While many studies have investigated the underpinnings of the mammalian motor system (i.e., the collection of neural networks that allow mammals to move in specific ways), some questions remain unanswered. One of these questions relates to the ways in which recurring or stable behaviors are maintained in the brain.

Some theories and research findings suggest that the neural activity underlying stable behaviors is itself very stable. Others, however, hinted at the possibility that the activity of individual motor neurons might change considerably over time, despite the production of similar behavioral patterns.

Researchers at Harvard University have recently tried to move toward the resolution of this long-standing debate, by observing the behavior and neural activity of rodents. Their findings, published in Nature Neuroscience, suggest that the activity of single neurons in brain regions associated with movement and physical behavioral patterns is highly stable over time.

Cornell University researchers have created an interface that allows users to handwrite and sketch within computer code—a challenge to conventional coding, which typically relies on typing.

The pen-based interface, called Notate, lets users of computational, digital notebooks open drawing canvases and handwrite diagrams within lines of traditional, digitized computer code.

Powered by a deep learning model, the interface bridges handwritten and textual programming contexts: notation in the handwritten diagram can reference textual code and vice versa. For instance, Notate recognizes handwritten programming symbols, like “n”, and then links them up to their typewritten equivalents.

Benchmarks orient AI. They encapsulate ideals and priorities that describe how the AI community should progress. When properly developed and analyzed, they allow the larger community to understand better and influence the direction of AI technology. The AI technology that has evolved the most in recent years is foundation models, highlighted by the advent of language models. A language model is essentially a box that accepts text and generates text. Despite their simplicity, these models may be customized (e.g., prompted or fine-tuned) to a wide range of downstream scenarios when trained on vast amounts of comprehensive data. However, there still needs to be more knowledge on the enormous surface of model capabilities, limits, and threats. They must benchmark language models holistically due to their fast growth, growing importance, and limited comprehension. But what does it mean to evaluate language models from a global perspective?

Language models are general-purpose text interfaces that may be used in various circumstances. And for each scenario, they may have a long list of requirements: models should be accurate, resilient, fair, and efficient, for example. In truth, the relative relevance of various desires is frequently determined by one’s perspective and ideals and the circumstance itself (e.g., inference efficiency might be of greater importance in mobile applications). They think that holistic assessment includes three components:

Will AI kill art? Not likely, says the artist Alexander Reben, who has been working with AI for years.

This CEO says generative AI is all flash, no substance — and ultimately, will fail to generate major new revenue streams for VCs’ billions.