to 2.5 Ma, culminating in lowstands similar to the Last Glacial Maximum lowstand at 21,000 years ago and that reoccurred throughout much of the Pleistocene. We attribute the middle Pleistocene transition in ice sheet variability (1.2 Ma to 0.62 Ma) to modulation of 41-thousand-year (kyr) obliquity forcing by an increase in ~100-kyr CO₂variability. 

Abstract Proxy‐based reconstructions suggest that equilibrium changes in global mean sea surface temperature (ΔGMSST) are nearly equivalent to changes in mean ocean temperature (ΔMOT) on glacial‐interglacial timescales over the past 900,000 years. However, the underlying mechanisms responsible for this relationship remain poorly understood. Here we use simulations from Paleoclimate Modeling Intercomparison Project Phase 3 and 4 (PMIP3/4) to investigate equilibrium ΔMOT and its linkage to sea surface temperature changes between the Last Glacial Maximum (LGM, ∼21,000 years ago) and pre‐Industrial. Results show that PMIP3/4 simulations generally underestimate proxy‐based ΔMOT. Regression analysis reveals that LGM MOT is strongly modulated by mid‐latitude SST cooling, with the Southern Ocean having a greater influence compared to other oceanic regions, thus helping explain why models with similar ΔGMSSTs exhibit significantly different ΔMOTs. Additionally, we find a strong relationship between simulated Antarctic sea‐ice coverage and Southern Ocean SST changes, with implications for constraining sea‐ice reconstructions. 

Language agents have shown some ability to interact with an external environment, e.g., a virtual world such as ScienceWorld, to perform complex tasks, e.g., growing a plant, without the startup costs of reinforcement learning. While recent work, e.g., Reflexion, has demonstrated how such agents can also self-improve by adding a textual memory of ''hints'' learned from prior experience, such improvements have been limited both in size and scope. In contrast, our goal is a language agent that can robustly improve performance over time, including when both the task and environment are varied. Our approach is to have the agent learn a textual representation of how the world works (rather than just isolated hints), expressed as a memory of causal abstractions, to guide future decision-making. In experiments, we find CLIN is able to continually improve on repeated trials on the same task and environment, outperforming state-of-the-art reflective language agents like Reflexion by 23 points in ScienceWorld and 1.4 points in ALFWorld benchmarks. CLIN can also transfer its learning to new environments and tasks, enhancing performance by 21 points in ScienceWorld and 11 points in ALFWorld 

Abstract. We use a recent reconstruction of global mean sea surface temperature change relative to preindustrial (ΔGMSST) over the last 4.5 Myr together with independent proxy-based reconstructions of bottom water (ΔBWT) or deep-ocean (ΔDOT) temperatures to infer changes in mean ocean temperature (ΔMOT). Three independent lines of evidence show that the ratio of ΔMOT / ΔGMSST​​​​​​​, which is a measure of ocean heat storage efficiency (HSE), increased from ∼ 0.5 to ∼ 1 during the Middle Pleistocene Transition (MPT, 1.5–0.9 Ma), indicating an increase in ocean heat uptake (OHU) at this time. The first line of evidence comes from global climate models; the second from proxy-based reconstructions of ΔBWT, ΔMOT, and ΔGMSST; and the third from decomposing a global mean benthic δ18O stack (δ18Ob) into its temperature (δ18OT) and seawater (δ18Osw) components. Regarding the latter, we also find that further corrections in benthic δ18O, probably due to some combination of a long-term diagenetic overprint and to the carbonate ion effect, are necessary to explain reconstructed Pliocene sea-level highstands inferred from δ18Osw. We develop a simple conceptual model that invokes an increase in OHU and HSE during the MPT in response to changes in deep-ocean circulation driven largely by surface forcing of the Southern Ocean. Our model accounts for heat uptake and temperature in the non-polar upper ocean (0–2000 m) that is mainly due to wind-driven ventilation, while changes in the deeper ocean (> 2000 m) in both polar and non-polar waters occur due to high-latitude deepwater formation. We propose that deepwater formation was substantially reduced prior to the MPT, effectively decreasing HSE. We attribute these changes in deepwater formation across the MPT to long-term cooling which caused a change starting ∼ 1.5 Ma from a highly stratified Southern Ocean due to warm SSTs and reduced sea-ice extent to a Southern Ocean which, due to colder SSTs and increased sea-ice extent, had a greater vertical exchange of water masses. 

Planning in a text-based environment continues to be a significant challenge for AI systems. Recent approaches have utilized language models to predict planning domain definitions (e.g., PDDL) but have only been evaluated in closed-domain simulated environments. To address this, we present Proc2PDDL, the first dataset containing open-domain procedural texts paired with expert-annotated PDDL representations. Using this dataset, we evaluate the task of predicting domain actions (parameters, preconditions, and effects). We experiment with various large language models (LLMs) and prompting mechanisms, including a novel instruction inspired by the zone of proximal development (ZPD), which reconstructs the task as incremental basic skills. Our results demonstrate that Proc2PDDL is highly challenging for end-to-end LLMs, with GPT-3.5’s success rate close to 0% and GPT-4o’s 38%. With ZPD instructions, GPT-4o’s success rate increases to 45%, outperforming regular chain-of-thought prompting’s 34%. Our analysis systematically examines both syntactic and semantic errors, providing insights into the strengths and weaknesses of language models in generating domain-specific programs. 

Chrysothrix susquehannensis was originally described from vertical rocks faces at a single locality in southeastern Pennsylvania, U.S.A. Here we document new occurrences discovered as result of fieldwork and the revision of herbarium specimens. We show that C. susquehannensis, although rare, is widespread in the Appalachian Mountains of eastern North America, with a disjunction in New Mexico in southwestern North America.