Carboncopies Connect - December 2025¶
Your donation helps turn the long-term vision of whole brain emulation into real scientific progress. Contributions to the Carboncopies Foundation directly support our research and infrastructure, funding critical hardware and software, enabling high-fidelity brain modeling experiments, and expanding our educational and public-outreach efforts. As a registered 501(c)(3) nonprofit dedicated to advancing public education and scientific research, Carboncopies ensures that every contribution is used to move this field forward. Donations are tax-deductible in the United States (EIN: 81-4083638).
If you believe that understanding and emulating the human brain is one of the most important scientific challenges of our time, we invite you to be part of this work. Please visit https://carboncopies.org/Donate to support our mission.
In this Newsletter¶
- The Carboncopies Foundation, a Re-introduction to our Effort
- It’s Time for Brain Emulation - Taking the Pulse of the Times
- Aspirational Neuroscience at SfN2025 - Memory Decoding Awards
- 2025 Brain Emulation Challenge Workshop: Functionalizing Brain Data, Ground-Truthing, and the Role of Artificial Data
- Memory Decoding Journal Club: Structure to Function Mouse Connectomics
- Consciousness & Subjective Experience Journal Club: Why We Experience Subjectively
- Brain Emulation Challenge
- High-Performance BrainGenix
- Neuroscience of WBE: Profound Implications
- Fidelity Tradeoffs in Whole Brain Emulation
- Bottleneck of BCI
- Janelia Publishes Complete Drosophila Central Nervous System
- Seminal Results of MICrONS
- Modeling the Fly Brain
- Predicting Fly Vision From Wiring
CARBONCOPIES HIGHLIGHTS¶
If it’s been a while since you last heard from us, welcome back! If you’re new, we’re glad you’re here.
The Carboncopies Foundation is a nonprofit organization dedicated to advancing whole brain emulation (WBE) and the neurotechnologies needed to make it possible. Our mission remains as ambitious as ever: to enable the scientific understanding and technological development required to emulate entire brains, with profound implications for medicine, neuroscience, and humanity’s long-term future.
What makes Carboncopies unique is that we focus on the entire pipeline of brain emulation. We are still the only organization dedicated specifically to whole brain emulation as an end-to-end challenge – from data acquisition and modeling, to validation, ethics, and human impact. Because of this, our work spans research, education, and community-building, all tightly interconnected.
Building and Supporting a Global Community¶
Community has always been central to Carboncopies. We welcome anyone interested in brain emulation and related neurotechnology, make it easy to get involved, and actively build international connections. We support new projects, foster collaborations, and keep our community informed about both scientific progress and open challenges.
A key part of this effort is our regular workshops, where we invite leading experts to discuss the most pressing topics in brain emulation, ongoing Carboncopies projects, and specific technical or conceptual bottlenecks facing the field.
We also run two regular journal clubs:
- The Memory Decoding Journal Club, a collaboration with the Brain Preservation Foundation’s Aspirational Neuroscience campaign. This group focuses on the challenge of decoding non-trivial memories from static brain data (e.g. a synaptic connectome). We review papers nominated for Memory Decoding awards and host a satellite event at the Society for Neuroscience annual meeting, where four outstanding papers are each awarded $25,000 toward an eventual $100,000 grand prize for the first true demonstration of non-trivial memory decoding from static data.
- The Consciousness & Subjective Experience Journal Club, which addresses foundational questions that are especially important for the future human applications of whole brain emulation.
Education and Outreach¶
Education is another pillar of our work. Carboncopies gives lectures at universities, supports the formation of student organizations focused on brain emulation and neurotechnology, and presents at academic and non-academic conferences and workshops, including the annual meeting of the Society for Neuroscience. We publish academic papers and maintain an active public presence through talks, writing, and social media to help broaden awareness of the field.
Research: From Roadmaps to Challenges¶
Our current research efforts center on three major projects.
First, we are developing an Ethics Framework for Whole Brain Emulation. We began this work in part because we observed how ethical considerations in AI were often addressed only after technologies were already deployed. Brain emulation, by contrast, will inevitably intersect with medicine and patient care. Our framework examines the responsibilities of researchers over time, equitable access and informed consent for patients, possible outcomes of the technology, and its broader societal implications.
Second, we continuously update the Roadmap toward Whole Brain Emulation. Over the past decade, the landscape has changed dramatically. While collecting high-resolution functional data from living brains remains difficult, advances in structural data acquisition – especially high-throughput electron microscopy – have been extraordinary. We now have complete connectomes for organisms such as Drosophila and zebrafish, and substantial datasets for mouse, human, and other brains, down to individual synapses and even synaptic vesicles. As a result, the dominant challenge has shifted: the bottleneck is no longer data collection alone, but how to build and validate working models from that data.
Which brings us to our largest research initiative: the Brain Emulation Challenge.
Despite the explosion of data, there is currently no validated working model of any brain, not even for simple organisms with a few hundred neurons. To address this, we are creating a challenge framework inspired by the role that standardized datasets and benchmarks – such as ImageNet and Kaggle competitions – played in accelerating progress in artificial intelligence. In the Brain Emulation Challenge, we generate synthetic brains with known ground-truth, along with corresponding datasets. Researchers can apply their analysis and reconstruction methods and receive explicit validation, revealing where and why their approaches succeed or fail. Our goal is to bootstrap meaningful validation and catalyze rapid progress in brain reconstruction and emulation.
(High resolution PDF available here.)
CiteRef: Koene, R.A., Liao, T., Jose, J., Mendoza, T., Behler, A., Shah, J., and Huynh, A., “The Brain Emulation Challenge: A Synthetic Data Platform to Evaluate Reconstruction Methods.” Poster presented at Society for Neuroscience Annual Meeting, San Diego, CA, November 2025, Program number PSTR197.18.
An International, Interdisciplinary Effort¶
Although Carboncopies is headquartered in California, we are a highly international organization, with nearly 100 team members across North America, Europe, and beyond. Our community includes neuroscientists, programmers, AI and machine learning researchers, philosophers working on consciousness and ethics, writers, and educators. We regularly host interns and have supported multiple Ph.D. dissertations in collaboration with universities, with students jointly mentored by academic advisors and Carboncopies researchers. These experiences have been deeply rewarding for everyone involved and reflect our commitment to training the next generation of researchers in this field.
Above: Geographic locations of the Carboncopies team. Circles highlight principal regions and line thickness gives a relative indication of the number of team members in each region.
