Sinclair Ta

Ray Bradbury understood something most people forget: attention is not just something you spend, it is something you train. His advice was simple but disciplined — read one short story, one poem, and one essay every night for a thousand nights — because the mind gets stronger by repeated contact with concentrated language and ideas. The deeper point is that rebuilding attention span is not about avoiding distraction for a week and calling it fixed. It is about giving your mind regular doses of depth until depth feels natural again. Bradbury’s method works because it mixes three things the brain needs: narrative, compression, and argument. What I like about this advice is that it treats attention as a craft, not a mood. You do not rebuild it by waiting to feel focused. You rebuild it by feeding the mind better material every day, long enough for your defaults to change. The image captures that well: attention span is not only about reading more, it is about reclaiming the ability to stay with something long enough for it to shape you. In Bradbury’s view, the real payoff is not just more focus — it is a fuller head, a richer inner world, and better raw material for thinking and writing.

Absolutely my way of thinking.💪 Getting back on track quickly is so important...😉

Anthropic’s move looks less like generosity and more like a calculated trial window: it lets paying users stress-test Claude Fable 5, gather proof that it is worth the cost, and ease backlash before the product shifts into metered usage. That matters because the model appears expensive to serve, and the free extension functions like a sampling period rather than a promise of unlimited access. The question is what this says about Anthropic’s product strategy. If a company keeps extending free access to a frontier model, it usually means one of three things: it wants adoption fast, it needs more real-world benchmarking, or it is still balancing capacity and pricing against user resistance. Here, the likely answer is all three — but the strongest signal is that Anthropic wants customers to feel the model’s value before the bill arrives. The model does not disappear; access simply becomes credit-based, with post-promo usage billed separately rather than bundled into the standard plan. In practice, that means heavy users, enterprise teams, and anyone doing long-context or agentic work will need to budget carefully or fall back to cheaper models. Anthropic is signaling that its best model is powerful enough to justify premium billing, but still not comfortable enough to leave fully unlimited inside flat-rate plans. So the extension is basically a bridge: one last chance to experiment widely before the product becomes a more tightly metered, more explicitly monetized frontier system. This also reveals fragility. If a company must repeatedly extend a “free” window, it suggests the launch economics or user expectations are not settled yet, and that the transition to paid usage may be bumpier than the marketing implies. In other words, the extension is friendly on the surface, but underneath it is a sign that Anthropic is still negotiating the tension between hype, cost, and adoption.

The Real Secret to High Performance Is What You Do Every Day The core message here is simple: most people are not limited by talent, but by the habits, fears, and environments that quietly keep them underperforming. The psychologist’s argument is that excellence is not a mood or a burst of motivation — it is a daily practice of choosing discomfort, staying present, and acting on purpose even when it feels awkward. What makes this idea powerful is how practical it is. He says people rise when they stop waiting to feel ready, stop caring so much about outside approval, and stop living in the emotional residue of past failures. In his view, confidence is not just a personality trait; it is built through mastery experiences, social reinforcement, comparison, and how you interpret your own physical state under pressure. The deeper insight is that performance is shaped as much by identity as by skill. People who reach the top usually do not just want the outcome — they are drawn to the craft itself, and they learn to love the process, not just the reward. That is why he keeps coming back to flow, presence, and solving hard problems together: if the work feels meaningful and the environment rewards truth and growth, people can do far more than they think. What stays with me most is his warning that fear shrinks your world. If you care too much what people think, overreact to setbacks, or define yourself by old wounds, you end up building a ceiling around your own potential. The alternative is quieter but stronger: small wins, repeated daily, in a setting that lets you improve without being crushed by mistakes. High performance is not about becoming a different person overnight; it is about building the habits, confidence, and environment that let your best self show up repeatedly. buff.ly/0oNJ36X

@SahilBloom There’s something powerful about being comfortable with being unfinished. Growth gets easier when you stop performing perfection and start embracing iteration. You don’t need all the answers, you just need enough humility to keep learning and enough discipline to keep moving.

A mentor once told me this: Fall in love with feeling like a work in progress. Resist the urge to look polished. You don’t need to pretend you have it all figured out. Nobody does. Embrace your unfinished form. Just commit to getting a better each day and trust where it leads.

