Brady Long

🚨 BREAKING: Google Research just dropped the textbook killer. Its called "Learn Your Way" and it uses LearnLM to transform any PDF into 5 personalized learning formats. Students using it scored 78% vs 67% on retention tests. The education revolution is here.

A 27-year-old PhD student at Université de Montréal got into an argument at a bar in 2014 about whether AI could ever generate a realistic face. He went home that night, wrote the code drunk, and accidentally invented the architecture behind every deepfake, every Midjourney image, and every AI-generated face that has ever fooled a human being. His name is Ian Goodfellow. And the story is one of those moments in AI history that sounds made up but is actually documented. May 26, 2014. Montreal. A bar called Les 3 Brasseurs, which translates to The Three Brewers. Goodfellow was out drinking with friends to celebrate a fellow PhD student who had just graduated. He was two weeks away from a NIPS conference deadline he was not planning to submit to. He was supposed to be writing a deep learning textbook with his advisor Yoshua Bengio. Some of his friends were stuck on a problem. They were trying to build a neural network that could generate realistic photos by itself. Every approach they had tried produced blurry faces with missing ears or three arms. They asked Goodfellow how he would fix it. He thought about it over his beer. Then he said, what if you pitted two neural networks against each other? One network generates fake images. A second network tries to tell which ones are fake. They train at the same time. The forger gets better. The detective gets better. Eventually the forger gets so good the detective cannot tell the difference. His friends said it would never work. They told him training one neural network was hard enough. Two was insane. He went home. His girlfriend was already asleep. He sat down at his laptop and coded the entire thing in one sitting through the early hours of the morning. It worked the first time. He called it a Generative Adversarial Network. GAN for short. The paper he wrote in the next two weeks became one of the most cited papers in machine learning history. Yann LeCun, the godfather of convolutional networks, called it the most interesting idea in the last 10 years of deep learning. Every photorealistic AI face you have ever seen comes from this idea. The This Person Does Not Exist website. Every early deepfake. The architecture that would later evolve into the diffusion models behind Midjourney, DALL-E, and Stable Diffusion. All of it traces back to one tipsy moment in a Montreal pub. Goodfellow joined OpenAI as one of its first employees in 2016. He left for Google Brain a year later. He wrote the deep learning textbook that became the standard reference in every graduate program on Earth. In 2019 Apple hired him as Director of Machine Learning in their Special Projects Group. The work was classified. He stayed for three years. In 2022 he resigned over Apple's return-to-office policy. He joined DeepMind that same year, where he still works today. The man who invented the technology behind every deepfake rarely talks publicly about deepfakes. He has said it scares him. He has said GANs were never meant for what they became. A drunk argument in a Montreal bar started the era of AI-generated reality. And the person who started it would probably prefer if he hadn't.

@mil000 Feel like everyone in SF must hate you at this point lol

It’s a D tier accelerator that blows half of their LP money on just the warehouse. It’s not “early hacker Silicon Valley”

Newest company added to my lists: x.com/scobleizer/lis… -- 8,700+ AI companies now on them. Good morning. What a night last night. Took @LinusEkenstam to a secret dinner at @pika_labs where we both got a preview of a new way to build fun videos after being at @fdotinc's startup festival. Blew me away. The only way to keep up now is to have an AI keep up for you. That's what I've built at alignednews.com/ai It even reads other AIs at, say, di.gg and techmeme.com to keep up on what they are finding. In the past few weeks we've seen more than 150 companies launch. And next week even more are launching at the Frontier Tower in San Francisco. Never seen anything like this. AI is speeding everyone up and making new things possible at such a rate. Oh, and X's algo is improving too. I feel sorry for people who are on Instagram or LinkedIn or Reddit. X is way better to keep up with founders, builders, companies, the whole AI industry. It's like staying with an old gas car after many have switched to an autonomous vehicle. I feel sorry for them too. Just did a meeting with @rronak_ who is building a new continually learning AI system that launches next week. This world is speeding up. And I love it.

@RyanDeQuiroz Yeah idk why I forgot to include sleep there too lol

@thisguyknowsai Focus on your sleep for a week and watch how much of a machine you seem to have magically turned into.

self care is not a luxury. It’s a necessity. You can’t expect to do anything great with no exercise, hours of useless scrolling, bad eating and a cloudy mind. So many people are looking for the right workflow to optimize everything other than themself.

