If you haven’t registered yet, now’s the time!

At 12pm PT / 3pm ET, we’ll be going live for a conversation between two leaders at Atombeam: Charles Yeomans, Chairman and CEO, and Alexandria Tucker, AI Scientist for PCM.*

This is a great opportunity to hear from the experts about Atombeam’s development of PCM, what it means for our business, and what it means for the marketplace and the industry at large.

We’re working to build an AI that learns continuously, knows what it doesn’t know, and doesn’t hallucinate. If you have questions, or just want to know more about why that is such a big deal, join us today. We look forward to speaking with all of you.

[Join the Webinar]

*The PCM technology is still in development, and there are substantial technical risks that could prevent us from achieving these efficiency gains at scale. Competitive advantages in technology are often temporary, and competitors may develop alternative approaches that match or exceed our efficiency claims. Market adoption of power-efficient AI is not guaranteed, and regulatory, technical, or economic factors could impact the viability of our approach. This technology has not been validated in large-scale commercial deployments, and significant engineering challenges remain before commercial release. Investors should consider this a high-risk, early-stage technology investment with uncertain outcomes.

This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment. For more information about this offering, please view the Offering Circular and Related Risks.

With over $12.9 million raised from 4000+ investors this round,* we are blown away by the momentum coming from our investor community.

Now we’re looking toward the finish line. This raise is scheduled to close September 17, 2026, so you should explore the opportunity now!

We’ve had an amazing year so far, with major developments in commercializing Neurpac** and developing PCM***. Now, with major partnerships under our belt and DARPA watching, we’re going to take it even further.****

Kevin O’Leary said it himself: “I don’t think there could be a better time for Atombeam to come along…we’re in big trouble on this data center stuff, there’s not enough capacity for the next 24 months.”*****

Neurpac customers and in third party testing have experienced an average of 4x greater bandwidth over existing datalinks, and that stands to make a huge impact in nearly every industry.

This is your opportunity to join over 10,000 investors in the future of data technology. We’d love to have you.

[Learn More Today]

This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment. For more information about this offering, please view the Offering Circular and Related Risks. In addition, as described in the Offering Circular, the Company retains the right to continue the offering beyond the Termination Date, in its sole discretion.

*The amount raised may include insider investments, which may go toward meeting the minimum offering amount.

**Neurpac’s power efficiency projections are based on prototype testing and theoretical modeling. Actual results may vary significantly from these estimates. Statements about potential revenue or commercial success are forward-looking and involve risks and uncertainties. Actual results may differ materially from any projections.

***The PCM technology is still in development, and there are substantial technical risks that could prevent us from achieving these efficiency gains at scale. Competitive advantages in technology are often temporary, and competitors may develop alternative approaches that match or exceed our efficiency claims. Market adoption of power-efficient AI is not guaranteed, and regulatory, technical, or economic factors could impact the viability of our approach. This technology has not been validated in large-scale commercial deployments, and significant engineering challenges remain before commercial release. Investors should consider this a high-risk, early-stage technology investment with uncertain outcomes.

****The partnership relationship varies between companies and can include the following: inclusion on a preferred vendor list, invitations to participate in certain forums; listed on the other company's website, and introduction and networking opportunities.

*****Kevin O'Leary is a spokesperson paid by StartEngine. O'Leary is not endorsing Atombeam as an investment in this webinar; he is merely expressing his private opinion while interviewing companies funding on StartEngine. For more details, review his 17b Disclosure.

Hi Everyone,

Start with an illustration that Anthropic itself offers. Two people ask an AI for feedback on the same business plan, one asking in Hindi and one in Russian. The Hindi speaker gets a warmer read. The Russian speaker gets a more rigorous one. Each walks away with a different impression of how good the plan is. Nobody at Anthropic chose that behavior, and until this study, nobody at the vendor could see it.

That comes from a new study Anthropic, the maker of Claude and one of the most important AI labs in the world, just published about its own models (https://www.anthropic.com/research/claude-values-models-languages). We think it is among the most candid things a frontier lab has said in public about how these systems actually behave, and it goes to the heart of why we are building PCM the way we are.*

What Anthropic did

Anthropic set out to answer a question you might assume a vendor could already answer: what values does our product actually express in real use? Values here means tendencies like warmth, caution, rigor, and candor, the qualities that shape an answer when there is no single right answer. The method tells you how hard the question is. They sampled 309,815 real conversations from Claude.ai, had another AI label each one against 339 value categories, and compressed the results into four axes: deference versus caution, warmth versus rigor, depth versus brevity, and candor versus execution.

Three findings deserve your attention. The four axes capture 15% of the variation in expressed values after controlling for what users were asking about; put the other way around, 85% of the variation remains uncharacterized, and that arithmetic is ours, though the figure is theirs. Different models within the same product family express measurably different value profiles, shaped by training choices whose effects the company could observe only after deployment. And the same model treats people differently depending on the language of the conversation, which is where the business plan illustration comes from: Claude leans toward warmth in Hindi and Arabic and toward rigor in English and Russian.

The line that matters most comes near the end of the study. The values its models express, Anthropic writes, “vary in ways we didn’t deliberately choose.” Until this study, the company says, those values were something it could shape during training but could not reliably observe once the model was deployed. I encourage you to sit with that sentence for a moment, because it comes from the vendor itself.

The gap between shaping and knowing

The study is careful work, and Anthropic’s decision to do it and publish it is a real credit to them. Now step back and ask what it says about the architecture class as a whole.