The best startups rarely begin by trying to conquer the world. They begin by solving one painful problem for a small group of people unusually well. Precision creates traction. Traction creates insight. Insight creates expansion. Grand visions are often discovered after the small thing starts working.

An initial startup idea can't usually be both grand and precise. In practice they're usually either grand and vague or precise and small. Precise and small is better. You know who your initial users are, and you expand outward. With grand and vague you can't even get started.

@pmarca The bigger shift won’t be when AI stops sounding like AI. It’ll be when we stop caring who wrote it and start judging ideas purely by their clarity, originality, and usefulness. Style imitation is the easy part. Genuine insight is still the real moat.

Peter Thiel’s Real Message: Build Something No One Else Can Copy Peter Thiel’s core argument is that the best businesses are not the ones fighting hardest in crowded markets, but the ones creating their own category and then defending it with something truly hard to replicate. He keeps circling back to one big idea: a company should aim to become a creative monopoly, meaning it offers something so distinctive that no close substitute exists. What makes that idea powerful is how practical it is. Thiel says the strongest businesses usually combine a few durable advantages — proprietary technology, network effects, economies of scale, and branding — instead of relying on generic execution alone. In his view, the real mistake founders make is chasing obvious competition instead of asking, “What valuable company is nobody building?”. He also argues that great companies start small on purpose. Rather than launching broad and vague, the smartest founders begin with a narrow market, dominate it, and only then expand outward in sequence. That’s why he treats sales and distribution as part of product design, not an afterthought, because even the best product fails if nobody can be convinced to adopt it. The deeper theme is that good founders are not lottery tickets — they make plans, choose people carefully, search for hidden opportunities, and commit to a long game. Thiel’s real challenge to readers is simple: don’t imitate the crowd, don’t optimize for short-term applause, and don’t build something replaceable. Build something rare enough that the world would miss it if it disappeared. buff.ly/wDJdLlu @FoundersPodcast

Really solid primer because it connects how AI models work technically with what actually matters economically. What stood out to me is the shift from treating AI as a mysterious black box to understanding the mechanics underneath: tokenization, inference costs, agentic workflows, context windows, and the trade-offs between open-source and proprietary models. Those details matter because every architectural choice eventually shows up somewhere else, as latency, compute cost, data risk, maintenance overhead, or product capability. The most important takeaway for me is that AI economics will increasingly shape AI architecture. A model can be technically impressive, but if it costs too much to run at scale, requires excessive human intervention, or creates poor unit economics, it may still be the wrong system for the job. So yes, this is a good piece for understanding AI models, but even more importantly, it helps explain why the future of AI will not be decided by model intelligence alone. It will also be decided by efficiency, infrastructure, openness, and economics.

Get it here: buff.ly/HwhR0yC

A very big deal for scientific AI. 15 terabytes of physics simulations are now openly available. But the size is probably the least interesting part of the story. The Well, created by Polymathic AI with researchers across institutions including the Flatiron Institute, Princeton, Cambridge, NYU, Berkeley and Los Alamos, brings together 16 large-scale simulation datasets spanning fluid dynamics, acoustic scattering, active biological matter, reaction-diffusion systems, magnetohydrodynamics, stellar evolution and supernova physics. That alone sounds impressive. But what makes The Well genuinely useful is not just volume. It is standardization. The datasets range from roughly 6.9 GB to 5.1 TB, yet they are organized through a shared data specification, stored in HDF5, sampled on uniform spatial grids at constant time intervals, and exposed through a unified PyTorch interface. In practical terms, that means researchers can move between very different physical systems without rebuilding an entirely new data pipeline every time. And that matters because scientific AI has a very different data problem from language or vision. A language model can learn from huge amounts of naturally occurring text. A vision model can train on billions of existing images. But high-quality physics data often has to be created deliberately by solving complex numerical equations on expensive computing infrastructure. In some cases, even generating one simulation dataset can be costly. For example, one of The Well's Euler-equation datasets was generated in double precision over 80 hours using 160 CPU cores. Other datasets involve vastly more complicated systems, including relativistic magnetohydrodynamics around black holes and simulations of stellar explosions. This is where machine-learning surrogate models become interesting. A traditional numerical solver repeatedly computes how a physical system evolves according to its governing equations. That may be extremely accurate, but also computationally expensive. A neural surrogate instead learns an approximation of that evolution from simulation data.

