The traditional process of discovering new antibiotics is slow, costly, and prone to failure. Addressing this, a team of researchers from Stanford University and McMaster University has developed an AI system called SyntheMol-RL, which designs novel antibiotic candidate molecules from scratch.

This AI optimizes for two critical factors: efficacy against resistant bacteria and realistic synthetic feasibility. Unlike previous AI models that might generate theoretically potent but unmanufacturable compounds, SyntheMol-RL integrates synthesis limitations from the outset.

Out of 79 molecules generated by the model, 13 showed high activity against Staphylococcus aureus, a problematic bacterium in clinical settings. Notably, a compound named synthecin successfully inhibited MRSA (Methicillin-resistant Staphylococcus aureus) infections in a mouse wound model. This demonstrates that while AI doesn't eliminate the need for experimental testing, it significantly accelerates the process by reducing the number of necessary trials, improving candidate quality, and exploring previously inaccessible chemical spaces.

Antimicrobial resistance causes over 1.27 million deaths annually, making the rapid discovery of new drugs imperative. SyntheMol-RL offers a new approach to designing solutions with greater precision from the beginning, rather than relying solely on trial and error, potentially marking a crucial turning point in the fight against superbugs.

👉 Read the full article on Gizmodo en Español

The primary obstacle to nuclear fusion energy has not been generating energy itself, but rather controlling the unpredictable and chaotic behavior of plasma. To tackle this challenge, the Oak Ridge National Laboratory (ORNL), under the U.S. Department of Energy, has deployed the exascale supercomputer Frontier.

Frontier, one of the world's most powerful machines capable of over a quintillion operations per second, combines its immense computational power with artificial intelligence to run thousands of advanced plasma simulations in seconds. This capability has transformed plasma behavior prediction, making it far more reliable than previous methods that required extensive calculation times and were prone to error, successfully reducing prediction errors by over 50%.

Understanding plasma has broader scientific implications beyond fusion reactors, including simulating supernovas, predicting solar flares, and better comprehending the interaction between cosmic radiation and Earth's magnetic field. For fusion reactors, maintaining plasma stably at extremely high temperatures of around 150 million degrees Celsius is crucial, and Frontier provides the ability to anticipate plasma turbulence and adjust the system before reactions destabilize.

This advancement signifies a paradigm shift in fusion research. Instead of merely reacting to problems, understanding and predicting them creates a genuine opportunity to master chaos. This is a critical step towards realizing fusion energy, indicating that the game is changing not through a technological breakthrough, but through enhanced predictive capabilities.

👉 Read the full article on Gizmodo en Español

Google has launched a Tensor Processing Unit (TPU) chip into Earth orbit as part of its ambitious Project Suncatcher, aiming to move a portion of future data center infrastructure off-planet. The experiment's goal is to verify if AI chips can withstand the harsh space environment, including vacuum, radiation, and thermal changes.

The first prototype, built in collaboration with Planet and launched aboard a SpaceX Falcon 9 rocket, has successfully established contact and is functioning as planned. Space offers significant advantages, such as access to near-constant solar energy, potentially up to eight times more than equivalent facilities on Earth.

Google envisions a future constellation of 81 satellites, each equipped with dozens of TPUs, connected by high-speed laser links to distribute large computational loads. The next critical step is planned for 2027, involving two computing satellites attempting laser communication while in motion.

However, realizing this vision requires a substantial reduction in launch costs (to approximately $200/kg by 2035) and a dramatic increase in the launch frequency of rockets like Starship. This experiment demonstrates Google's willingness to physically test futuristic ideas for AI infrastructure, marking a crucial step in exploring the potential for off-Earth data centers.

👉 Read the full article on Gizmodo en Español