Sparks of Artificial General Intelligence: Early experiments with GPT-4

Contrastive-Signal-Dependent Plasticity: Forward-Forward Learning of Spiking Neural Systems

“…Furthermore, the adaptive thresholds we presented in Equations 4, 16, and 22 are each calculated as a function of the total number of spikes across a layer ℓ at time t whereas it would be more bio-physically realistic to have each LIF unit within a layer adapt its own specific scalar threshold. 5 Finally, with respect to the event-driven forward-forward learning process itself, while we were able to desirably craft a simple update rule that operated at the spike-level (through a biologically plausible activation trace) there is still the drawback that the rule, much like its rated-coded sources of inspiration [25,54], requires positive and negative data samples to compute synaptic adjustments. In this study, we exploited the fact that labels were readily available to serve as top-down context signals which made synthesizing negative samples easy -all we needed to do was simply select one of the incorrect class labels to create a negative context.…”

“…As was further argued in [54], moving towards mortal computation would also likely entail challenging another important separation made in deep learning -the separation between inference and credit assignment. Specifically, in deep neural networks, including the more recent neural transformers [11,16,5] that drive large language models (typically pre-trained on gigantic text databases), are fit to training datasets in such a way that learning, carried out via the backpropagation of errors (backprop) algorithm [64], is treated as a separate computation distinct from the mechanisms in which information is propagated through the network itself. In contrast, an adaptive system that could take advantage of mortal computing will most likely need to engage in intertwined inference-and-learning [60,58,55,53], the framing that neurobiological learning and inference in the brain are not really two completely distinct, separate processes but rather complementary ones that depend on one another, the formulations of which are motivated and integrated with the properties of the underlying neural circuitry (and the hardware that instantiates it) in mind.…”

Automatic Generation of SBML Kinetic Models from Natural Language Texts Using GPT

“…Based on instructions such as “design an operon with a Tac promoter upstream of the gfp gene and place a BbsI recognition site before and after it”, corresponding DNA sequences can be created by LLMs. Moreover, GPT can draw pictures using scalable vector graphics (SVG) and TikZ codes [ 36 ]. In the future, LLMs may be able to draw biochemical network maps using Systems Biology Graphical Notation (SBGN) [ 37 ] or CADLIVE notation [ 9 , 11 ].…”

Assessing GPT-4’s Role as a Co-Collaborator in Scientific Research: A Case Study Analyzing Einstein’s Special Theory of Relativity

“…The advent of sophisticated artificial intelligence models, such as GPT-4, presents the opportunity to use these models as co-collaborators in scientific research. 1,2 GPT-4, and its GPT-3.5 predecessor, have exhibited human-level performance across various domains, including passing the US Medical Licensing Exams and the Multistate Bar Exam with remarkable accuracy. [3][4][5][6] These accomplishments suggest that GPT-4 could aid researchers in complex and controversial areas where human collaboration might be constrained or biased.…”

“…GPT-4, a highly advanced language model developed by OpenAI, is one such example. 1,2 This case study aims to assess GPT-4's performance in evaluating the mathematical consistency of Einstein's SRT equations and its ability to collaborate in discovering potential inconsistencies.…”