News Archive - Mindplex

Lila Sciences wants to build super-smart AI for science

A company called Lila Sciences aims to build scientific superintelligence to discover new science. The company wants to build artificial intelligence (AI) that handles tons of data and predicts things. This AI would also help scientists plan and run experiments. It would come up with ideas, and tests them in real labs. Lila wants to change how we discover things in life, chemical, and materials sciences.

The company raised $200 million from big investors like Flagship Pioneering and General Catalyst. Lila’s CEO Geoffrey von Maltzahn says that smart AI running the scientific method is a huge deal. The scientific method means guessing, testing, and learning. According to the company, Lila’s AI will do this at big scale and high speed, using robots to turn ideas into real results.

Noubar Afeyan, another founder, says humans still know little about the world’s rules. He thinks big experiments with AI can find new things. Lila’s platform, called Autonomous Science, mixes smart AI with robotic lab units. Humans guide it, but it runs experiments on its own. It’s already better than humans and other AI in some areas. For example, it made designed genetic medicines and found new antibodies. It also created cheap metals for green hydrogen and better materials for capturing carbon.

AI will design and run experiments

AI already digs up insights from data. Now, according to Lila’s team, it’s time for AI to design and run experiments.

The New York Times has covered the launch of Lila Science (unpaywalled copy). The article emphasizes that some big names have joined the company. George Church, a Harvard geneticist known for his pioneering research in genome sequencing and DNA synthesis, is Lila’s chief scientist. John Gregoire, a prominent researcher in new materials for clean energy, previously at Caltech, is Lila’s head of physical sciences research.

“It seems beyond anything I’m familiar with in scientific discovery,” says Nobel Prize in Chemistry winner David Baker as reported by The New York Times.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Growing giant structures in space with biology

DARPA wants to grow giant biological structures in space, New Scientist reports. According to a DARPA program manager, the combination of biology and mechanical engineering could allow to “manufacture at unprecedented sizes in space.”

SingularityHub has also covered the story.

In February, DARPA’s Biological Technologies Office has issued a call for ideas on “Large Bio-Mechanical Space Structures.”

Metabolic engineering speeds up growth in living things. Extremophiles are tough organisms that survive extreme conditions, like space. Biological self-assembly means living materials can form structures on their own. New design ideas from biology show how to make strong, useful shapes. DARPA thinks these tools can create structures over 500 meters long in microgravity. This could change how we build in space.

Imagine growing space elevator tethers, nets to catch space junk, or huge radio science tools. Biology could also make wings for space stations or patches to fix damage from tiny space rocks. These ideas are hard to do with normal methods. Growing them in space with biology might be easier and cheaper.

A mix of biology and engineering could build giant, useful things in space

DARPA wants to know how to make big, strong biological structures that stay stiff and stable in space.

This could cut down what we launch from Earth. Sending stuff to space costs a lot, but growing structures in space skips most of that. Biology naturally grows and assembles itself fast. But biology alone won’t work. It needs help from mechanical and structural engineering to make stiff, load-bearing shapes in microgravity.

Think of a tent: poles hold it up, and biology acts like the fabric cover. Add electronics, and it can do special jobs. The goal is to use less traditional material and grow more in space.

These structures could need feedstock delivered to them. Some organisms need air, so equipment must control atmosphere, pressure, and temperature. Others don’t need air but still need the right conditions to grow. The structures must grow in the right direction. They might need electronics or strong materials mixed in. DARPA hopes this mix of biology and engineering can build giant, useful things in space, saving money and opening new possibilities.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Zuchongzhi-3 breaks quantum speed records

Chinese scientists have built Zuchongzhi-3, a quantum computer with 105 qubits. This prototype also has 182 couplers, which connect qubits to share data. Scientists from the University of Science and Technology of China led the project.

They worked with experts from Shanghai, Henan, Jinan, Xidian University, and the Chinese Academy of Sciences. Zuchongzhi-3 runs a task called random quantum circuit sampling incredibly fast. It beats the world’s top supercomputer by 1015 times and Google’s latest result by a million times. The journal Physical Review Letters published this as its cover story.

Quantum supremacy means a quantum computer can do things regular computers can’t. In 2019, Google’s Sycamore, with 53 qubits, finished a task in 200 seconds. A regular supercomputer would take 10,000 years. But in 2023, Chinese researchers used smarter regular computer methods. They cut that time to 14 seconds with many GPUs. A supercomputer called Frontier did it in 1.6 seconds. This ended Google’s supremacy claim. Then, Chinese researchers achieved supremacy with Jiuzhang in 2020 using light. In 2021, Zuchongzhi-2, with 66 qubits, did it too. In 2023, Jiuzhang-3, with 255 light qubits, beat supercomputers by 1016 times. Google’s updated Sycamore, with 67 qubits, also showed supremacy in 2024.

