Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values, meaning that it is helpful, honest, and harmless. As an initial foray in this direction we study simple baseline techniques and evaluations, such as prompting. We find that the benefits from modest interventions increase with model size, generalize to a variety of alignment evaluations, and do not compromise the performance of large models. Next we investigate scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination, and ranked preference modeling. We find that ranked preference modeling performs much better than imitation learning, and often scales more favorably with model size. In contrast, binary discrimination typically performs and scales very similarly to imitation learning. Finally we study a 'preference model pre-training' stage of training, with the goal of improving sample efficiency when finetuning on human preferences.

No abstract

Chronic pain patients receiving opioid drugs are at risk for opioid-induced hyperalgesia (OIH), wherein opioid pain medication leads to a paradoxical pain state. OIH involves central sensitization of primary and secondary afferent neurons in the dorsal horn and dorsal root ganglion, similar to neuropathic pain. Gabapentin, a gamma-aminobutyric acid (GABA) analog anticonvulsant used to treat neuropathic pain, has been shown in animal models to reduce fentanyl hyperalgesia without compromising analgesic effect. Chronic pain patients have also exhibited lower opioid consumption and improved pain response when given gabapentin. However, few human studies investigating gabapentin use in OIH have been performed in recent years. In this review, we discuss the potential mechanisms that underlie OIH and provide a critical overview of interventional therapeutic strategies, especially the clinically-successful drug gabapentin, which may reduce OIH.

There is a continuing debate concerning “adjustments” to heart period variability [i.e., heart rate variability (HRV)] for the heart period [i.e., increases inter-beat-intervals (IBI)]. To date, such arguments have not seriously considered the impact a demographic variable, such as gender, can have on the association between HRV and the heart period. A prior meta-analysis showed women to have greater HRV compared to men despite having shorter IBI and higher heart rate (HR). Thus, it is plausible that men and women differ in the association between HRV and HR/IBI. Thus, the present study investigates the potential moderating effect of gender on the association between HRV and indices of cardiac chronotropy, including both HR and IBI. Data from 633 participants (339 women) were available for analysis. Cardiac measures were assessed during a 5-min baseline-resting period. HRV measures included the standard deviation of inter-beat-intervals, root mean square of successive differences, and autoregressive high frequency power. Moderation analyses showed gender significantly moderated the association between all HRV variables and both HR and IBI (each p < 0.05). However, results were not consistent when using recently recommended HRV variables “adjusted” for IBI. Overall, the current investigation provides data illustrating a differential association between HRV and the heart period based on gender. Substantial neurophysiological evidence support the current findings; women show greater sensitivity to acetylcholine compared to men. If women show greater sensitivity to acetylcholine, and acetylcholine increases HRV and the heart period, then the association between HRV and the heart period indeed should be stronger in women compared to men. Taken together, these data suggest that routine “adjustments” to HRV for the heart period are unjustified and problematic at best. As it relates to the application of future HRV research, it is imperative that researchers continue to consider the potential impact of gender.

The possession of a technique which permitted rapid estimations of cardiac output and which, demanding no intelligent cooperation, seemed especially suitable for use on ward patients, has permitted an extensive study on the action of common drugs on the heart and circulation in clinical conditions. This study contains about 450 estimations of cardiac output performed on 85 patients.Coincidentally with these estimations the action of drugs on pulse rate, on blood pressure, on respiratory rate and volume, and on metabolic rate was observed. Orthodiagrams and electrocardiograms were secured also. Therefore, certain parts of our study dealt with effects already well known.The results of such estimations have permitted the calculation of heart work, of peripheral resistance, of arteriovenous oxygen difference, and of the ratio of heart work to heart size, the latter a factor of decisive importance in our conception of cardiac stimulation and depression. Therefore our study demonstrates the effect of drugs on these functions also.Most of the drugs selected are commonly used in cases of cardiac and circulatory disease. We have studied the actions of digitalis, epinephrine, ephedrine, caffeine, theophylline, carbaminoylcholine, sodium nitrite, nitroglycerine, pitressin, quinidine, morphine and strychnine. We have studied the effects of drugs in those clinical conditions in which physicians are accustomed to employ them.But when suitable cases were not available the effects were studied in other conditions. Almost without exception our results support the general conceptions of drug action derived from animal experiments. PROCEDUREAll estimations were performed in the morning. The patients received no food after their evening meal and no water after midnight. They were taken from the ward in bed or in a wheel chair. An electrocardiogram and an orthodiagram were obtained first. Then the subjects lay down for at least 45 minutes. Duplicate estimations of cardiac output and metabolism were then made, together with repeated determinations of pulse rate, blood pressure, respiratory rate and volume.If the study concerned a rapidly acting drug, this was administered soon after the control estimations. The patient was watched until evidence of the drug's action became manifest objectively. Duplicate estimations of cardiac output and metabolism were then made, the purpose being to make these determinations at the height of action. Orthodiagrams and electrocardiograms were secured immediately afterward. Cardiac output was estimated by the method of Starr and Gamble (1), the analyses being performed by the katharometer method of Donal, Gamble and Shaw (2). Metabolism was estimated from samples of expired air drawn from a mixing bottle containing a fan.Respiratory volume was obtained by reading the spirometer at frequent intervals. Respiratory rate was counted repeatedly during the period of observations. It is well known that subjects breathing from a spirometer under 3 mm. H,O negative pressure, and through valves, tend to breathe...