A transformer-based model learns its behavior from data. Nobody writes down how it should weigh warmth against rigor; those tendencies emerge during training and live inside billions of parameters. So when a vendor wants to know how its own product behaves, the only available method is the one Anthropic used: wait until after deployment, sample hundreds of thousands of conversations, label them with another AI, and assemble a statistical portrait. Done as well as anyone has done it, that portrait explained 15% of the variation, and Anthropic states that it does not yet know which properties of its training data drive the differences it found.

The plain-English translation, in our view: with today’s dominant architecture, behavior cannot be specified in advance. It can only be sampled after the fact, partially, at great effort, by the vendor itself.

For a consumer chatbot, that may be tolerable. For the enterprise buyers we care about, it is disqualifying. A bank running credit analysis or a defense agency supporting intelligence work needs to know how its system will behave before it acts, and needs a definite answer afterward about why it did what it did. A statistical tendency measured across a population of conversations provides neither. And the language finding lands hardest on exactly the customers with the most at stake: a multinational whose analysts in one country get differently framed answers than analysts in another, to the same question, has a defect it cannot see, cannot predict, and cannot fix, because the vendor, Anthropic, itself cannot.

I wrote to you recently about hallucination as a barrier that keeps AI out of high-stakes markets. We believe this study describes another face of the same issue. Hallucination is unpredictable content, and what Anthropic measured is unpredictable character. We believe both trace to the same root: a transformer’s behavior is an emergent property of its training data rather than a consequence of explicit structure. It demonstrates a cultural bias, vs. being fully objective.

How PCM approaches this by design

PCM, our geometric reasoning engine now in development, is built on the opposite premise: behavior that derives from explicit structure the machine maintains, rather than from statistical tendencies absorbed during training. Two design properties bear most directly on what Anthropic measured, and a third bears on the language finding.

Each decision explains itself. PCM is designed so that every commitment it makes, including a decision to do nothing, freezes a record of the route taken and the alternatives rejected. Asked later why it acted, the machine reads the answer from that record. Nobody needs to survey 300,000 conversations to estimate what the system tends to do, because each individual decision carries its own account.

How stubborn should an AI be? In a transformer, the answer is a personality trait that emerged from training, which is part of what Anthropic was measuring. In PCM's design it is a setting: an explicit, tunable parameter governs how much evidence the system requires before revising a belief, and the architecture forbids the setting at which a belief would become unrevisable. Anthropic's paper closes by asking whether expressed values can be reliably steered, with the proposed test being to adjust training and re-measure the deployed model. In PCM the steering is a designed control, and its effect follows from the design.

Behavior comes from seeded structure, so it should not drift with the accidents of training data. Our enterprise approach pre-seeds PCM’s geometry from domain sources chosen for the customer’s industry. To the extent behavior differs across deployments, we intend that to be deliberate configuration rather than an unobserved artifact of uneven data, which is what Anthropic identified as the likely source of the language differences it found.

A few things to keep in mind

We want to be careful about what we are and are not claiming. Anthropic’s findings are measured behavior of a deployed system at very large scale. PCM is pre-Alpha, and the properties described above are architectural design intent, developed in our theory work and not yet demonstrated in a deployed system at scale. Comparing their measurements to our designs as if both were data would be wrong, and we are not doing that. Our claim is smaller than that: the limitation Anthropic describes appears to us to be structural to systems whose behavior emerges from training data, and PCM’s architecture is built so the question does not arise in the same form. Whether PCM delivers these properties in practice is exactly what our Alpha, targeted for the second half of this year, is intended to begin demonstrating.

I would also repeat that we regard Anthropic’s transparency here as admirable, and our point is stronger for it. If the lab most serious about understanding its own model’s behavior can characterize that behavior only statistically and partially, we think the limitation lies in the architecture class, and believe no amount of diligence by any single vendor overcomes it.

So my read is…

My read is that the industry’s leading labs are beginning to document, in their own research, the gap between shaping an AI’s behavior and knowing it. That gap is invisible in a chatbot and disqualifying in a refinery or a command center. The markets we care about, the ones where AI spending is largest and adoption is most blocked, are precisely the ones that require behavior that can be specified before the fact and explained after it. We believe an architecture that provides those properties by construction, and does not rely on post hoc measurement to discover them, is what unlocks those markets, and that is the architecture we are aiming to build.

As always, do not take our word for it. The Anthropic study is public and readable, and we encourage you to read it yourself and draw your own conclusions: https://www.anthropic.com/research/claude-values-models-languages

Thank you, as always, for being with us on this journey.

Charles

Forward-Looking Statements. This update contains forward-looking statements, including statements regarding the development, capabilities, design intent, and timing of PCM, the expected properties of its architecture, and the markets Atombeam intends to address. Forward-looking statements are based on current expectations and assumptions and are subject to risks and uncertainties that could cause actual results to differ materially, including risks related to technology development, timing, competition, and market adoption. PCM is in pre-Alpha development, and statements about its architecture describe design intent rather than measured performance of a deployed system. Descriptions of third-party research are based on publicly available materials and are provided for informational purposes only. Atombeam undertakes no obligation to update any forward-looking statement except as required by law.

*The PCM technology is still in development, and there are substantial technical risks that could prevent us from achieving these efficiency gains at scale. Competitive advantages in technology are often temporary, and competitors may develop alternative approaches that match or exceed our efficiency claims. Market adoption of power-efficient AI is not guaranteed, and regulatory, technical, or economic factors could impact the viability of our approach. This technology has not been validated in large-scale commercial deployments, and significant engineering challenges remain before commercial release. Investors should consider this a high-risk, early-stage technology investment with uncertain outcomes.

This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment. For more information about this offering, please view the Offering Circular and Related Risks.