John Martin

I just released DFoS Troposphere v1.0 — a hardened, penetration-proof Linux distribution built around my patent-pending DialectForge protocol. Every process gets memory encryption. Filesystem is fully encrypted and transparent. USB devices auto-quarantine. Network uses dynamic secure protocols with anomaly detection and emergency key rotation. Three-tier trust model (Native / Adopted / Quarantine). Password-protected safe mode included. I baked in a public “Break Me” pentest challenge. Thousands of attack surfaces tested — it’s still standing. This is the working proof-of-concept. A real-world test run for something much bigger I’m building next. The last month nearly broke me. I tested it on a dying laptop while barely able to afford basics. This project pushed me to every limit — financially and physically. I can’t bootstrap the next phase alone anymore. If you’re serious about structural OS security, an investor, or a company that gets what a true patented security moat looks like — download it, test it, try to break it, and DM me. Serious inquiries only. The era of bolted-on security is ending. Download + SHA256 verification link in my first reply 👇 #DFoS #CyberSecurity #SecureOS

I think people may see what I post and think I’m all over the place. I am not. I invented the security protocol. I used it to make the vpn so I could see that it works and find the bugs. Then I made the DFoS kernel to see that I could do that and prove DialectForge works there, then I started DFoS stratosphere which is its own system not requiring Linux, DOS, or Unix. And Now Lu Ai. The prompt injection proof, non llm not needing a transformer, gpu training, or rebuilding of weights. Believe that every piece is connected. LU Ai demo coming soon

@roger @AnthropicAI @Tibbzzee you can also solve the efficiency problem architecturally by making the software behave more like a brain and you stop needing brute force hardware to compensate for a brute force design. I’m training, building and running my ai on a ryzen 9 with 16 gb ram. Never touching my gpu.

@fuzzlongfellow @AnthropicAI @Tibbzzee Also analogue circus no? I’m not any expert but it’s been something I’ve thought about and wondered if both the watts and capabilities would grow if the circuits were analogue. How far off my crackers am I?

We've signed an agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, coming online starting in 2027, to train and serve frontier Claude models.

Just gonna leave this here….

Right now it’s being built, ran and trained on a pc with a ryzen 9 and 16bg ram. It doesn’t take massive amounts of power to run. Training is easy without the need of a gpu. The weights are never frozen so it never needs retraining and actually detects gaps in its knowledge and researches to fill. Soon I will deploy a demo instance on GitHub for people to try.

@fuzzlongfellow @Dr_Singularity What is the physical manifestation of this architecture? How do you deploy it?

wow, insane AI news We may have just crossed the line where AI research becomes automated and self improving. This paper introduces ASI-Evolve, a system where AI doesn’t just use tools… it becomes the researcher. Instead of humans designing better models, AI now runs a full scientific loop on itself: learns from past research designs new ideas runs experiments analyzes results improves itself… again and again It already produced real results: Discovered 100+ new neural architectures Beat human designed improvements by ~3x Improved training data pipelines significantly Invented new RL algorithms outperforming existing ones AI/acc

@brockpierson I miss the simplicity of win 3.1

33 years ago today Windows 3.1 Launched! Did you ever use this OS?

So I have been working on building my Lu AI utilizing my patent pending security protocol. not an llm, no transformer, no need for guardrails and immune to prompt injection. It to has RSI among other features needed to be the natural evolution from llms. I’m training now and will have a demo in a few days.

Recursive Self-Improvement (RSI) is an AI capability where a system iterates on its own code, architecture, or training data to produce a superior version of itself I think we have RSI now

@TiffanyFong It definitely gets old

the older i get, the more i hate alcohol.

@AnthropicAI add a restart item under file menu, that way dispatch can restart Claude with computer use.

@RootMonsteR @three_cube The legacy migration problem is real, and it’s why we designed DialectForge to work as a layer, not a rip-and-replace. Enterprises shouldn’t have to rebuild everything to get quantum resistance.

This is the right approach. Per-session negotiated keys with no static material to harvest is exactly how you future-proof it. The problem is the 99% of enterprises still running legacy crypto that'll take years to migrate. The ones building PQC-native from scratch like you are the exception not the rule.