Zuchongzhi-3 is the top quantum machine in its class

Zuchongzhi-3 improves on Zuchongzhi-2. It lasts 72 microseconds before losing coherence. It runs single-qubit gates, or simple commands, with 99.90% accuracy. Two-qubit gates, which link qubits, hit 99.62% accuracy. Readouts, or results, score 99.13% accuracy. The team tested it with an 83-qubit, 32-layer sampling task. It outran the best regular algorithm by 15 orders of magnitude. It also beat Google’s 2024 result by 6 orders of magnitude. This makes it the top quantum machine in its class.

The scientists now explore quantum error correction with a 2D grid setup for better qubit connections. This helps data move faster. They work on a surface code to catch errors, aiming to expand it. Zuchongzhi-3’s success earned praise. Experts call it a major leap from Zuchongzhi-2. Physics Magazine highlighted its importance in a special article.

A Forbes headline reads “Quantum Singularity Ahead? China’s Zuchongzhi-3 Reshapes Quantum Race.”

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Robots learn from human nudges

Imagine a robot helping you wash dishes. You ask it to grab a soapy bowl, but it misses slightly. Researchers from MIT and NVIDIA found a way to fix this. You can point to the bowl on a screen, trace a path for the robot, or nudge its arm. This method helps the robot understand what you want without needing to retrain its brain.

Normally, robots use a machine-learning model, which is a system that learns from examples. Old methods need new examples and retraining if the robot fails. This takes time and skill. The new framework skips that. It uses real-time feedback, like a nudge, to guide the robot. The robot then picks a valid action, or movement, that matches your goal.

Tests showed this method works better. It beat another approach by 21 percent. The framework could help factory robots do home tasks, even in new places. People shouldn’t need to train robots with data, says Felix Yanwei Wang, the lead researcher. He’s a graduate student in electrical engineering and computer science. His team includes experts from MIT and NVIDIA. They’ll share this at a big robot conference.

Real time feedback

Robots often learn a policy, or set of rules, from a generative AI model. This model makes valid paths for the robot. But these paths don’t always fit what users want. Retraining fixes this, but it’s slow. The new method lets users adjust the robot right away. It offers three ways: pointing, tracing, or nudging. Nudging is best because it keeps all the details clear.

To avoid mistakes, like knocking things over, the method uses sampling. This picks a safe action close to what you want. It blends your input with the robot’s learned rules. Tests in a toy kitchen proved it works well. If the robot picks the wrong bowl, you can fix it instantly. Over time, it could learn from these fixes for next time. The team wants to make it faster and test it in new places. This could make robots easier to use at home.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Chinese AI agent Manus surprises the tech world

A new artificial intelligence (AI) agent called Manus tackles hard real-world jobs, South China Morning Post reports (unpaywalled copy).

The agent stays private for now, offering access only to invited users through a web preview. It shows off its skills in a video on its website, manus.im. The video proves Manus can make a website step by step.

“Manus, derived from the Latin word for ‘hand’, is a general AI agent that turns your thoughts into actions,” reads the manus.im website.

The website brags about other cool tricks. It plans trips to places like Japan, and studies Tesla’s stock in detail. It builds courses for teachers. Manus compares insurance plans, and even helps businesses find suppliers. These jobs show Manus handles big tasks well. Its makers say Manus beats OpenAI’s Deep Research. They measure this with the GAIA benchmark.

People love the Manus demo video. It hit X on a Wednesday night. By Thursday noon, over 200,000 folks watched it. Many left comments begging for invite codes. This buzz reminds some of DeepSeek, another Chinese AI that got big fast. Manus keeps quiet about its makers and tech details, though. That mystery fuels the hype.

An AI agent that thinks, plans, and executes tasks independently

The makers of Manus see it as a game-changer. General-purpose AI agents like it interact with the world, gathering info and finishing tasks to meet goals. This style could shape AI’s future.

A Forbes article comments that Manus “changes everything.” Manus “represents something entirely different – it is not just another model. It is an agent, an AI system that thinks, plans, and executes tasks independently, capable of navigating the real world as seamlessly as a human intern with an unlimited attention span.”

In other AI news from China, Alibaba unveiled its latest artificial intelligence reasoning model, CNN reports. According to Alibaba, its capabilities beat those of rival models from OpenAI and startup DeepSeek. CNN also notes that the release of Alibaba’s new AI model comes a day after the launch of Manus.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

AI helps create a new lung disease drug

A company called Insilico Medicine made a new drug named Rentosertib. This drug helps people with a lung disease called Idiopathic Pulmonary Fibrosis, or IPF. IPF is a serious condition that makes breathing harder over time. About 5 million people have it worldwide. Most live only 3 to 4 years after diagnosis. Current drugs slow it down but don’t fix it. Insilico wants Rentosertib to do more.

