William Wright, University of California, San Diego and Takaki Komiyama, University of California, San Diego

Every day, people are constantly learning and forming new memories. When you pick up a new hobby, try a recipe a friend recommended or read the latest world news, your brain stores many of these memories for years or decades.

But how does your brain achieve this incredible feat?

In our newly published research in the journal Science, we have identified some of the “rules” the brain uses to learn.

Learning in the brain

The human brain is made up of billions of nerve cells. These neurons conduct electrical pulses that carry information, much like how computers use binary code to carry data.

These electrical pulses are communicated with other neurons through connections between them called synapses. Individual neurons have branching extensions known as dendrites that can receive thousands of electrical inputs from other cells. Dendrites transmit these inputs to the main body of the neuron, where it then integrates all these signals to generate its own electrical pulses.

It is the collective activity of these electrical pulses across specific groups of neurons that form the representations of different information and experiences within the brain.

For decades, neuroscientists have thought that the brain learns by changing how neurons are connected to one another. As new information and experiences alter how neurons communicate with each other and change their collective activity patterns, some synaptic connections are made stronger while others are made weaker. This process of synaptic plasticity is what produces representations of new information and experiences within your brain.

In order for your brain to produce the correct representations during learning, however, the right synaptic connections must undergo the right changes at the right time. The “rules” that your brain uses to select which synapses to change during learning – what neuroscientists call the credit assignment problem – have remained largely unclear.

Defining the rules

We decided to monitor the activity of individual synaptic connections within the brain during learning to see whether we could identify activity patterns that determine which connections would get stronger or weaker.

To do this, we genetically encoded biosensors in the neurons of mice that would light up in response to synaptic and neural activity. We monitored this activity in real time as the mice learned a task that involved pressing a lever to a certain position after a sound cue in order to receive water.

We were surprised to find that the synapses on a neuron don’t all follow the same rule. For example, scientists have often thought that neurons follow what are called Hebbian rules, where neurons that consistently fire together, wire together. Instead, we saw that synapses on different locations of dendrites of the same neuron followed different rules to determine whether connections got stronger or weaker. Some synapses adhered to the traditional Hebbian rule where neurons that consistently fire together strengthen their connections. Other synapses did something different and completely independent of the neuron’s activity.

Our findings suggest that neurons, by simultaneously using two different sets of rules for learning across different groups of synapses, rather than a single uniform rule, can more precisely tune the different types of inputs they receive to appropriately represent new information in the brain.

In other words, by following different rules in the process of learning, neurons can multitask and perform multiple functions in parallel.

Future applications

This discovery provides a clearer understanding of how the connections between neurons change during learning. Given that most brain disorders, including degenerative and psychiatric conditions, involve some form of malfunctioning synapses, this has potentially important implications for human health and society.

For example, depression may develop from an excessive weakening of the synaptic connections within certain areas of the brain that make it harder to experience pleasure. By understanding how synaptic plasticity normally operates, scientists may be able to better understand what goes wrong in depression and then develop therapies to more effectively treat it.

These findings may also have implications for artificial intelligence. The artificial neural networks underlying AI have largely been inspired by how the brain works. However, the learning rules researchers use to update the connections within the networks and train the models are usually uniform and also not biologically plausible. Our research may provide insights into how to develop more biologically realistic AI models that are more efficient, have better performance, or both.

There is still a long way to go before we can use this information to develop new therapies for human brain disorders. While we found that synaptic connections on different groups of dendrites use different learning rules, we don’t know exactly why or how. In addition, while the ability of neurons to simultaneously use multiple learning methods increases their capacity to encode information, what other properties this may give them isn’t yet clear.

Future research will hopefully answer these questions and further our understanding of how the brain learns.

William Wright, Postdoctoral Scholar in Neurobiology, University of California, San Diego and Takaki Komiyama, Professor of Neurobiology, University of California, San Diego

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Manas Gaur, University of Maryland, Baltimore County

ChatGPT and other AI chatbots based on large language models are known to occasionally make things up, including scientific and legal citations. It turns out that measuring how accurate an AI model’s citations are is a good way of assessing the model’s reasoning abilities.

An AI model “reasons” by breaking down a query into steps and working through them in order. Think of how you learned to solve math word problems in school.

Ideally, to generate citations an AI model would understand the key concepts in a document, generate a ranked list of relevant papers to cite, and provide convincing reasoning for how each suggested paper supports the corresponding text. It would highlight specific connections between the text and the cited research, clarifying why each source matters.

The question is, can today’s models be trusted to make these connections and provide clear reasoning that justifies their source choices? The answer goes beyond citation accuracy to address how useful and accurate large language models are for any information retrieval purpose.