As I have been saying for the last year, Google researchers now confirm that quantum computers will soon break ECC and RSA cryptography. Really nothing to see here, just the death knell of confidentiality over the internet 😂 x.com/projecteleven/…

Spot on. The harvest-now-decrypt-later threat is already here, not 2029. We’ve been building post-quantum from day one at DialectForge. Per-session negotiated encryption with no static keys to stockpile. The migration doesn’t have to take 3 years if you architect it right from the start

This is an everything problem. ECC and RSA don't just secure one thing, they're behind every bank login, every medical record, every SSH key, every VPN you've ever touched. Google just moved their post-quantum deadline to 2029. Three years to gut and replace the cryptographic backbone of the entire internet. The scary part? Nation states are already stockpiling encrypted traffic right now, just waiting. Your 2026 data becomes their 2029 goldmine.

Thank you @AnthropicAI for the $100 credit. It came at just the right time for me to restart a project from scratch.

@bindureddy What will I build? Already building it. An AI that stores knowledge as discrete atoms instead of smearing it across billions of weights. Every fact individually addressable, removable, verifiable. No retraining. No versions. She just grows.

SPUD is coming 🥔✨ OpenAI's next model isn't just an update—it's 2 years of research, a fresh pre-train. Better reasoning, true context understanding, agentic capabilities that work. This is the bridge to AGI. And it will change everything. What will you build? 🤖

This is exactly the challenge @DialectForge is built to solve. We created a patent-pending AI-negotiated protocol that turns any network traffic into undetectable random noise. No static keys, nothing to harvest, quantum-proof, and it re-encodes itself every ~117 seconds. We even shipped a penetration-proof custom Linux OS with it baked into the kernel. If widely adopted, it would eliminate the need for huge chunks of the traditional cybersecurity industry — endless antivirus/firewall subscriptions, breach monitoring, managed detection services, etc. — because the threats are prevented at the protocol level instead of “managed” for recurring revenue. Same as this gene therapy: curing the problem isn’t a sustainable business model for an industry that profits from the status quo. dialectforge.com

🚨 do you understand what scientists just did to deafness.. researchers injected a modified harmless virus directly into the cochlea the spiral cavity in your inner ear.. carrying a working copy of a gene called OTOF.. the gene that transmits sound signals from your ear to your brain.. without it, your ear hears everything.. your brain receives nothing.. 10 completely deaf patients.. single injection.. within weeks all 10 could hear.. 10 out of 10.. here's what nobody wants to say.. cochlear implants cost between $30,000 and $100,000 per patient.. hearing aids sell for up to $7,000.. the global hearing industry is worth over $9 billion a year.. every year.. recurring.. because deafness has never been cured.. just managed.. one injection ends all of that.. and in 2018 goldman sachs analysts literally wrote this in a report about gene therapy.. "curing patients is not a sustainable business model" that's a goldman sachs equity research note.. sent to investors.. warning them that companies developing one-time cures were a risky bet because cured patients stop buying products.. the science to fix single broken genes has existed in research labs for years.. the same platform used here already cured a form of blindness in 2017.. cured spinal muscular atrophy in babies in 2019.. there are over 10,000 known single-gene disorders.. millions of people labelled "incurable".. the platform exists.. the proof is 10 out of 10.. the question was never whether they could fix it.. it's whether fixing it was good for business.

@theliamnissan Sounds like your now logged into a pineapple

For IT guys above my pay grade. What would cause me to have to re-enter my WiFi password on every device in my house? My Nest thermostat, every laptop, PC, Xboxes, etc. The WiFi password never changed, but I had to reenter it on every device

@coinbase You guys should really have a look at DialectForge.

The internet has protocols for data, email and identity. But not payments. Today the x402 Foundation launches under the Linux Foundation — an open standard for payments over HTTP. Now agents, APIs, and apps can move money as easily as they move data. Learn more ↓

@Prathkum I don’t think you need the source code to build it if you just use Claude, Claude built it in the first place

Now that Claude Code source code is publicly available, the real question is: who is building a better version of it?

I’m really locked in on my Lu AI project right now. You know the one that’s not an LLM, doesn’t need a transformer, has zero guardrails, and never retrains weights. Pure graph based chaos discovery that learns, reflects, and improves under my full control. I’m at the point where I really don’t wanna say much more. Hopefully a demo is coming soon. #LuAI #AGI #LocalAI #Neurosymbolic #ChaosDiscovery #BuiltFromScratch #NoTransformers

These @AnthropicAI Claude limit issues are inconvenient, but also an opportunity to improve my ai and project management skills. Before going to bed I set several projects through dispatch and co work, I set all task to use my Flowsate mcp so that I don’t loose progress. I scheduled a task to check FlowState after reset and continue. I also set cowork up to hand tasks off to ollama models via scripts using some of my in house mcp tools. I actually think this process will allow me to get more done.