It’s Time for Brain Emulation - Taking the Pulse of the Times¶
Over the past few years, the scientific and technical foundations required for brain emulation have advanced at a remarkable pace. End-to-end pipelines for acquiring high-resolution brain data have matured from experimental efforts into increasingly robust and scalable workflows, spanning imaging, segmentation, neurite tracing, 3D reconstruction with expert proofreading, and the assembly of connectome databases with identified cell types.
As a result, full connectomes are now available for entire nervous systems such as the fruit fly Drosophila melanogaster, and for zebrafish through resources such as ZAPBench. In parallel, large partial connectomes covering substantial volumes of mouse and human brain tissue are now publicly accessible, offering unprecedented structural insight into mammalian neural circuits.
Momentum is also growing within the neuroscience community to move beyond structure alone and confront the challenge of functionalizing connectome data. In 2024, two independent Nature papers – by Philip Shiu and colleagues (Nature, 2024), and by Janne Lappalainen and colleagues (Nature, 2024) – demonstrated serious, large-scale efforts to reconstruct and simulate functional components of the fruit fly brain. These works signal a clear shift toward treating connectomes not just as anatomical maps, but as substrates for predictive, executable neural models.
This growing engagement was also evident at the third Memory Decoding Challenge satellite event at SfN 2025, produced by Aspirational Neuroscience (a collaboration between the Carboncopies Foundation and the Brain Preservation Foundation). Interest reached an all-time high: attendance by neuroscientists exceeded the room capacity provided by SfN, the competition drew an exceptionally strong field of nominated papers, and the experts panel, composed of leading scientists, demonstrated clear recognition of the importance of this challenge for the future of neuroscience.
Beyond academia, prominent neuroscientists are now openly engaging with whole brain emulation–adjacent challenges through new ventures and collaborations. Examples include Sebastian Seung’s Memazing, which has received public support from Yann LeCun; Timothy Busbice’s ConnectomicAGI; and EON Systems, involving Philip Shiu. These efforts reflect a growing conviction that connectomics-driven functional modeling is becoming both scientifically tractable and strategically important.
At the Carboncopies Foundation, we continuously monitor these developments to understand where the true bottlenecks on the roadmap to whole brain emulation (WBE) lie. While data acquisition and reconstruction continue to improve rapidly, it has become increasingly clear that today’s central challenge is not merely building functional models from connectomes, but rigorously comparing them. The field urgently needs standardized benchmarks that allow different functionalization and emulation approaches to be verified, validated, and evaluated for both accuracy and performance.
This need is precisely what motivated the creation of our BrainGenix platform. As we approach the launch of the Brain Emulation Challenge, our goal is to provide the neuroscience and computational modeling communities with a shared experimental framework – one that enables systematic progress toward faithful, scalable brain emulation.
The pieces are coming together. For the first time, the question is no longer whether brain emulation is becoming feasible, but how quickly we can converge on the methods that truly work.
Aspirational Neuroscience at SfN2025 - Memory Decoding Awards¶
The third Aspirational Neuroscience event was held this November at the 2025 Society for Neuroscience (SfN) Annual Meeting in San Diego. The growing visibility and engagement surrounding this year’s event underscored the accelerating interest in connectomics-driven approaches to understanding memory.
Titled “Long-term Memory Encoding and Connectome Decoding Meetup,” the event featured four $25,000 Aspirational Neuroscience Awards, presented to the first authors of four outstanding papers judged to represent leading progress toward the Memory Decoding Challenge grand prize of $100,000. This grand prize is intended to recognize the first demonstration of decoding a non-trivial memory (or other learned function) using only a static map of synaptic connectivity. While ambitious, we are hopeful that such a milestone could be achieved within the next few years.
In addition to the awards, the event included a five-member expert panel discussion on the challenges and prospects of memory decoding. The panel featured Michal Januszewski (Google Research), Sven Dorkenwald (Allen Institute), Helene Schmidt (Ernst Strüngmann Institute), Andrew Payne (E11 Bio), and Randal Koene (Carboncopies Foundation). The discussion was expertly moderated by Paul Middlebrooks, host of the Brain Inspired podcast.
Interest in the event exceeded that of the 2019 and 2023 meetings by a wide margin. Attendance far surpassed the 115-person capacity of the room allocated by SfN, and even several late-arriving, leading neuroscientists were unable to find space to attend, an unmistakable signal of the field’s growing engagement with these questions.
The panel discussion made clear that leading researchers in neural connectomics and memory recognize the importance of converging on a shared understanding of memory, its encoding, and how it might be identified within connectome data. At the same time, as in previous years, the discussion surfaced persistent and important questions about the interpretation and scope of the Memory Decoding Challenge. These included how to define or quantify a threshold for what constitutes “non-trivial” memory decoding; whether decoding requires the neural implementation of a decoder itself to be extracted and reconstructed from synaptic connectivity; and how best to define a memory engram, an arrangement of synaptic connections that meaningfully represents a memory.
These questions have provided valuable guidance for Aspirational Neuroscience. In response, our two organizing foundations have initiated a collaborative project to further sharpen the grand challenge by developing concrete proposals for example studies. These studies, spanning neural circuits across multiple animal models, are intended to clarify the kinds of experimental and analytical approaches that could ultimately qualify for the grand prize.
The accompanying diagram (above) illustrates how each example study will be described within Aspirational Neuroscience. It also highlights how the Brain Emulation Challenge, developed by the Carboncopies Foundation, will align with the Memory Decoding Challenge by providing synthetic, ground-truth systems for each example. These in-silico systems will enable any researcher to rigorously test ideas, methods, and experimental protocols in a controlled and transparent framework.
2025 Aspirational Neuroscience Award–Winning Papers¶
This year’s awards were presented to the following papers:
- Connectome-constrained networks predict neural activity across the fly visual system, Lappalainen et al. (2024)
- Binary and analog variation of synapses between cortical pyramidal neurons, Dorkenwald et al. (2022)
- Functional connectomics reveals general wiring rule in mouse visual cortex, Ding et al. (2025)
- Systems consolidation reorganizes hippocampal engram circuitry, Ko et al. (2025)
For the full list of nominated papers, please see the 2025 Nominations page at aspirationalneuroscience.org.
COMMUNITY¶
2025 Brain Emulation Challenge Workshop: Functionalizing Brain Data, Ground-Truthing, and the Role of Artificial Data¶
In February, the Carboncopies Foundation convened a focused workshop on the Brain Emulation Challenge, bringing together researchers working at the intersection of connectomics, computational neuroscience, machine learning, and neuroimaging. The workshop examined a central question (video) facing the field today: how do we move from increasingly detailed brain data to validated, functional brain models – and how do we know when those models are actually correct?
