I will be doing everything within my power to ruin Trump’s UFC event tonight.

This is a special kind of loathsomeness: a blend of historical deafness , grotesque stupidity, and comically ludicrous self- importance . As if the little people's rage against immigration somehow is superior to the war against the 3rd Reich and entitles this comic book nobody to lecture the actual heroes

Earlier this year I was getting frustrated with Claude's charts, fed this book to claude and had it generate a Tufte skill. Instantly got simpler/more beautiful visualizations. gist.github.com/aparente/e48…

Elon Musk is the Ivar Kreuger of our time, and the OpenAI trial is PROVING it in real time. If you don't know who Kreuger was, you should: In the 1920s he was the most admired businessman in the world. The "Match King." He controlled 90% of global match production, lent money to sovereign governments, and his securities were the most widely held in America. But after his death in 1932, auditors spent 5 years untangling over 400 subsidiary companies and discovered the whole thing was held together with fictitious assets, forged bonds, and the unquestioning loyalty of people too dazzled to ask questions. Investors lost $750 million (~$17 billion in today's money). His deficits exceeded Sweden's national debt. Doesn't this sound familiar? The Musk playbook is the most DANGEROUS house of cards I've witnessed in my career. This week in federal court, Musk took the stand to argue that Sam Altman stole a charity. 3 days later he'd contradicted himself under oath so many times that the judge told his lawyers she suspected plenty of people don't want to put the future of humanity in Mr. Musk's hands. OpenAI's attorney asked if Tesla is pursuing AGI. Musk said no. The attorney then pulled up Musk's OWN post from March 4 where he wrote Tesla will be one of the companies to make AGI. His own words entered into evidence against him. BY HIM. Then the attorney asked if xAI used OpenAI's models to train Grok (which violates OpenAI's terms of service). Musk called it a general practice among AI companies. Pressed for a direct answer, he said "partly." Think about that: Musk is in court accusing OpenAI of betrayal while admitting under oath that xAI violated the very same company's terms of service to build Grok. Then came the credibility test: Musk was asked to name his companies that benefit society. He listed Tesla, SpaceX, Neuralink, and X without hesitation. Every one of them is an uncapped for-profit enterprise. Then why did xAI start as a benefit corporation and quietly flip to a for-profit C-corp? No clean answer. This is someone who repeatedly launches entities with noble-sounding charters and converts them into for-profit corporations once the money gets serious. Then his money manager Jared Birchall took the stand: OpenAI's lawyer asked about the donor-advised funds at Vanguard and Fidelity that Musk used to send his $38 million. Did Musk have any legal right to direct where the money went once it entered the DAF? Birchall couldn't answer. Said the legal question was beyond his expertise. The entire lawsuit hinges on that donation creating enforceable obligations. But the man who managed Musk's money just told a federal jury he can't confirm Musk had any enforceable claim over those funds. Now step back... This is a man who promised full autonomy by 2018, a million robotaxis by 2020, and unsupervised FSD by June 2025. EVERY deadline was missed. He claimed he invested $100 million in OpenAI. The real number was $38 million. His defense? His "reputation" made up the difference. Kreuger had 400 subsidiaries and used one entity to prop up another through structures nobody could follow. Musk has Tesla, SpaceX, xAI, Neuralink, the Boring Company, and X. He shifts AI talent from Tesla to xAI, has xAI building the brains for Tesla's Optimus robot, and uses X as a megaphone while the algorithm amplifies his narrative to 200 million followers. Kreuger's investors trusted the man, NOT the math. They loved the confidence. They stopped asking questions because the aura of genius made questioning feel foolish. The same psychology applies to Musk's empire today. Kreuger's reckoning took 5 years of forensic auditing after his death. But Musk is providing his in REAL TIME: contradicting his own posts under oath, admitting to the practices he's suing others for, watching his logic collapse under cross-examination. Different decade. Different industry. Same ending. The truth always catches up.

Diligence in 2026 is wild. My friends in PE are now spending the weekend before IC trying to rebuild the company they're acquiring in Claude Code. If the clone works, the deal dies. Cheapest moat test in human history.

