Toggle light / dark theme

Once upon a time, Earth was barren. Everything changed when, somehow, out of the chemistry available early in our planet’s history, something started squirming – processing available matter to survive, to breed, to thrive.

What that something was, and when it first squirmed, have been burning questions that have puzzled humanity probably for as long as we’ve been able to ask “what am I?”

Now, a new study has found some answers – and life emerged surprisingly early.

Through analysis of gene families that were duplicated before the last universal common ancestor (LUCA), we estimate the date of the LUCA at approximately 4.2 billion years ago. Our reconstruction of the genome of the LUCA contains around 2,500 protein-encoding genes across 2.5 megabases, and we suggest that the LUCA was a complex anaerobic acetogen that lived within a pre-existing ecosystem.

A University of Maryland spinoff firm, Wave Engine Corporation, has created a simpler, more affordable jet propulsion system for drones.

The digitally controlled modern-day pulsejet engine features no moving parts and claims to offer major improvements in the cost reduction and rapid production of future jet-powered aircraft.

In March, the Baltimore-based company demonstrated the full flight capability of its J-1 engine on an Unmanned Aerial Vehicle (UAV).

One of the variables in TD algorithms is called reward prediction error (RPE), which is the difference between the discounted predicted reward at the current state and the discounted predicted reward plus the actual reward at the next state. TD learning theory gained traction in neuroscience once it was demonstrated that firing patterns of dopaminergic neurons in the ventral tegmental area (VTA) during reinforcement learning resemble RPE5,9,10.

Implementations of TD using computer algorithms are straightforward, but are more complex when they are mapped onto plausible neural machinery11,12,13. Current implementations of neural TD assume a set of temporal basis-functions13,14, which are activated by external cues. For this assumption to hold, each possible external cue must activate a separate set of basis-functions, and these basis-functions must tile all possible learnable intervals between stimulus and reward.

In this paper, we argue that these assumptions are unscalable and therefore implausible from a fundamental conceptual level, and demonstrate that some predictions of such algorithms are inconsistent with various established experimental results. Instead, we propose that temporal basis functions used by the brain are themselves learned. We call this theoretical framework: Flexibly Learned Errors in Expected Reward, or FLEX for short. We also propose a biophysically plausible implementation of FLEX, as a proof-of-concept model. We show that key predictions of this model are consistent with actual experimental results but are inconsistent with some key predictions of the TD theory.

A new superconducting compound offers a bridge to more practicals with a potentially attractive range of applications, according to new research. And the new material’s strange magnetic behavior recalls classics of decades ago—but this time in a material that’s already demonstrated its near-room-temperature bona fides.

Lanthanum hydrides—which combine atoms of the rare earth metal lanthanum with atoms of hydrogen—contain a range of superconducting materials of varying properties. One noteworthy material is lanthanum decahydride (LaH10), which boasts the world’s highest accepted superconducting transition temperature, at −23 °C. (The catch is that to achieve this feat, lanthanum decahydride must be subjected to 200 billion pascals of pressure.)

Now a different lanthanum hydride (La4H23) has revealed similar if not quite equally impressive superconductivity stats. (Its transition temperature is −168 °C at 122 billion Pa.) However, the new lanthanum hydride also has revealingly peculiar magnetic properties that suggest an unexpected family resemblance to the superstar of the superconductivity world, cuprates.

World’s smallest violin for AI execs.


Researchers are ringing the alarm bells, warning that companies like OpenAI and Google are rapidly running out of human-written training data for their AI models.

And without new training data, it’s likely the models won’t be able to get any smarter, a point of reckoning for the burgeoning AI industry.

“There is a serious bottleneck here,” AI researcher Tamay Besiroglu, lead author of a new paper to be presented at a conference this summer, told the Associated Press. “If you start hitting those constraints about how much data you have, then you can’t really scale up your models efficiently anymore.”