Insilico used artificial intelligence, or AI, to create Rentosertib. Their AI system, Pharma.AI, found a target called TNIK. TNIK is a protein linked to IPF. Another AI tool, Chemistry42, designed Rentosertib to hit that target. This process took just 18 months, which is super fast. Normally, making drugs takes years. The United States Adopted Names Council gave Rentosertib its official name recently.

AI can speed up medicine-making

Rentosertib is a product of generative AI. It’s the first drug where AI found both the target and the medicine. Tests show it works well so far. In early trials with healthy people, Rentosertib was safe and easy to take. These Phase I trials checked pharmacokinetics, or how the body handles the drug. Results looked good in New Zealand and China.

Next, Insilico tested Rentosertib on IPF patients in a Phase IIa trial. This 12-week study showed it’s safe at all doses. Patients taking higher doses saw better lung function, measured by forced vital capacity, or FVC. At 60mg, patients improved FVC by 98.4 mL. The placebo group, with no real drug, lost 62.3 mL. Patients also felt better, with less coughing.

Rentosertib had mild side effects and no big problems. Insilico plans bigger trials now. They hope it becomes the first AI-made drug to reach patients. The name Rentosertib honors Feng Ren, a key scientist at Insilico. This drug could change how we fight IPF. It shows AI can speed up medicine-making. Insilico aims to help people with IPF and prove AI’s power in science.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

A new kind of computer powered by human brain cells

Scientists Australian company Cortical Labs built a new type of computer called the CL1, New Atlas reports.

This computer uses human brain cells mixed with silicon hardware. This computer is the first of its kind available to buy. Cortical Labs announced CL1 in Barcelona on March 2, 2025. Other news outlets like Tom’s Hardware, TechSpot, and ABC New Australia have covered the launch of the CL1.

The CL1 runs on Synthetic Biological Intelligence, or SBI for short. SBI means combining living cells with tech to make a smart system.

Regular computers use only silicon chips. The CL1, though, grows human brain cells on these chips. These cells form neural networks. This setup lets the computer learn fast and adapt to new tasks. The company claims it outdoes traditional artificial intelligence, or AI, in speed and flexibility.

A basic but effective brain model

The CL1 needs less power than typical AI systems. Our brains use about 20 watts, like a dim light bulb. Big AI setups can use millions of watts. The CL1’s low energy use makes it more sustainable. It could change how we do science and medicine. Researchers might use it to test drugs or study the brain. It’s not fully ready yet, but people can buy it starting later in 2025.

Cortical Labs first got attention in 2022. Back then, they made a system with 800,000 brain cells play Pong. That showed how powerful this idea could be. Now, the CL1 takes it further. The company wants to find the simplest brain-like system possible. They call this the Minimal Viable Brain. It’s about making a basic but effective brain model.

This tech mixes biology and machines in a new way. Some wonder about its limits and ethics. Still, the CL1 offers a fresh path for computing. It blends human cells with silicon to create something unique. Scientists and companies will soon explore its full potential. This could lead to breakthroughs we can’t even imagine yet.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Other interesting news for 03.07.25

This post collects links to other interesting news that we have considered but not included in our regular news this week.

An ancient RNA-guided system could simplify delivery of gene editing therapies

In Early-Phase Pancreatic Cancer Clinical Trial, Investigational mRNA Vaccine Induces Sustained Immune Activity in Small Patient Group

Machine learning on blockchain: A new approach to engineering computational security

Next top model: Competition-based AI study aims to lower data center costs

Multi-photon 3D nanoprinting: Advances, challenges and the road ahead

Unveiled: New Vision for American Science and Technology (VAST)

ChatGPT on the couch? How to calm a stressed-out AI

HKU researchers unveil neuromorphic exposure control system to improve machine vision in extreme lighting environments

More headlines

“Vortion,” a new magnetic state able to mimic neuronal synapses

How the brain turns sound into conversation: A new study uncovers the neural pathways of communication

Unlocking the future: How machine learning transforms big data analytics

Study provides novel insights into GenAI feedback

Scientists unravel the spiraling secrets of magnetic materials for next-generation electronics

From classical hydrodynamics to quantum hydrodynamics and back again – how the Navier-Stokes equations describe quantum systems

SNU researchers develop world’s first technology to observe atomic structural changes of nanoparticles in 3D

AI and Adaptive Optics propel free-space quantum communication into a new era

Planetary science: More potential loc ations for ice on Moon

Cold atoms on a chip

Controlling electrons in molecules at ultrafast timescales

Butterfly-inspired 4D printing of smart hydrogels enables precise micro-nano deformation

