I’ve been using OpenAI’s “Deep Research” capability for the last several days, trying to evaluate how it can be used to improve the research and reporting we do for our network.

Deep Research is an AI agent. It uses reasoning to synthesize large amounts of online information and conduct complex multi-step research tasks. The goal of this agentic AI system is to enable researchers to accomplish in minutes what it would take an experienced researcher to do in days.

The way it works is you give it a prompt, it will ask some clarifying questions, then will set about finding, analyzing and syntheisizing from as many online sources as it needs to to create a comprehensive report.

The model behind Deep Research is OpenAI’s latest 03, which is scoring incredibly high on a wide range of benchmarks for AI including one known as “Humanity’s Last Exam” which evaluates AI across extremely hard expert-level questions. When using just the o3 model, scores on the exam come in at 13 percent accuracy. When using Deep Research (which includes browsing and reasoning) scores on this test come in at 26%, a dramatic improvement.

(It is amazing to think of how fast accuracy is improving, CPT-4o is still a new and very capable model and Deep Research just blows it away in capability).

The way Deep Research works is based on extensive end-to-end reinforcement learning on hard browsing and reasoning tasks across multiple domains. It has been trained to plan and execute multi-step activities to fid the data it needs and to backtrack and reject data as needed, reacting to real-time info when necessary. It can also browse uploaded files, can plot and interest over graphs, use python, and cite sources.

Here is an example of a simple research project that just took 10 minutes to complete. Deep Research was asked:

“Find surprising results about brain research, something that would surprise the average person”

It responded with a clarifying question, asking me:

“Would you like surprising neuroscience findings across all areas, or do you have specific interests (e.g., memory, consciousness, neuroplasticity, AI-brain interfaces, mental health, perception, etc.)? Also, should I prioritize very recent research (last 1-2 years) or include older groundbreaking studies as well?”

I just replied that all topics were good.

Response from Deep Research was;

I’ll gather more surprising and counterintuitive neuroscience findings across various topics, including cognition, perception, memory, consciousness, and brain function. I will focus on recent research that challenges conventional wisdom and would be unexpected to the average person. I’ll update you once the findings are ready.

Surprising Findings in Recent Neuroscience Research

Neuroscience is continually uncovering findings that defy expectations and challenge what we thought we knew about the brain. Below are several recent discoveries across cognition, perception, memory, consciousness, and brain anatomy that are particularly unexpected. Each finding is summarized with a link to the original research.

Mice Resist Optimal Strategies in Learning Tasks

Surprise: Even after learning the “best” way to get rewards, mice deliberately don’t stick to it. In an MIT study, mice were trained on a simple reversal learning game (turn a wheel left or right for reward, then the rule switches). The expected outcome was that mice would adopt a consistent win-stay/lose-shift strategy (like humans do), but surprisingly, they never fully abandoned random exploring. Even when one side yielded a reward 100% of the time, mice would occasionally try the other side (Study decodes surprising approach mice take in learning | Brain and Cognitive Sciences). Researchers suggest mice might be intentionally testing for changes in their environment rather than mindlessly exploiting a known rule – a counterintuitive insight into animal cognition that challenges the notion that lab animals always maximize reward.

Humans May Sense Earth’s Magnetic Field Unconsciously

Surprise: Humans, long assumed to lack a magnetic sense, show brain responses to Earth-strength magnetic fields. A Caltech/Tokyo study placed people in a dark, electrically shielded chamber and subtly rotated magnetic fields around them. Participants felt nothing, but EEG recordings revealed distinct changes in their brain’s alpha waves in response to the magnetic field shifts. This provides the first solid evidence of human magnetoreception – an unconscious sixth (or perhaps eleventh) sense in humans (Evidence for a Human Geomagnetic Sense | ScienceDaily). It upends the conventional wisdom that only animals like birds or bees can detect magnetic fields, suggesting our brains have a hidden sensory capability that the average person isn’t aware of.

Non-Brain Cells Exhibit Memory-Like Behavior

Surprise: Memory might not be exclusive to neurons – even cells outside the brain can “learn.” Neuroscientists found that human kidney and other non-neural cells can store information from past stimuli in a way analogous to memory. When exposed to repeated, spaced chemical signals in a lab (mimicking how neurons receive bursts of neurotransmitters during learning), these cells activated a so-called “memory gene” and responded more strongly to repeated pulses than to continuous stimulation (Cells Beyond the Brain Show Surprising Memory Abilities – Neuroscience News). In other words, they remembered the pattern of stimulation. This challenges the old view that only the brain encodes memories – suggesting that the massed vs. spaced learning effect (cramming vs. spacing out learning) may be a fundamental cellular property, not just a brain function. The findings, published in Nature Communications, open the door to the idea that organs could “remember” past exposures, a concept that would surprise most people.

Signs of Conscious Activity Detected Near Death

Surprise: The brain may exhibit a flurry of conscious-like activity at the brink of death, even when it was thought to be shutting down. In a study of cardiac arrest patients who received CPR, about 20% of survivors later reported lucid near-death experiences – such as perceiving themselves outside their body or vividly reviewing their life – despite having no discernible consciousness at the time. Remarkably, EEG recordings during CPR revealed brief bursts of organized brain waves (gamma, alpha, theta, etc.) up to an hour into resuscitation (Lucid dying: Patients recall death experiences during CPR | ScienceDaily). These brain wave patterns are typically associated with conscious thought and memory retrieval, suggesting the mind can flicker back on momentarily even when the heart has stopped. Such findings, from research led by NYU and published in 2022, challenge the assumption that the brain flatlines immediately during clinical death. They imply the “door to consciousness” may remain ajar longer than expected, offering a scientific footing for reports of near-death experiences that would astonish most people.

New “Protective Shield” Layer Discovered in the Brain

Surprise: Scientists have identified an entirely new anatomical structure in the brain – after we thought we had mapped them all. In 2023, researchers reported in Science the discovery of a thin, sheet-like layer within the meninges (the membranes covering the brain) that had never been noticed before. This layer, described as a “protective shield”, partitions the space where cerebrospinal fluid flows (the subarachnoid space) into two compartments (Newfound ‘protective shield’ in the brain is like a watchtower for immune cells | Live Science). It appears to act as a waste filter and immune sentry, helping clear out metabolites while letting immune cells surveil the brain for infection. The finding was unexpected because it reveals a previously hidden component of brain anatomy performing critical maintenance and monitoring roles. It overturns the assumption that we knew all the brain’s parts, reminding us that even our basic anatomy can still hold surprises.

Each of these discoveries highlights how recent neuroscience is challenging conventional wisdom – from animal behavior to human perception, from the cellular basis of memory to the final moments of consciousness, and even the brain’s physical structure. These surprising findings not only fascinate, they reshape our understanding of how the brain and mind truly work.