I’m a computer scientist. My colleagues − researchers from the AI Institute at the University of South Carolina, Ohio State University and University of Maryland Baltimore County − and I have developed the Reasons benchmark to test how well large language models can automatically generate research citations and provide understandable reasoning.

We used the benchmark to compare the performance of two popular AI reasoning models, DeepSeek’s R1 and OpenAI’s o1. Though DeepSeek made headlines with its stunning efficiency and cost-effectiveness, the Chinese upstart has a way to go to match OpenAI’s reasoning performance.

Sentence specific

The accuracy of citations has a lot to do with whether the AI model is reasoning about information at the sentence level rather than paragraph or document level. Paragraph-level and document-level citations can be thought of as throwing a large chunk of information into a large language model and asking it to provide many citations.

In this process, the large language model overgeneralizes and misinterprets individual sentences. The user ends up with citations that explain the whole paragraph or document, not the relatively fine-grained information in the sentence.

Further, reasoning suffers when you ask the large language model to read through an entire document. These models mostly rely on memorizing patterns that they typically are better at finding at the beginning and end of longer texts than in the middle. This makes it difficult for them to fully understand all the important information throughout a long document.

Large language models get confused because paragraphs and documents hold a lot of information, which affects citation generation and the reasoning process. Consequently, reasoning from large language models over paragraphs and documents becomes more like summarizing or paraphrasing.

The Reasons benchmark addresses this weakness by examining large language models’ citation generation and reasoning.

Testing citations and reasoning

Following the release of DeepSeek R1 in January 2025, we wanted to examine its accuracy in generating citations and its quality of reasoning and compare it with OpenAI’s o1 model. We created a paragraph that had sentences from different sources, gave the models individual sentences from this paragraph, and asked for citations and reasoning.

To start our test, we developed a small test bed of about 4,100 research articles around four key topics that are related to human brains and computer science: neurons and cognition, human-computer interaction, databases and artificial intelligence. We evaluated the models using two measures: F-1 score, which measures how accurate the provided citation is, and hallucination rate, which measures how sound the model’s reasoning is − that is, how often it produces an inaccurate or misleading response.

Our testing revealed significant performance differences between OpenAI o1 and DeepSeek R1 across different scientific domains. OpenAI’s o1 did well connecting information between different subjects, such as understanding how research on neurons and cognition connects to human-computer interaction and then to concepts in artificial intelligence, while remaining accurate. Its performance metrics consistently outpaced DeepSeek R1’s across all evaluation categories, especially in reducing hallucinations and successfully completing assigned tasks.

OpenAI o1 was better at combining ideas semantically, whereas R1 focused on making sure it generated a response for every attribution task, which in turn increased hallucination during reasoning. OpenAI o1 had a hallucination rate of approximately 35% compared with DeepSeek R1’s rate of nearly 85% in the attribution-based reasoning task.

In terms of accuracy and linguistic competence, OpenAI o1 scored about 0.65 on the F-1 test, which means it was right about 65% of the time when answering questions. It also scored about 0.70 on the BLEU test, which measures how well a language model writes in natural language. These are pretty good scores.

DeepSeek R1 scored lower, with about 0.35 on the F-1 test, meaning it was right about 35% of the time. However, its BLEU score was only about 0.2, which means its writing wasn’t as natural-sounding as OpenAI’s o1. This shows that o1 was better at presenting that information in clear, natural language.

OpenAI holds the advantage

On other benchmarks, DeepSeek R1 performs on par with OpenAI o1 on math, coding and scientific reasoning tasks. But the substantial difference on our benchmark suggests that o1 provides more reliable information, while R1 struggles with factual consistency.

Though we included other models in our comprehensive testing, the performance gap between o1 and R1 specifically highlights the current competitive landscape in AI development, with OpenAI’s offering maintaining a significant advantage in reasoning and knowledge integration capabilities.

These results suggest that OpenAI still has a leg up when it comes to source attribution and reasoning, possibly due to the nature and volume of the data it was trained on. The company recently announced its deep research tool, which can create reports with citations, ask follow-up questions and provide reasoning for the generated response.

The jury is still out on the tool’s value for researchers, but the caveat remains for everyone: Double-check all citations an AI gives you.

Manas Gaur, Assistant Professor of Computer Science and Electrical Engineering, University of Maryland, Baltimore County

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Breanne Hayes Haney, West Virginia University

Curious Kids is a series for children of all ages. If you have a question you’d like an expert to answer, send it to curiouskidsus@theconversation.com.