A method to purify proteins with light: Gentler and more efficient than previous procedures

The chances of anything coming from Mars

SLAC scientists created the most powerful ultrashort electron beam in the world

Scientists design experimental protein booster for rare genetic diseases

Sneaky clocks: uncovering Einstein’s relativity in an interacting atomic playground

Pusan National University researchers developed an advanced AI model for accelerating therapeutic gene target discovery

Feeling is believing: Bionic hand “knows” what it’s touching, grasps like a human

Precision cancer treatment using magnet -guided, heat-activated nanoparticles

Breakthrough in noninvasive monitoring of molecular processes in deep tissue

Tiny “rhinoceros beetle” robot does micro-scale manipulation in extreme conditions

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Reinforcement learning pioneers win ACM Turing award

Andrew Barto and Richard Sutton earned the 2024 ACM A.M. Turing Award. This prize honors their work on reinforcement learning, a key part of artificial intelligence (AI) technology.

Barto, a professor emeritus at the University of Massachusetts Amherst, teamed up with Sutton, a professor at the University of Alberta. They started in working together in the 1980s. Reinforcement learning, or RL, teaches machines to act better by using rewards – like treats for a pet.

In RL, an agent is something that sees and does things, like a robot or program. Rewards tell the agent if its actions are good or bad. Barto and Sutton built this idea using math called Markov decision processes, or MDPs. MDPs help agents decide in random settings, aiming to get the most rewards over time. Unlike regular MDPs, RL lets the agent learn without knowing everything upfront. This makes RL useful for many problems.

They created key tools for RL, like temporal difference learning. This method helps predict rewards faster and better. They also made policy-gradient methods and used neural networks to store what the agent learns. Their work mixed learning with planning, showing how agents can study their world to make smarter choices.

The Nobel Prize of Computing

Their 1998 book, Reinforcement Learning: An Introduction, taught thousands of people about RL and still guides research today. Years later, RL joined with deep learning to do amazing things.

With RL, robots learn to move objects, like solving a Rubik’s Cube. RL improves internet ads, chip designs, and even supply chains. Plus, it helps explain how human brains work with dopamine, a chemical for rewards. Barto and Sutton’s ideas keep growing, making tech and science better. The Turing Award, often called the “Nobel Prize of Computing,” gives them $1 million from Google. Their work proves old ideas can spark huge changes, exciting everyone in tech.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Yet another thrilling Starship test flight

SpaceX launched its eighth Starship test flight from Starbase in Texas, showcasing remarkable engineering prowess. The Super Heavy booster ignited all 33 Raptor engines and propelled the massive rocket skyward. About two and a half minutes into the flight, Super Heavy executed a hot-staging maneuver. It shut down 30 engines, leaving three burning as Ship 34 separated cleanly.

The booster then performed a boostback burn, guiding itself back toward the launch tower. With pinpoint accuracy, the tower’s chopstick arms caught Super Heavy, marking another triumph for SpaceX’s reusable rocket vision. Cheers likely echoed through mission control as the team nailed this complex feat again. Super Heavy’s catch marked the third time SpaceX achieved this milestone, reinforcing confidence in the system.

A bump in the road

Ship 34, the upper stage, began its ascent with high hopes. It aimed to deploy four mock Starlink satellites and test its heat shield. Six Raptor engines fired up, pushing it higher into the sky. However, trouble struck around eight minutes in. Four engines shut down unexpectedly, one after another. Ship lost attitude control, tumbling as telemetry signals faded. SpaceX confirmed the vehicle broke apart over the Bahamas, ending its journey prematurely. Observers in Florida, Jamaica, and the Turks and Caicos Islands witnessed the dramatic breakup.

Despite this setback, the mission wasn’t a total loss. The team gathered critical data to refine future flights, proving every test pushes them closer to their goals.

Ship 34’s failure mirrored challenges from Flight 7, hinting at persistent engine or propellant issues. Yet, SpaceX views these hiccups as stepping stones. Elon Musk once called such events “barely a bump in the road,” reflecting an optimistic outlook. The test highlighted Starship’s potential to revolutionize space travel. It carried the Block 2 design further, testing upgrades vital for lunar and Mars missions.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.

Exciting News! Our Mobile App is Here!

Welcome Back

No account? Create One

Join

Already have an account? Sign in

forgot password

AI will design and run experiments

A mix of biology and engineering could build giant, useful things in space

Zuchongzhi-3 is the top quantum machine in its class

Real time feedback

An AI agent that thinks, plans, and executes tasks independently

AI can speed up medicine-making

A basic but effective brain model

More headlines

The Nobel Prize of Computing

A bump in the road