A recurring theme throughout the day was the growing gap between our ability to collect large, high-resolution datasets and our ability to reliably functionalize them. While modern connectomics and imaging techniques now provide unprecedented structural detail, translating that structure into predictive, executable models remains a major scientific challenge.
Several talks showcased promising progress. Philip Shiu (video) presented a large-scale computational model of the Drosophila brain built directly from its connectome. Despite using a relatively simple spiking neuron model, the system successfully predicted neural activity and behavior, including feeding and locomotion responses. This demonstrates that even simplified models, when properly constrained by anatomy, can be surprisingly powerful. Janne Lappalainen (video) described complementary work using machine learning to infer neural function from the fruit fly connectome, showing how structural constraints combined with task-relevant optimization can reproduce known circuit computations, such as motion detection.
At the same time, multiple speakers emphasized why validation and ground truth remain difficult. Konrad Kording (video) explored why inverse problems in neuroscience, inferring function from data, are fundamentally hard, especially at scale, and argued that new approaches based on molecular-level measurements may be required to overcome these limits. Razvan Marinescu (video) introduced differentiable MRI simulators as a way to bridge physics-based modeling, synthetic data, and non-invasive brain measurements, offering a potential path toward calibrating large-scale brain models using real-world constraints.
Randal Koene (video) framed these efforts within a broader systems-identification perspective, arguing that progress in brain emulation depends on clear goal definitions, appropriate scale separation, and iterative validation. He introduced the Brain Emulation Challenge (video) and the accompanying Generative Meta-Analysis platform (BrainGenix), which use synthetic, ground-truth neural systems to benchmark and compare different modeling approaches in a controlled and transparent way. The goal is not to replace biological data, but to accelerate method development and reduce ambiguity about what works, when, and why.
The workshop concluded with an open discussion highlighting the importance of shared benchmarks, better alignment between subfields, and more systematic use of existing data. Participants broadly agreed that the field is entering a new phase – one in which the key bottleneck is no longer data availability, but the ability to rigorously validate functional brain models.
To read the full report of the workshop, please visit the Events page in the Community section at carboncopies.org. Complete recordings of each talk at the workshop are available in a YouTube playlist or on our Carboncopies Media Server.
Memory Decoding Journal Club: Structure to Function Mouse Connectomics¶
The following section summarizes one of the many outstanding papers nominated for a 2025 Aspirational Neuroscience award, its presentation and review at our biweekly/weekly Memory Decoding Journal Club. The journal club is organized in collaboration with the Brain Preservation Foundation, reviews all award-nominated papers, as well as additional important recently published work in the field. You are all invited to participate and contribute to these discussions, once every two weeks on Tuesdays at 3 pm US Pacific Time, via carboncopies.org/aspirational-neuroscience. You can also nominate papers for the awards or the grand prize at aspirationalneuroscience.org.
In 1979 brilliant molecular biologist and biophysicist Francis Crick pronounced that it is impossible to create a wiring diagram for a cubic millimeter of brain tissue including the ways in which their neurons operate. Crick would have had difficulty imagining the complexity and speed of the neuroscience data collection pipelines developed since Sebastian Seung’s keynote on connectomics at SfN2007 (“The Once and Future Science of Neural Networks.” Presidential/Featured Lecture, Society for Neuroscience Annual Meeting, Neuroscience 2007, San Diego). Since then, many studies have demonstrated significant progress in pushing the boundaries towards creating a detailed map of the brain, the morphology of neural circuits, and the characteristics of neuronal function. A recent publication by the impactful MICrONS Consortium, “Functional Connectomics Spanning Multiple Areas of Mouse Visual Cortex”, shows us the potential in solving this issue by mapping the connectome of the brain and relating these parts to function.
Above, Figure 1 from the MICrONS paper (Nature, 2025), showing the many types of data collected, combined and composed for the study. On the right, the relationships between the various data types are illustrated.
The study involved in-vivo mouse behavior, exposed to natural and synthetic stimuli while data was collected in multiple stages. The first stage measured active responses by means of calcium imaging of excitatory neurons across cortical layers while the mouse was responding to the visual stimuli. The second stage aimed to map connectivity by recording the same cubic millimeter of mouse brain via serial section transmission electron microscopy (TEM). By using scalable convolution networks and tailored computational systems, researchers were able to reconstruct neurons and their synaptic connections in 3D while employing rigorous proofreading to ensure accuracy. Lastly, they co-registered the calcium-imaging and TEM data to connect the neuronal responses to neurons and their connectivity. The entire dataset is made to be publicly available so that other researchers can explore and run their own analyses and models.
The outcomes of this research have broad implications for future studies as it paves the way in mapping structure and relating it to function. The technologies and methods used here for imaging, data processing, alignment, and proofreading are scalable and could accelerate connectomics in other systems such as other species or eventually, whole-brain maps.
Read the published paper in Nature.
Recordings of each session of the Memory Decoding Journal Club are available on the Aspirational Neuroscience YouTube channel or on our Carboncopies Media Server.
Consciousness & Subjective Experience Journal Club: Why We Experience Subjectively¶
Over the years, the Carboncopies Foundation has also hosted a number of more philosophically inclined workshops, projects and journal clubs. Our second journal club belongs to that category, although with a fairly explicit aim to weave its topic closely into new insights from neuroscience: The Consciousness & Subjective Experience Journal Club. While biweekly through most of 2025, this JC is now held once a month on Sundays.
You can receive notifications and invitations to the JC by letting us know of your interest at [email protected].
The journal club emerged to address reoccurring frequently asked questions that were directed to our Ethics Framework for WBE project and has focused mostly on aspects of what David Chalmers dubbed the “hard problem” of consciousness.
We recently hosted a summary session, in which Randal Koene consolidated insights from the compendium of readings and approaches reviewed and discussed in 2025: Transparent Anthropic Coordinate Transformation (TACT), a model of consciousness and subjective experience.
The Carboncopies Foundation’s core scientific goal is whole brain emulation: collecting brain data, reconstructing brain structure and function, and creating emulations that faithfully reproduce an individual’s cognition and behavior. A successful human brain emulation must therefore reproduce all aspects of cognition, including conscious awareness and subjective experience.