Are twins allergic to the same things? – Ella, age 7, Philadelphia

Allergies, whether spring sneezes due to pollen or trouble breathing triggered by a certain food, are caused by a combination of someone’s genes and the environment they live in.

The more things two people share, the higher their chances of being allergic to the same things. Twins are more likely to share allergies because of everything they have in common, but the story doesn’t end there.

I’m an allergist and immunologist, and part of my job is treating patients who have environmental, food or drug allergies. Allergies are really complex, and a lot of factors play a role in who gets them and who doesn’t.

What is an allergy?

Your immune system makes defense proteins called antibodies. Their job is to keep watch and attack any invading germs or other dangerous substances that get inside your body before they can make you sick.

An allergy happens when your body mistakes some usually harmless substance for a harmful intruder. These trigger molecules are called allergens.

The antibodies stick like suction cups to the allergens, setting off an immune system reaction. That process leads to common allergy symptoms: sneezing, a runny or stuffy nose, itchy, watery eyes, a cough. These symptoms can be annoying but minor.

Allergies can also cause a life-threatening reaction called anaphylaxis that requires immediate medical attention. For example, if someone ate a food they were allergic to, and then had throat swelling and a rash, that would be considered anaphylaxis.

The traditional treatment for anaphylaxis is a shot of the hormone epinephrine into the leg muscle. Allergy sufferers can also carry an auto-injector to give themselves an emergency shot in case of a life-threatening case of anaphylaxis. An epinephrine nasal spray is now available, too, which also works very quickly.

A person can be allergic to things outdoors, like grass or tree pollen and bee stings, or indoors, like pets and tiny bugs called dust mites that hang out in carpets and mattresses.

A person can also be allergic to foods. Food allergies affect 4% to 5% of the population. The most common are to cow’s milk, eggs, wheat, soy, peanuts, tree nuts, fish, shellfish and sesame. Sometimes people grow out of allergies, and sometimes they are lifelong.

Who gets allergies?

Each antibody has a specific target, which is why some people may only be allergic to one thing.

The antibodies responsible for allergies also take care of cleaning up any parasites that your body encounters. Thanks to modern medicine, people in the United States rarely deal with parasites. Those antibodies are still ready to fight, though, and sometimes they misfire at silly things, like pollen or food.

Hygiene and the environment around you can also play a role in how likely it is you’ll develop allergies. Basically, the more different kinds of bacteria that you’re exposed to earlier in life, the less likely you are to develop allergies. Studies have even shown that kids who grow up on farms, kids who have pets before the age of 5, and kids who have a lot of siblings are less likely to develop allergies. Being breastfed as a baby can also protect against having allergies.

Children who grow up in cities are more likely to develop allergies, probably due to air pollution, as are children who are around people who smoke.

Kids are less likely to develop food allergies if they try foods early in life rather than waiting until they are older. Sometimes a certain job can contribute to an adult developing environmental allergies. For example, hairdressers, bakers and car mechanics can develop allergies due to chemicals they work with.

Genetics can also play a huge role in why some people develop allergies. If a mom or dad has environmental or food allergies, their child is more likely to have allergies. Specifically for peanut allergies, if your parent or sibling is allergic to peanuts, you are seven times more likely to be allergic to peanuts!

Identical in allergies?

Back to the idea of twins: Yes, they can be allergic to the same things, but not always.

Researchers in Australia found that 60% to 70% of twins in one study both had environmental allergies, and identical twins were more likely to share allergies than fraternal (nonidentical) twins. Identical twins share 100% of their genes, while fraternal twins only share about 50% of their genes, the same as any pair of siblings.

A lot more research has been done on the genetics of food allergies. One peanut allergy study found that identical twins were more likely to both be allergic to peanuts than fraternal twins were.

So, twins can be allergic to the same things, and it’s more likely that they will be, based on their shared genetics and growing up together. But twins aren’t automatically allergic to the exact same things.

Imagine if two twins are separated at birth and raised in different homes: one on a farm with pets and one in the inner city. What if one’s parents smoke, and the others don’t? What if one lives with a lot of siblings and the other is an only child? They certainly could develop different allergies, or maybe not develop allergies at all.

Scientists like me are continuing to research allergies, and we hope to have more answers in the future.

Hello, curious kids! Do you have a question you’d like an expert to answer? Ask an adult to send your question to CuriousKidsUS@theconversation.com. Please tell us your name, age and the city where you live.

And since curiosity has no age limit – adults, let us know what you’re wondering, too. We won’t be able to answer every question, but we will do our best.

Breanne Hayes Haney, Allergy and Immunology Fellow-in-Training, School of Medicine, West Virginia University

This article is republished from The Conversation under a Creative Commons license. Read the original article.