What the f*ck just happened 🤯 I've been stress-testing Codex 5.5 for two weeks and the use cases are honestly insane. This is not "write me a Python script" energy. This is "rebuild my entire technical life in a weekend" energy. Here's everything I shipped: → Faster home internet after letting it audit my network stack → Local 6B SLM running 3x faster with zero accuracy loss → MacBook Pro performing like it's brand new again → Lightweight suite for writing and testing Metal kernels → Custom skill that talks to Claude Code in real-time → End-to-end SFT dataset pipeline powered by DeepSeek v4 → Computer use workflow that fine-tunes models inside Google Colab → 4 autopilot routines that test workflows 3 times a day on their own Nobody at OpenAI told me to do any of this. I just stopped asking Codex for snippets and started giving it actual problems to own. The shift happens the second you treat it like a senior engineer instead of an autocomplete tool. Codex 5.5 is not the upgrade everyone expected. It's the upgrade nobody is ready to use properly.

@CavsandEffect str8 up cheek clappin

Aye fucking yo the Spurs are beating OKC’s ass😂😂😂😂

@howietl I may or may not be 1 of the 500 but this is sick

We’re giving away $10,000,000 to founders building agent-first businesses. Autonomous, proactive agents will run tomorrow's companies. We're backing 500 founders building them. The Founding 500. hyperagent.com

A hacker group just stole 4,000 of GitHub's own private repositories and put them up for sale for $50,000, and the way they got in is the scariest part of the whole story. They did not hack GitHub's servers. They poisoned a VS Code extension. One GitHub employee installed it, and the attackers walked through the front door using that employee's own credentials. The group calls themselves TeamPCP. They name their malware after the sandworms from Dune, and they have been running the most sophisticated supply chain attack campaign in the history of cybersecurity. Here is how the whole thing unfolded. In March, they poisoned Trivy, one of the most trusted security scanners in the world, used inside over 10,000 development workflows globally. They injected credential-stealing malware into Trivy's official GitHub Action, and the malware ran silently before the security scan even started. Every log showed "scan completed successfully" while the malware was quietly stealing AWS keys, SSH credentials, database passwords, and Kubernetes tokens in the background. It took Aqua Security five full days to remove them. Using the credentials they stole from Trivy, they breached Cisco Systems and cloned over 300 private repositories, including the source code for unreleased AI products and repositories belonging to Cisco's own customers, which included major banks, government agencies, and BPO firms. In April, they hit Checkmarx, another security vendor, and poisoned five official Docker images in 83 minutes. The scanner worked perfectly. It just silently sent every secret it touched to the attackers. That single breach automatically cascaded into Bitwarden, the password manager, because Bitwarden's CI/CD system pulled the poisoned Docker image and the attackers injected malware into Bitwarden's official CLI package published on npm. One compromised security scanner poisoned a password manager. Automatically. No human involved in the chain. In May, they hit TanStack, a set of libraries downloaded millions of times per week. 84 malicious package versions across 42 packages, and here is the part that should genuinely terrify you. The malware scraped the raw memory of GitHub's build servers, extracted authentication tokens, used those tokens to bypass two-factor authentication, and then published the infected packages with completely valid cryptographic signatures. Every security verification tool on earth said the packages were legitimate, because they were signed by the real pipeline, using real keys. The attackers just happened to be inside the pipeline when it signed them. They defeated the entire trust model of modern software supply chains in a single week. The same week, they hit the Nx Console VS Code extension, which has 2.2 million installations. The malware specifically targeted Claude Code configurations and hunted for AI assistant credentials. That is a first. Supply chain malware engineered to steal your AI's access keys. Then on May 19, they revealed the GitHub breach itself. 4,000 internal repositories listed for sale at $50,000, with a warning attached. If nobody buys it, they leak everything for free. Their malware is self-propagating. Once it infects one package, it automatically finds every other package that developer maintains, steals the publish tokens, and infects all of them. Then those packages infect the next developer, and the next. It jumps between npm and PyPI on its own. The group does not even do the extortion themselves. They sell stolen credentials to ransomware gangs, and one of those gangs used TeamPCP's data to threaten Cisco with leaking FBI and NASA personnel records. The scariest part is the part almost nobody is saying out loud. They did not break any encryption. They did not find any zero-days. They exploited the fact that the entire software industry blindly trusts its own build tools. Every security scanner, every Docker image, every VS Code extension, every GitHub Action is a potential weapon the moment someone poisons it upstream. And right now, nobody can tell the difference between a legitimate build and a compromised one, because the compromised ones have valid signatures too.