This raises two critical concerns. First, a medical or restorative brain emulation should not result in a so-called philosophical zombie: an entity that behaves identically to a person but lacks the accompanying subjective experience. Second, as neuroscience, AI, and emulation technologies advance, society will increasingly face ethical questions about which systems, human, animal, or artificial may possess conscious experience, are subject to suffering, or well-being. Avoiding inadvertent harm requires a clearer understanding of what subjective experience is and how it arises.
Surveying the scientific and philosophical literature reveals a wide diversity of consciousness theories, often based on very different assumptions. These range from panpsychist and fundamentalist approaches, to biologically embodied theories, to higher-order and representational models. Many popular theories either redefine consciousness in ways that avoid the core mystery, or relocate it to poorly specified physical domains, such as quantum effects, without offering testable predictions.
For the purposes of brain emulation, the most useful theories are those that can be grounded in neural circuitry, make falsifiable predictions, and explain how subjective experience emerges from cognitive function. In practice, this narrows the field to representational and self-model–based approaches, such as those developed by Thomas Metzinger and others working on world- and self-model theories.
A key premise of the TACT model is that we never have access to “pure” experience, only to reported experience. Even introspection consists of the brain generating internal reports, such as noting in working memory that “I experienced X.” There is no demonstrable awareness of experience that is independent of such reporting processes. This observation undermines the idea that subjective experience is fundamentally inaccessible to scientific explanation. If conscious experience is always mediated by reporting mechanisms, then it is amenable, in principle, to functional and mechanistic analysis.
Living in a Virtual Model¶
At any moment, the brain receives only sparse, noisy, and bandwidth-limited sensory input. Most processing occurs subconsciously, resulting in abstracted internal models of the world and of the self. Conscious experience consists of a small, continuously updated subset of these representations, what Metzinger describes as a transparent, full-immersion simulation.
Crucially, the processes that generate this simulation are transparent to introspection. We experience the contents of the model, but not the mechanisms that produced them. This transparency creates the powerful illusion that the self is a fundamental entity located at the center of experience, rather than a constructed reference point within a representational system.
The Core Insight of TACT¶
The Transparent Anthropic Coordinate Transformation model proposes that subjective perspective experience (i.e. an experience like that of a homunculus in a Cartesian Theater) necessarily arises once a system contains:
- A consciously accessible world model,
- A learned self-model embedded within it,
- And relational structures that reference all other information relative to that self-model.
No additional module or special neural correlate is required for this. The subjective, first-person perspective is an unavoidable consequence of how information is organized and referenced within the system. Much like the anthropic principle in cosmology, only systems capable of generating self-referential reports (e.g. humans) are asking questions about the nature of their own existence, and those reports will necessarily be centered on the self-model.
What feels like an irreducible, primary inner reality is therefore an observation-selection effect: the product of missing information about underlying processes combined with the certainty of self-model contents.
Time, Postdiction, and the Phenomenal “Now”¶
Experimental evidence further supports this picture. Conscious perception appears to be delayed and discretized, with reports of experience integrating information over roughly 300–400 milliseconds. Studies involving transcranial magnetic stimulation, split-brain patients, and decision-making experiments show that conscious awareness often reflects a post-hoc narrative rather than direct access to real-time processing.
Neuroscientifically, this aligns with evidence for dual processing modes in the brain: fast, feed-forward subconscious activity and slower, rhythmically organized conscious processing associated with working memory and theta-band oscillations. Conscious experience corresponds to a buffered, integrative “phenomenal Now” constructed atop these mechanisms.
Implications for the Hard Problem and for WBE¶
TACT does not claim to explain every aspect of consciousness, such as the fine-grained character of individual qualia. However, it addresses a central component of the hard problem: why our conscious experience must appear centered, subjective, and irreducible.
Importantly for brain emulation, the model implies that subjective experience does not require any non-physical essence or additional ingredient beyond correct functional organization. A faithfully reconstructed brain, implemented biologically or artificially, would not be a philosophical zombie. Subjective experience would emerge necessarily from the operation of world- and self-modeling mechanisms.
This removes a major conceptual barrier to medical and restorative whole brain emulation and supports the Carboncopies Foundation’s position that accurate reconstruction of brain structure and function is sufficient, in principle, to reproduce both a person’s mind and their subjective sense of being.
Photo in image: David Chalmers, philosopher known for proposing the "hard problem" of consciousness.
This is a topic with many nuanced details. For a better understanding, beyond this brief summary, we strongly encourage you to watch the complete presentation on YouTube or on our Carboncopies Media Server.
RESEARCH¶
Whole brain emulation (WBE) matters to neuroscience because it directly confronts one of the field’s deepest limitations: our difficulty translating vast amounts of neural data into causal, testable understanding. Traditional neuroscience excels at collecting correlations, between neurons, circuits, and behavior, but struggles to integrate these into coherent, mechanistic models of whole brains (see Jonas & Kording, 2017). WBE reframes the problem by demanding models that work: if an emulation built from neural data can reproduce observed dynamics and behavior, then the model captures something essential about how the brain functions. This shifts neuroscience toward a more rigorous, engineering-style discipline, where hypotheses are validated by reconstruction and functional performance, not just statistical association. In doing so, WBE provides a unifying framework for integrating data across scales, from synapses to systems, and for systematically testing theories of learning, memory, cognition, and subjective experience.
For future medicine, WBE offers a path toward genuinely personalized and predictive brain healthcare. Instead of treating neurological and psychiatric disorders through broad, population-level interventions, emulated brain models could enable clinicians to explore disease mechanisms, progression, and treatment responses in-silico, before intervening in a patient’s brain. This could transform how we understand disorders such as Alzheimer’s disease, epilepsy, depression, and traumatic brain injury, enabling precise targeting of circuits, stimulation protocols, or pharmacological strategies. In the longer term, WBE also opens the possibility of preserving and restoring brain function when biological tissue fails, redefining what recovery and continuity of mind could mean. Together, these advances point toward a future where neuroscience research and medicine are not only more explanatory, but also more humane, effective, and forward-looking.
Brain Emulation Challenge¶
Brain Emulation Challenge: What and Why?¶
The final and most difficult step toward whole brain emulation is translation: converting high-resolution brain data into executable, validated models that reproduce meaningful neural function. While modern neuroscience can now generate extraordinarily detailed structural and physiological datasets, there is no reliable way to fully evaluate whether reconstruction and system-identification methods actually recover the causal mechanisms underlying brain function. The Brain Emulation Challenge addresses this gap by providing synthetic, ground-truth neural systems where both structure and function are fully known, enabling rigorous validation of emulation approaches.