I didn’t realize this was something people got help with. She never helped with school stuff either. I just thought that we were supposed to figure it out on our own

Yep, he’s spot on there👌

My thoughts The most interesting part of this story isn't simply whether a few files were allegedly taken. It is the increasingly difficult question of where human expertise ends and corporate intellectual property begins. When someone spends 10, 20 or 24 years at a company, they inevitably carry knowledge with them when they leave. You cannot erase experience from a person's brain. Silicon Valley itself was built on talented people moving between companies, taking what they had learned and using that experience to build something new. But there is an obvious line between taking your skills with you and physically taking confidential files, prototypes, proprietary specifications or unreleased product information. Apple's lawsuit alleges that line was crossed repeatedly and deliberately. Whether it can prove that is now the crucial question. I also think there is a much bigger strategic story here. For years, AI companies mainly threatened software businesses. Now they are beginning to challenge the interface through which we interact with technology itself. OpenAI doesn't necessarily need to build a better iPhone to threaten Apple. It may simply need to create a device or interface that makes opening individual apps, navigating screens and perhaps even carrying a conventional smartphone feel less necessary. That is the real existential threat. Apple owns one of the most valuable interfaces in human history: the smartphone screen. OpenAI's long-term ambition appears to be moving beyond being an app that lives on someone else's device toward controlling a new interface of its own. And that turns a former partner into a potential rival. My biggest takeaway is this: **the next phase of the AI race won't just be fought over who has the smartest model. It will be fought over who controls the device, the interface and ultimately the relationship between humans and AI.** That is why this lawsuit matters far beyond the courtroom. The irony is hard to miss: some of the people trying to build the device that could potentially challenge the iPhone are the same people who spent years helping Apple perfect it. Whether that's healthy talent mobility or illegal appropriation of trade secrets is precisely what the court will now have to decide.

Apple just sued OpenAI over alleged trade secret theft—and this is much bigger than a normal tech lawsuit Apple has filed a lawsuit against OpenAI, accusing the AI company of misappropriating trade secrets and breaching contracts as it builds out its own consumer hardware ambitions. The case was filed on July 10, 2026, in the U.S. District Court for the Northern District of California. At the center of the lawsuit are two former Apple employees: Tang Tan, who spent 24 years at Apple and most recently served as vice president of product design for the iPhone and Apple Watch, and Chang Liu, a former senior systems electrical engineer who worked at Apple for eight years. Both later joined OpenAI. Apple’s allegations are serious. According to the complaint, Tan allegedly used confidential Apple project code names during OpenAI's recruitment process, asked candidates to bring Apple hardware components to interviews, sought information about unreleased products and even coached departing employees on how to get around Apple’s security procedures. Liu, meanwhile, is accused of failing to return an Apple-issued laptop after leaving the company and using it to download confidential technical documents related to unannounced technologies, engineering specifications and proprietary projects. Apple also alleges that he shared confidential information with other Apple employees seeking jobs at OpenAI and advised at least one candidate on what to study before an interview. Perhaps the most consequential claim is that Apple believes this was not simply the misconduct of a few rogue employees. The company alleges a broader pattern involving OpenAI's recruitment practices and senior leadership, and claims that some of its confidential information has already been used in the development of OpenAI's hardware products. Apple says it raised concerns directly with OpenAI in February but received no response. OpenAI has denied any interest in competitors' trade secrets and says it remains focused on creating innovative technology. These allegations have not been proven in court, and at this stage they remain Apple's claims. What makes this case particularly fascinating is the context. Apple and OpenAI were partners not long ago, with ChatGPT integrated into Apple's ecosystem beginning in 2024. But OpenAI has since made a major push into hardware, including its multibillion-dollar acquisition of io, the hardware startup associated with legendary former Apple design chief Jony Ive. io Products is also named in Apple's complaint, though Ive himself is not. So this is no longer simply Apple versus an AI software company. It is potentially Apple versus a future hardware competitor built by some of the very people who once helped create Apple's most important products. buff.ly/2R737Av