@mil000 I'd try mixing it up with something positive. Instead of constant darkness and despair

Stanford just published its 2026 AI Index Report and one number inside it should be read by every parent, every teacher, and every school board member in the country. 4 out of 5 American high school and college students are already using AI for schoolwork. Only 6% of teachers say their school has clear policies on how that should happen. That gap is not a technology problem. It is one of the most significant institutional failures in the history of modern education, and it is happening right now, quietly, inside every classroom in America while the adults responsible for those classrooms are still writing the memos. To understand why this matters, you have to understand the speed at which it happened. Stanford's researchers found that generative AI reached 53% population adoption within three years, which is faster than the personal computer and faster than the internet. Students did not wait for permission. They did not wait for curriculum updates or policy frameworks or guidance from administrators. They found the tools, figured out how to use them, and integrated them into their schoolwork the same way every generation of students has always integrated whatever was available to them, except this time the tool is capable of writing their essays, solving their problem sets, structuring their arguments, and doing the cognitive work that teachers have spent decades designing assignments to produce. Half of all middle and high schools in the United States have no AI policy whatsoever. The ones that do have policies have them in a form so unclear that only 6% of teachers describe them as actionable. Which means that in practice, across almost every school in the country, a student using AI to complete an assignment and a student not using AI are being evaluated against the same standard, measured by the same rubric, and assessed for the same skills, without anyone in the building having agreed on what any of that actually means in a world where the distinction between the two has become nearly impossible to detect. The Stanford researchers also found what students are using AI for, and it is not the edge cases. They are using it for research, essay editing, and brainstorming, which are precisely the three activities that sit at the center of what schools have traditionally used to teach students how to think. Those are not peripheral tasks that AI has quietly automated. They are the core of what academic work was designed to develop, and they are now being handed off at scale to a tool that most institutions have not formally acknowledged exists inside their walls. China and the United Arab Emirates have already responded by mandating AI education starting with the 2025 to 2026 school year. Both countries made a national decision that the right response to students using AI was to teach them to use it deliberately, critically, and with an understanding of what it can and cannot do. The United States has not made that decision at any level of government, which means the first genuinely AI-native generation of workers is being formed right now, mostly without guardrails, and the employers who will eventually hire them have not yet thought carefully about what that means for the skills they will and will not be walking in the door with. The report describes this as an institutional readiness failure. That is the polite version of what it is.

@thisdudelikesAI I’m gonna save this!

What do you guys think about this? To be fair I think there’s more nuance. I agree writing can be a powerful tool for self. But I don’t think writing is as much a tool for critical thinking as critical thinking is a tool for writing. And historically, the only reason people learned to write was to spread ideas to others.

@bigaiguy Kinda valid

10 WEBSITES EVERY INTERNET USER SHOULD CHECK TONIGHT. Bookmark all of them. Most people don't know half of these exist. 1. haveibeenpwned.com Shows every data breach your email is in and what got leaked. 2. behindtheemail.com Shows every social profile, photo, and login tied to an email address. 3. tempmail.com Free disposable email for any signup you don't trust. 4. 10minutemail.com Burner inbox that self-destructs in 10 minutes. 5. justdeleteme.xyz A directory of direct links to delete your account from any major service. 6. exposing.ai Check if your face was used to train AI image models without consent. 7. dnsleaktest.com Tells you if your VPN is actually hiding your real location or leaking it. 8. amiunique.org Shows how trackable your browser fingerprint is, even in incognito mode. 9. shouldiremoveit.com Tells you which programs on your PC are useless bloatware or spyware. 10. virustotal.com Drop any file or link. It scans against 70+ antivirus engines instantly. The internet is hostile by default. These websites are your free defense.

SHOCKING: Claude has a secret mode called “Startup Lab.” It turns Reddit complaints, customer reviews, and your own skills into startup ideas with MVPs, pricing, and validation plans. Here are the 12 prompts founders are using before building anything:

We're hiring for our Enterprise Operations team! You'll be assigned to our most valuable accounts and your job is to make their experience seamless. What it takes: - Operator mentality. You enjoy fixing problems. - On call. Things can break at any time. Can travel to customer sites ~one week every 45 days. - Leader. You've managed a 25-person team successfully. - Financially sharp. Can sit with a CFO and add value. - Enterprise-oriented. You understand how MNCs operate & decide. - AI-native. Our tools and workflows are AI-first. You should be too. - Self-directed. Every customer at this stage has unique and precise needs, so there's no unique playbook. Get inside the customer's business and understand how they operate. This means being on top of everything. To the point where you anticipate their needs and solve them proactively. If there's a problem you can't solve yourself, you know who on our team to pull in. You'll own the customer end to end - fully responsible for their experience. People in this team are extremely smart and hard-working. If that's you, join them!

Looking for more posts like this? Follow me @thisguyknowsai and you'll get 1x post every single day. And don't forget to repost the thread to reach more. Thanks for checking!

Prompt 12: The Kill Switch Test "Here is my startup idea: [describe it]. Try to kill it. Give me the top 5 reasons this will fail, the most common objection a customer will have on a sales call, and what I need to prove in the next 30 days to know this is worth continuing." The best investors stress-test every idea before funding it. Do this before you build. Not after.