Inspired by the role of standardized datasets in accelerating machine learning, such as ImageNet and Kaggle competitions, the Brain Emulation Challenge will offer tiered synthetic datasets with explicit success metrics. Using the open-source BrainGenix platform, we generate realistic in-silico neural circuits with embedded, hidden functions and produce multimodal “virtual experimental” data, including simulated EM, electrophysiology, and calcium imaging. Participating researchers attempt to reconstruct and emulate these systems using their own methods, receiving quantitative feedback on where and how their approaches succeed or fail. By iterating over progressively larger and more complex challenges, the community can rapidly debug methods, compare approaches, and drive cumulative progress toward validated brain emulation.
How the Challenge Works¶
Each challenge begins with a batch of synthetic ground-truth neural systems created using BrainGenix. From these systems, we collect virtual experimental data using realistic stimulation protocols and simulated measurement physics. Participants receive only the resulting datasets, not the hidden function embedded in the synthetic brain. They are tasked with reconstructing the system’s structure and dynamics to produce a working emulation.
Because the ground truth is known, submitted models can be directly evaluated. Performance is scored using primary behavioral success metrics (“Does the reconstructed system reproduce the intended cognitive or functional behavior?”), along with secondary measures of dynamic and structural similarity. This feedback allows researchers to identify systematic weaknesses, to compare competing methods, and understand the conditions under which particular approaches succeed. Success on synthetic ground-truth does not guarantee success on biological data, but it eliminates major sources of ambiguity and greatly increases confidence that a method is on the right track.
How BrainGenix Supports the Challenge¶
The Brain Emulation Challenge is powered by Carboncopies’ BrainGenix platform, which combines biologically grounded neural morphogenesis with executable neural dynamics. An extended version of Netmorph generates realistic populations of neurons and their axonal and dendritic morphologies, producing a “connectome reservoir” with candidate synaptic sites. BrainGenix-NES then selects and tunes synapses, applies neuron and synapse models, and entrains specific circuit functions through learning rules such as STDP, creating a synthetic brain with a hidden, purposeful function.
From these systems, BrainGenix performs virtual data acquisition that closely mirrors real neuroscience experiments, including high-throughput EM, Neuropixel-like electrophysiology, and calcium imaging, complete with configurable noise and artifacts. The result is a controllable, transparent virtual brain laboratory where complexity and realism can be scaled systematically. This allows the Challenge to offer progressively harder problems, isolate specific failure modes, and support method development across many levels of sophistication.
The image is a snapshot from Neuroglancer displaying the detailed 3D neurite structure of a synthetic ground-truth example generated by Netmorph and BrainGenix-NES. Each neuron’s morphology is shown in a unique color for ease of inspection.
Alignment with the Memory Decoding Challenge¶
The Brain Emulation Challenge aims to align with the Memory Decoding Challenge by providing synthetic, ground-truth systems for proposed experimental targets and protocols. Together, these efforts enable researchers to test analysis methods, decoding strategies, and experimental designs in a fully transparent setting, before deploying them on biological data. This alignment strengthens both initiatives by grounding ambitious neuroscience goals in rigorous, testable frameworks.
High-Performance BrainGenix¶
BrainGenix is a research software platform for Brain Emulation and computational neuroscience research. It fills a gap in the existing ecosystem of neuroscience software, by aiming specifically at solving the challenges faced by researchers working on WBE related technologies. At present, platform development is focused on helping solve the Translation problem (converting collected brain data into a simulatable and validatable model of the brain’s neural circuits). We generate artificial neuronal systems, by neural morphogenesis or explicit specification, and simulate data acquisition on those to produce data corresponding to a known ground-truth. Subsequently, we evaluate methods and compare their in-domain and out-of-domain performance to the known ground-truth – thus allowing us to measure accuracy achieved and identify areas for improvement.
This quarter we have been working to rewrite the Virtual Scan Data Acquisition (VSDA) server into its own separate service. In addition, we are focusing on having the new server take advantage of GPUs via SYCL, and implementing distributed fault tolerant rendering across a cluster of heterogeneous GPUs.
Closing the Translation Problem: Synthetic Ground Truth for Brain Emulation¶
This year, Carboncopies Software Lead Thomas Liao completed his UCSC senior thesis, delivering a major milestone for the Brain Emulation Challenge and our long-term goal of whole brain emulation. A central obstacle to brain emulation is the Translation Problem: how to reliably convert high-resolution brain imaging data into accurate, functional computational models. Progress has been limited by a lack of validation. Without known ground-truth, there is no objective way to measure how well reconstruction and system identification algorithms perform.
Thomas’ thesis addresses this gap by building a high-performance version of the core synthetic data engine behind the Brain Emulation Challenge: BrainGenix, a distributed, GPU-accelerated platform for generating realistic connectomics datasets with fully known ground-truth. Instead of relying solely on experimental brain data, BrainGenix makes it possible to create virtual brain tissue, image it synthetically, and use the results to rigorously score methods used that attempt an emulation.
A key contribution of this work is a performance-portable architecture built using the open SYCL standard. This allows the same codebase to run efficiently across NVIDIA, AMD, and Intel GPUs without vendor lock-in, an essential requirement for deployment on diverse high-performance computing clusters. The system combines this with a scalable distributed design that can self-organize across many nodes, enabling large synthetic datasets to be generated efficiently and robustly.
At the core of the platform is a custom spatial indexing structure, the VoxelGridTree, which enables extremely fast spatial queries over scenes containing billions of neuronal primitives. Benchmarking shows orders-of-magnitude speedups over naive approaches for the read-heavy workloads required during virtual imaging and scheduling. While insertions are intentionally slower, this tradeoff is well-matched to BrainGenix’s “write once, read many” workload and results in dramatically faster overall rendering performance.
Figure 4.2.2 from the thesis: Insertion throughput (primitives / second) versus scene complexity (number of primitives) for four subdivision schemes (2×2×2, 3×3×3, 4×4×4, 8×8×8) across multiple thread counts. Performance degrades relative to naive insertion due to tree traversal overhead, per-node locking, and bounding box calculations. The 3×3×3 configuration demonstrates optimal insertion performance, balancing tree depth against computational overhead. Higher subdivision factors (8×8×8) show poorest performance despite less mutex contention.
The thesis also demonstrates substantial gains from GPU-accelerated rasterization and image generation, achieving multi-fold speedups over CPU baselines and identifying clear performance regimes where memory bandwidth becomes the limiting factor. These insights directly inform how BrainGenix schedules work and batches rendering tasks to maintain peak throughput at scale.