Ashley St. Clair was so deep in the right wing media apparatus that she had a baby by Elon, but since they threw her under the bus she’s been spilling all their secrets and it’s very entertaining

WATCH: The White House took down this video, but we still have it. Trump: We can't take care of daycare. We're a big country. We're fighting wars. It's not possible for us to take care of daycare, Medicaid, Medicare, all these things.

Running 400B model on iPhone! 0.6 t/s Credit @danveloper @alexintosh @danpacary @anemll

Tuesday night, protesters packed the City Council meeting in Surprise, Arizona, over a proposed ICE detention center. You need to hear what this man told the Council.

NVIDIA just dropped a banger paper on how they compressed a model from 16-bit to 4-bit and were able to maintain 99.4% accuracy, which is basically lossless. This is a must read. Link below.

I managed to unlock a crazy new hidden feature in Claude Code called Swarms. You're not talking to an AI coder anymore. You're talking to a team lead. The lead doesn't write code - it plans, delegates, and synthesizes. When you approve a plan, it enters a new "delegation mode" and spawns a team of specialists who: - Share a task board with dependencies - Work in parallel as teammates - Message each other to coordinate work Workers do the heavy lifting, coordinate amongst themselves, then report back.

Remarkable speech from @MarkJCarney, urging allies to stop playing along with great-power intimidation. Compliance won’t buy safety, he warns. Stop appeasing bullies. Carney draws applause. Citing Václav Havel, he says it’s "time for companies and countries to take their signs down." Under communism, ordinary people displayed slogans like "Workers of the world, unite!" even when they didn’t believe them, rituals of compliance meant to avoid trouble. Coercion, Havel argued, wasn’t always enforced through violence, but through quiet, everyday participation in a lie. Carney receives a standing ovation at the end of his speech. Extraordinary to see Canada asserting this kind of moral and strategic leadership, something we haven’t witnessed on the world stage in decades, especially at a moment when we have so much to lose if Trump chooses retaliation.

Amazing moment on French TV. A French judge explains how Trump sent people from the US Embassy basically trying to intimidate her during Le Pen's trial for embezzlement - something they've done to other judges around the world My English sub-titles 👇

Yesterday a quasi-judicial body in Italy fined @Cloudflare $17 million for failing to go along with their scheme to censor the Internet. The scheme, which even the EU has called concerning, required us within a mere 30 minutes of notification to fully censor from the Internet any sites a shadowy cabal of European media elites deemed against their interests. No judicial oversight. No due process. No appeal. No transparency. It required us to not just remove customers, but also censor our 1.1.1.1 DNS resolver meaning it risked blacking out any site on the Internet. And it required us not just to censor the content in Italy but globally. In other words, Italy insists a shadowy, European media cabal should be able to dictate what is and is not allowed online. That, of course, is DISGUSTING and even before yesterday’s fine we had multiple legal challenges pending against the underlying scheme. We, of course, will now fight the unjust fine. Not just because it’s wrong for us but because it is wrong for democratic values. In addition, we are considering the following actions: 1) discontinuing the millions of dollars in pro bono cyber security services we are providing the upcoming Milano-Cortina Olympics; 2) discontinuing Cloudflare’s Free cyber security services for any Italy-based users; 3) removing all servers from Italian cities; and 4) terminating all plans to build an Italian Cloudflare office or make any investments in the country. Play stupid games, win stupid prizes. While there are things I would handle differently than the current U.S. administration, I appreciate @JDVance taking a leadership role in recognizing this type of regulation is a fundamental unfair trade issue that also threatens democratic values. And in this case @ElonMusk is right: #FreeSpeech is critical and under attack from an out-of-touch cabal of very disturbed European policy makers. I will be in DC first thing next week to discuss this with U.S. administration officials and I’ll be meeting with the IOC in Lausanne shortly after to outline the risk to the Olympic Games if @Cloudflare withdraws our cyber security protection. In the meantime, we remain happy to discuss this with Italian government officials who, so far, have been unwilling to engage beyond issuing fines. We believe Italy, like all countries, has a right to regulate the content on networks inside its borders. But they must do so following the Rule of Law and principles of Due Process. And Italy certainly has no right to regulate what is and is not allowed on the Internet in the United States, the United Kingdom, Canada, China, Brazil, India or anywhere outside its borders. THIS IS AN IMPORTANT FIGHT AND WE WILL WIN!!!