Davide Mastracci -@DavideMastracci has written a genuinely useful guide on how to archive online content—and why more people should treat archiving as a basic digital habit rather than a niche skill for journalists or researchers. His central warning is simple: what exists online today may not exist tomorrow. Articles are edited, posts are deleted, accounts disappear, websites shut down, and valuable evidence can vanish with almost no notice. Mastracci argues that when you come across something important, especially something potentially controversial or likely to be deleted, you should archive it immediately. Not after posting about it. Not later that evening. Immediately. He also makes an important distinction between simply saving something and preserving it in a way that others can later verify. Screenshots are convenient, but weak as long-term evidence. They can be altered, fabricated or buried on someone’s device. Screen recordings are somewhat better, but still difficult for others to independently find and verify. That is why Mastracci recommends public archiving services such as the Wayback Machine, Archive Today and Perma.cc. The Wayback Machine is especially valuable because it allows anyone to save a webpage, revisit older versions and trace how a page may have changed over time. Archive Today can complement it, particularly when certain social media posts or webpages are not captured well elsewhere. Mastracci’s broader point is that redundancy matters: when something is truly important, relying on a single archive is risky. For audio and video, the challenge becomes harder. Traditional archiving services often preserve the webpage but not necessarily the playable media embedded within it. Mastracci points to tools such as Cobalt and yt-dlp for downloading copies directly, while also noting the legal and copyright questions that can arise when those files are redistributed publicly. What stood out to me most is that this is really a piece about collective memory. The internet often gives us the illusion that everything is permanent because everything feels searchable. But searchability is not permanence. A deleted post, an edited article or a vanished website can quietly change the public record. That matters not only for journalists, researchers and activists, but for anyone who cares about accountability, history and evidence. Archiving is not glamorous. It is often invisible work. But sometimes the reason we are able to prove what happened, understand how a story evolved or hold powerful people and institutions accountable is simply because someone thought to save the page before it disappeared. Davide Mastracci’s guide is a good reminder that preserving the internet should not be left entirely to institutions. Ordinary people can contribute too. The more people who build the habit of archiving important material, the harder it becomes for history to be quietly rewritten by deletion. The line I would personally take away from the whole piece is: The internet remembers less than we think, unless someone deliberately chooses to remember for it. buff.ly/figCWgH

At the Forbes Self-Made 250 Celebration, Joseph Neubauer, former chairman and CEO of Aramark and No. 214 on the Forbes Self-Made 250 list, sat down with Forbes senior writer Jabari Young to discuss mentorship, ambition and the value of understanding how businesses operate from within. Looking back on an early-career exit interview with David Rockefeller, Neubauer explained why he left Chase Manhattan Bank to join PepsiCo. He realized he did not want to remain a high-level adviser standing on the sidelines. He wanted to take action, make decisions and become a doer. Maintaining and scaling sovereign institutional leverage over a 101-year lifespan is a masterclass in elite execution. David didn't just sit on cash flow. He took Chase Manhattan Bank and aggressively weaponized it internationally...forcing American banking infrastructure into the Soviet Union and China during the height of the Cold War.

Expensive pain, pay them taxes!

Corporate athletes🤣

where on earth is this??!

GPT-6 may not just be smarter — it may be able to keep learning after it’s deployed. SEAL, the new MIT paper on self-adapting language models, describes a setup where an LLM generates its own “self-edits,” creates finetuning data, and updates its weights so the adaptation persists instead of resetting after each session. That is a big shift, but “alive” is still a metaphor, not a literal claim. What the paper shows is not consciousness or biology; it is a model that can participate in its own improvement loop by proposing changes, getting updated, and using downstream performance as feedback. In that computational sense, the system becomes less like a fixed tool and more like a machine that can revise itself in response to experience. The interesting part is how close that gets to something people intuitively call learning. Traditional LLMs are mostly static at deployment, while SEAL is built to turn new inputs into durable parameter updates, which means future behavior can genuinely change because of past interactions. That opens the door to more adaptive assistants, but it also raises harder questions about reliability, memory, and control. So the real headline is not “the model is alive.” It is that AI may be moving from one-shot prediction toward ongoing self-modification, which makes it feel more agentic, more cumulative, and more difficult to treat like a frozen product. That is why papers like this matter: they hint at a future where models do not just answer faster — they evolve while they are already in use. buff.ly/Cni92NL