Figure 4.3.2.2 - RGB Performance from the thesis: Three-channel bitmap generation showed reduced absolute throughput on both architectures due to increased memory traffic, with GPU performance reaching 942.3 Mpix/sec for 2048×2048 RGB images compared to 482.0 Mpix/sec on CPU. The relative GPU speedup (1.96× for RGB vs. 2.76× for grayscale at 2048×2048) suggests that memory bandwidth becomes the limiting factor for multi-channel data, though parallel acceleration remains advantageous.
Finally, Thomas validated the platform’s distributed coordination system, showing rapid and stable convergence across simulated clusters of up to 256 nodes, with reliable leader election and fault recovery, critical infrastructure for running BrainGenix as a large-scale, shared scientific resource.
Figure 4.4.3.2 from the thesis: illustrates the progressive convergence of per-node cluster size views for the n = 256 trial, representative of behavior across all scales.
Together, this work provides the technical foundation for a functional benchmark of brain emulation approaches, one that can evaluate not just structural accuracy, but whether reconstructed models preserve the computational behavior of neural circuits. It is a key step toward turning brain emulation from an aspirational idea into a rigorously testable engineering discipline, and a major contribution to the mission of the Carboncopies Foundation.
Neuroscience of WBE: Profound Implications¶
Whole brain emulation (WBE) is rapidly shifting from the realm of science fiction to a serious scientific frontier. A speculative but forward-looking review article by Mateusz Brzeziński, “Uploading Minds: The Race for Whole Brain Emulation (WBE) and Its Profound Implications” shows how breakthroughs in connectomics – from mapping the humble C. elegans worm to reconstructing the entire fruit fly brain and even a 1 mm³ slice of mouse cortex – are quietly assembling the toolkit required.
Yet behind this technological momentum lies a set of profound and potentially unsettling questions. Who will gain access to putative digitally extended life, and who might not? If a mind is reconstituted, through that, does a person truly carry on, or is this something entirely new? And if non-biological brains become a reality, how are rights, protections, or freedoms defined and extended?
The article turns a scientific race into a philosophical cliffhanger – inviting readers to imagine a future where consciousness may outlive its initial embodiment, but with the potential for ethical turbulence ahead.
Read the thought provoking article on the Bez Kabli forum.
Fidelity Tradeoffs in Whole Brain Emulation¶
If you are interested in how contemporary research approaches questions of accuracy and ethics in brain emulation, take a look at Functional Tests Guide Complex “Fidelity” Tradeoffs in Whole-Brain Emulation (Koene & Linssen, 2025). Linssen and Koene demonstrate that creating a digital mind is not merely an engineering task but a delicate balance between biological detail, computational cost, and the moral implications of reconstituting a person. Drawing on examples from neuroscience, modeling, and cognitive testing, the authors show that WBE cannot be evaluated by a single accuracy metric – it demands a multidimensional system of tests encompassing neural network dynamics, behavior, memory, skills, and individual traits.
Crucially, the article addresses the socio-ethical risks of emerging “tiers” of emulation, where financial resources could determine cognitive abilities, processing speed, or even subjective well-being. For this reason, the article serves not only as a technical indicator but also as a meaningful contribution to ongoing debates about digital rights, fairness, and the future governance of minds operating on non-biological substrates within an expanding ecosystem of intelligences.
Read the paper in the Journal of Ethics and Emerging Technologies.
Bottleneck of BCI¶
Zheng and Meister (2024) explore a striking paradox of brain function. Our sensory systems can process information at gigabit levels, yet our behaviour and conscious output run at only about 10 bits per second. Despite massive neural capacity, the mind communicates through an extremely narrow channel. The author suggests the bottleneck lies within what they call the “inner brain” where decisions, attention, and actions are processed in sequence.
For brain-computer interfaces (BCI), this poses a major challenge because even the most advanced systems cannot bypass the brain’s natural limits on throughput. This finding may have implications for the Carboncopies Foundation’s goal of whole brain emulation. Mapping the brain’s structure alone may not be sufficient without understanding the functional limits that shape cognition and behaviour. The paper reminds us that building realistic emulations or interfaces requires addressing both biological complexity and processing and output bottlenecks imposed by the brain’s architecture and physiology.
Read the Neuron paper at the Meister lab: The unbearable slowness of being: Why do we live at 10 bits/s?
Janelia Publishes Complete Drosophila Central Nervous System¶
This year, the FlyEM team at HHMI Janelia Research Campus, along with collaborators at Cambridge and Google Research, published a landmark result: the first complete connectome of an entire adult male Drosophila central nervous system, spanning the brain, optic lobes, and ventral nerve cord.
The dataset contains 166,691 fully proofread neurons, organized into 11,691 annotated cell types, with synaptic-resolution wiring across the entire CNS. Crucially, this is the first finished connectome of a male fly brain, enabling, for the first time, direct, whole-brain comparisons between male and female nervous systems at synaptic resolution.
Using this unprecedented resource, the authors identify 262 male-specific, 69 female-specific, and 114 sexually dimorphic neuron types, alongside over 7,200 isomorphic types shared across sexes. Although sex-specific and dimorphic neurons make up only a small fraction of the brain (≈5% in males), their influence is amplified through dimorphic connectivity, affecting nearly one-fifth of male brain neurons. This provides compelling evidence that modest, targeted wiring differences can propagate through neural circuits to produce large behavioral effects.
A striking organizational principle emerges: sex differences are concentrated in higher-order brain centers, while sensory inputs and motor outputs are largely conserved across sexes. Within these higher centers, sex-specific neurons form tightly clustered hubs and circuit “switches” that reroute sensory information into alternative downstream pathways, helping explain how the same sensory inputs can drive opposing, sex-specific behaviors such as courtship versus aggression.
Beyond its biological insights, this connectome represents a major technical milestone. It enables end-to-end circuit tracing from eyes to legs, links over 95% of neurons to known cell types and prior functional studies, and provides a gold-standard reference for comparative and computational neuroscience. The work also demonstrates how whole-brain connectomics can move beyond individual cell types to reveal circuit-level principles governing behavior.
(See video by Janelia Research on YouTube.)
For the field at large, and for long-term efforts like whole brain emulation, this achievement underscores both the power of complete, validated connectomes and the importance of comparative datasets. By showing how small, well-placed circuit differences scale up to brain-wide functional changes, the male Drosophila CNS connectome sets a new benchmark for understanding how structure gives rise to behavior.