If you are using a relational database of any kind, you should read “is query optimization a solved problem?” By Guy Lohman. It’s a classic, and for a good reason. Guy claims that query optimization barely advanced since 1989 because researchers are focusing on the wrong problems. He focuses on 3 key issues with query planning: a) host variables and parameter markers (b) the selectivity of join predicates (c) how we combine selectivities to estimate the cardinality For each one, he cites leading research in the area and then gives examples of rather basic problems that the research still doesn’t solve. The depressing part is that the paper in from 2014 and the situation isn’t much better in 2025. Still, very worthy read.

Such a great platform. There was a report of fireworks at the event that turned out to be about gunfire, and these chuckleheads went out and found a video of fireworks and pretending it's the truth. @grok care to check the facts?

This from @obie captures how I'm feeling about coding agents

ICE cut off her wedding ring. Sue Tincher is a 55-year-old American citizen: a grandmother, 5’4’, and white. Just in case you thought you’d be safe. She walked to her neighbor’s house after getting alerts that ICE was nearby. She stood across the street and asked an officer if they were ICE. They told her to “get back.” She didn’t move. Seconds later, they threw her to the ground, handcuffed her, and hauled her away. She spent five hours in leg shackles at a federal building. Agents cut off her wedding ring and threatened to pepper-spray her in the truck. Her husband spent all day trying to find where they’d taken her. Federal officials wouldn’t tell him. Her "crime" was simply standing on a public street, watching, and asking questions. Read that again: here in Minnesota, a U.S. citizen was arrested, restrained, and disappeared for hours, not for interfering, not for resisting, but for asking a question. If they can arrest Sue Tincher for standing on a public sidewalk, they can arrest anyone. Immigration attorneys said they’re seeing constitutional violations every single day now. I’m running because we need leaders who will call this what it is- un-American, unconstitutional, and unacceptable. Sue Tincher stood up. I’m standing up. We must all stand up. mprnews.org/story/2025/12/09…

Imo this is along the lines of how talking to an LLM via text is like typing into a DOS Terminal and "GUI hasn't been invented yet" of some of my earlier posts. The GUI is an intelligent canvas.

The recently released DeepSeek-OCR paper has huge implication for AI memory, long‐context problems and token budgets. It frames the OCR model not only as a document‐reading tool but as an experiment in how models can “remember” more by storing data as images rather than text tokens. With this paper, DeepSeek really found a new way to store long context by turning text into images and reading them with optical character recognition, so the model keeps more detail while spending fewer tokens. DeepSeek's technique packs the running conversation or documents into visual tokens made from page images, which are 2D patches that often cover far more content per token than plain text pieces. The system can keep a stack of these page images as the conversation history, then call optical character recognition only when it needs exact words or quotes. Because layout is preserved in the image, things like tables, code blocks, and headings stay in place, which helps the model anchor references and reduces misreads that come from flattened text streams. The model adds tiered compression, so fresh and important pages are stored at higher resolution while older pages are downsampled into fewer patches that still retain gist for later recovery. That tiering acts like a soft memory fade where the budget prefers recent or flagged items but does not fully discard older context, which makes retrieval cheaper without a hard cutoff. Researchers who reviewed it point out that text tokens can be wasteful for long passages, and that image patches may be a better fit for storing large slabs of running context. On the compute side, attention cost depends on sequence length, so swapping thousands of text tokens for hundreds of image patches can lower per step work across layers. There is a latency tradeoff because pulling exact lines may require an optical character recognition pass, but the gain is that most of the time the model reasons over compact visual embeddings instead of huge text sequences. DeepSeek also reports that the pipeline can generate synthetic supervision at scale by producing rendered pages and labels, with throughput around 200,000 pages per day on 1 GPU. The method will not magically fix all forgetting because it still tends to favor what arrived most recently, but it gives the system a cheaper way to keep older material within reach instead of truncating it. For agent workloads this is appealing, since a planning bot can stash logs, instructions, and tool feedback as compact pages and then recall them hours later without blowing the token window. Compared with vector databases and retrieval augmented generation, this keeps more memory inside the model context itself, which reduces glue code and avoids embedding drift between external stores and the core model. --- technologyreview .com/2025/10/29/1126932/deepseek-ocr-visual-compression