CiteRef: Berg, S. et al. (2025). Sexual dimorphism in the complete connectome of the Drosophila male central nervous system. bioRxiv 2025.10.09.680999; doi: https://doi.org/10.1101/2025.10.09.680999
Seminal Results of MICrONS¶
One of the most significant neuroscience milestones of recent years came to fruition in 2025 through published data and studies from the MICrONS Consortium, which has delivered the first dense, multimodal reconstruction of a cubic millimetre of mammalian cortex, an achievement that was widely considered impractical only a decade ago.
A single cubic millimetre of brain tissue contains tens of thousands of neurons and hundreds of millions of synapses, generating on the order of a petabyte of data. The MICrONS project overcame this challenge by combining advances in large-scale electron microscopy, machine learning–based reconstruction, automated proofreading, and collaborative data infrastructure. The result is an unprecedented dataset comprising ~200,000 neurons and 523 million synapses from mouse primary visual cortex and surrounding areas.
What sets MICrONS apart from previous connectomics efforts is that it pairs this dense anatomical reconstruction with in-vivo functional recordings from ~75,000 of the same neurons. This makes MICrONS the first truly large-scale example of functional connectomics, allowing direct links between neural wiring and neural computation. The flagship paper describing this dataset is available here:
Functional connectomics spanning multiple areas of mouse visual cortex
New Tools for Connectomics at Scale¶
To make this effort possible, the consortium developed and released foundational tools that will benefit the entire field. NEURD provides automated proofreading and feature extraction for large connectomes, helping manage error rates that would otherwise overwhelm human curators:
https://www.nature.com/articles/s41586-025-08660-5
To support collaboration on petabyte-scale datasets, the team also introduced CAVE (Connectome Annotation Versioning Engine), a purpose-built system for versioning and coordinating large, evolving connectomic datasets:
https://www.nature.com/articles/s41592-024-02426-z
Together, these tools represent a major advance in the infrastructure required for scalable, collaborative neuroscience.
Mapping Cell Types and Circuit Architecture¶
On the anatomical side, MICrONS enables a census of cortical cell types based on connectivity, morphology, and ultrastructure, complementing genetic and transcriptomic classifications. Using unsupervised methods applied to the dense reconstructions, the team identified and mapped neuron types across cortical layers and areas, revealing highly specific patterns of connectivity, particularly among inhibitory neurons.
Key results include detailed atlases of inhibitory specificity and connectivity-based classification of cell types, described in papers such as:
- Inhibitory specificity from a connectomic census of mouse visual cortex
- Perisomatic ultrastructure efficiently classifies cells in mouse cortex
- Connectomics of predicted Sst transcriptomic types in mouse visual cortex
- An unsupervised map of excitatory neuron dendritic morphology in the mouse visual cortex
These results demonstrate how connectivity-based analysis can reveal organizational principles and cell distinctions that are difficult to capture with genetics or morphology alone.
From Wiring Diagrams to Digital Twins¶
The functional side of MICrONS is equally transformative. By combining dense wiring diagrams with neuronal activity, the consortium uncovered general wiring rules that relate functional similarity to synaptic connectivity, rules that go well beyond simple spatial proximity. Dense reconstruction also revealed higher-order connectivity patterns that cannot be inferred from pairwise interactions alone.
Perhaps most excitingly, the MICrONS team used this data to construct a digital twin of the imaged cortex, enabling large-scale in-silico experiments. These models allow researchers to test hypotheses about cortical computation, predict responses to novel stimuli, and guide future biological experiments.
Key papers include:
- Functional connectomics reveals general wiring rule in mouse visual cortex
- Foundation model of neural activity predicts response to new stimulus types
Why This Matters¶
MICrONS marks a turning point for neuroscience. It shows that millimetre-scale, synaptic-resolution, functionally grounded connectomics is now achievable in mammals. For the broader goals of understanding brain computation, and ultimately enabling whole brain emulation, this dataset provides both a proving ground for theories and a benchmark for future tools and models.
The full dataset is openly available and can be explored at:
https://www.microns-explorer.org
As connectomics continues to scale, MICrONS stands as a clear demonstration that the combination of dense anatomy, function, and computation can move the field from descriptive wiring diagrams toward predictive, mechanistic models of brain function.
You can watch a recording of our Memory Decoding journal club session discussing the functional connectomics MICrONS paper on the Carboncopies Media Server.
Modeling the Fly Brain¶
One of the award-nominated studies at the Aspirational Neuroscience 2025 (AN2025) event takes an ambitious step towards whole-brain simulation. Researchers have built a detailed computational model of the adult Drosophila brain, capturing more than 125,000 neurons and 50 million connections.
The model reveals how sensory signals such as taste and touch give rise to behaviours like feeding and grooming. By simulating sensory input, the researchers could predict which neurons drive movement, and those predictions were later confirmed in real experiments. The work shows how connectome-based modeling can turn vast neural wiring maps into living, testable systems, bringing us closer to understanding how brains transform sensation into action.
Read the study here: A Drosophila computational brain model reveals sensorimotor processing
CiteRef: Nature (2018). 3D image reveals hidden neurons in fruit-fly brain. Nature; doi: https://doi.org/10.1038/d41586-018-05782-x
This map shows the precise location and arrangement of the 50 largest neurons of the fly brain connectome. These 50, along with another 139,205 brain cells in the brain of an adult fruit fly, were painstakingly mapped by a Princeton University-led team of neuroscientists, gamers and professional tracers. Activity within these neurons (brain cells) drives everything the organism does, from sensory perception to decision-making to controlling flight. The brain cells are connected by more than 50 million connections (synapses).
Credit: Tyler Sloan and Amy Sterling / FlyWire / Princeton University
Predicting Fly Vision From Wiring¶
An award-winning study at Aspirational Neuroscience 2025 (AN2025) shows how wiring diagrams can be transformed into working models of neural computation. The research team built a detailed simulation of the fruit fly visual system using real connectivity from 64 cell types and more than 45,000 neurons. The model was trained on natural video clips, allowing it to discover how to detect motion without being given any rules about vision.
Even without using recorded neural activity during training, the model reproduced several well known features of fly vision. It captured the split between ON and OFF pathways, the direction selectivity of T4 and T5 neurons, and the timing patterns that help these cells detect motion. The predictions also matched data from more than 25 experimental studies
Beyond reproducing known findings, the model suggested new roles for several neuron types, including possible parallel motion pathways that have not been tested yet. This work highlights how connectome based simulation can turn static wiring maps into dynamic hypotheses about neural function and provide a powerful guide for future experiments.
Read the study here: Connectome-constrained networks predict neural activity across the fly visual system
CiteRef: Lappalainen, J. K. et al. (2024). Connectome-constrained networks predict neural activity across the fly visual system. Nature; doi: https://doi.org/10.1038/s41586-024-07939-3
EDUCATION¶
Books: The Future Loves You¶
What if death was reframed as a solvable engineering problem? In The Future Loves You: How and Why We Should Abolish Death, neuroscientist Ariel Zeleznikow-Johnston argues the Why, What, and How of the cessation of death. By linking brain preservation and whole brain emulation, it shows how emerging technologies could turn mortality from destiny into a technically preventable condition.
Synthesizing contemporary research in neurobiology, cognitive science, and brain preservation, he argues that personal identity arises from the brain’s physical and informational architecture. From this standpoint, death refers not to the loss of heartbeat or breath, but to the irreversible loss of that structural substrate. Throughout medical history – from the advent of CPR and defibrillation to the development of mechanical ventilation – advances in life-support technologies have repeatedly shifted our operational definition of death, revealing that earlier criteria often reflected the limits of intervention rather than intrinsic biological boundaries. By placing emerging preservation technologies in dialogue with modern bioethical concerns, this book invites readers to reconsider the scientific and moral contours of a world where mortality could become optional.
Dr. Ariel Zeleznikow-Johnston presents his book at the Foresight Institute: Dr Ariel Zeleznikow-Johnston | The Future Loves You
Where to buy: The Future Loves You: How and Why We Should Abolish Death: Zeleznikow-Johnston, Dr. Ariel: 9780241655894
Media: Upload¶
Have you ever wondered what it would be like to live on in digital form? What might life be like? Well, Amazon Prime’s Upload is a fun show that tackles these questions with lots of ideas based in science, comedy and drama. The story takes place in the year 2033, where the main character Nathan Brown dies in a self-driving car accident and is uploaded into a virtual afterlife called Lakeview that is controlled by a mega corporation. As we watch Nathan explore his new consciousness in the afterlife, we see a nearly perfect simulation of reality. Objects and people are recreated exactly as their archetypes in real life. In one scene, we see Nathan physically get uploaded and reconstructed into Lakeview, in which he looks exactly the same as he did in real life and has even retained all his memories. As the show goes on, we see how Nathan and other characters interact physically and verbally with each other as well as other objects that have been created in this world. This accomplishment, reconstituting one’s embodiment and conscious awareness in another substrate, explores socially intriguing avenues that might be enabled by whole brain emulation.
Potentially, whole brain emulation may be achieved by reverse engineering a biological brain and constructing a synthetic brain from that, reproducing memories, characteristics and cognitive abilities. In the show, we see through Nathan’s perspective what effects a common-place and commercially available process could have. And as we learn more about the fictional world of the show, we are invited to consider the moral and ethical implications. The show is worth a watch as light-hearted SciFi based on the premise of whole brain emulation.
Check out the trailer for Season 1 here.
Crossword Puzzle¶
|
Across
|
Down
(Look here for the crossword answers.) |
OPPORTUNITIES¶
Team Member Spotlight¶
Meet Aiden Behler, a senior studying computer science whose curiosity about neuroscience and computing led him to Carboncopies. While searching online for connections between brains and computers, Carboncopies’ Brain Emulation Challenge strongly resonated with him.
As a Data Science Intern, Aiden developed his very first neuron simulation model alongside his team – an experience he describes as both challenging and rewarding, and one that deepened his understanding of and passion for neuroengineering. He especially valued the open, collaborative culture and the Memory Decoding Journal Club, which made his time with the team truly memorable.
Although he continues at the Foundation in a smaller part-time role, following the end of his intensive internship, what Aiden has learned and how he has grown continue to keep him connected to the Carboncopies family.
Volunteer - Citizen Science Roles¶
Complex Systems Research
We're seeking a volunteer with expertise in complex systems, system identification, control theory, inverse problems, and/or applied mathematics to contribute to our research initiatives and technical projects.
Creative Consultant
We are seeking a volunteer to generate imaginative, compelling ideas for content that communicates our mission and inspires our audiences, sparking new initiatives.
General Research Assistant
We are seeking a volunteer with a strong interest in the technological achievement and development of whole brain emulation (WBE). You will acquire requisite skills on the job and support the practical application of multiple R&D workstreams.
PR & Donor Relations
We're seeking a motivated PR volunteer to help establish and maintain relationships with corporate donors, focusing on securing IT resource donations and managing donor communications.
Social Media Content Creator
We are looking for a volunteer to help us create engaging social media content, executing the content calendar.
Social Media Manager - Volunteer
We're looking for a creative and organized volunteer to manage our social media accounts and help grow our online community to increase awareness of our mission and activities.
Technical Onboarding Specialist
We are looking for a volunteer to guide new technical volunteers and interns throughout the process of joining the team so they may effectively contribute to our mission.
Internships¶
If you are interested in our mission and projects and would like to join, please email us at [email protected] with your resume/CV, indication of the field of interest, preferred start date, duration and hrs/week. We’re happy to work with students and structured OTP internships. For more information about internships at the Carboncopies Foundation, please visit https://carboncopies.org/Internships/.
Control Systems Research 2026 Internship (winter start, 4+ months)
We're seeking an intern with expertise in complex systems, system identification, control theory, inverse problems, and/or applied mathematics to contribute to our research initiatives and technical projects.
Data Science Engineering 2026 Internship (winter start, 4+ months)
We are seeking a data science intern who will focus on analyzing electrophysiology data from neurons and developing quantitative metrics.
Python/C++ Software Engineering 2026 Internship (winter start, 4+ months)
We're seeking a skilled programming intern with experience in C++ and/or Python to help develop and improve our software projects and technical infrastructure.
Large Language Model Engineering 2026 Internship (winter start, 4+ months)
We're seeking an enthusiastic AI intern with a desire to build special-purpose LLM applications to support our internal team’s work.
Javascript / React Programming 2026 Internship (winter start, 4+ months)
We're seeking a skilled Javascript programming intern interested in web-facing visualization and graph interaction applications to help build out the interface to our whole brain emulation roadmap.
Website Designing and Building 2026 Internship (winter start, 4+ months)
We're seeking an intern with understanding of HTML, Javascript, Markdown and Git to help us design, build and publish websites for our projects and collaborations.
Please feel free to forward or post our opportunities or this newsletter to your lab, or colleagues and friends who are likely to be interested in whole brain emulation and the Carboncopies Foundation.