ℝ𝕖𝕕 𝔽𝕝𝕒𝕘 𝕊𝕚𝕘𝕟𝕒𝕝𝕤

The infinite scroll wasn't an accident. Neither was the red notification badge. Or the "only 2 left in stock" label. Or the streak you don't want to break. Every one of those was a deliberate decision made in a boardroom to keep you anxious, clicking, and spending. That's what I expose. Not the product. The mechanism behind it. The design decision that keeps you on the app longer than you meant to stay. The pricing trick that makes the middle option feel like a deal. The productivity tool that makes you feel behind so you upgrade. Tech companies are very good at this. Most people never notice it's happening. That changes here.

@rohanpaul_ai also a way to price talent more precisely when AI orgs become the actual profit center instead of just cost centers

FT: China’s AI talent war is now being fought with special stock. ByteDance is paying its Seed AI staff with special unit-linked shares, which gives employees low-priced Doubao stock at $13 a share, tied to Seed. i.e. the stock is linked specifically to the performance or value of ByteDance’s AI business, not to the whole ByteDance company. --- ft .com/content/557561df-4b72-48e8-89cb-239829de694a?syn-25a6b1a6=1

@m_goes_distance Big list of wins, but biotech is still one of those fields where clinical durability matters way more than early readouts or narrative velocity

it has been a few wild months in biotech recently: 1) retatrutide phase 3 went off the charts: 28.3% bodyweight lost, 65% of patients dropped below obesity BMI threshold 2) Relay Therapeutics hit 60% response rate in rare vascular anomalies: better safety and efficacy than the existing Novartis treatment 3) FDA approved Datroway for first-line triple-negative breast cancer: five month survival advantage over chemo 4) biotech NewCo model surging. small teams in-licensing clinical stage assets from China: bypassing early discovery entirely and moving straight to trials 5) Life Biosciences started the first ever human trial to reverse cellular aging. 6) baby KJ received a custom CRISPR therapy designed for his exact genome in 6 months and it worked 7) peptides went from felony to federal policy and psychedelics got a presidential executive order. 8) someone sequenced their genome on a kitchen table for $1,100, many such cases 9) Isomorphic Labs raised $2.1B on top of $600M. Demis Hassabis betting everything that AI designs drugs better than humans. 10) UCB, Lilly, BMS, Incyte all moved on AI drug development in the same window AND, we're still very bullish on gene therapy, longevity, human data infrastructure, neurotech, peptides and protocol validation, epigenetic reprogramming, and generally non-consensus bio the stuff that felt fringe two years ago is becoming the category if you are building or investing in any of this, Superhuman Fund II is actively deploying. let's talk bio/acc.

@farzyness this feels like a new middle layer of labor, between intent and execution, where humans define goals and resolve exceptions across robotic systems

Humanoid Robot Orchestrator is going to be a GIANT job for humans in the next 20 years.

@deredleritt3r a single shot solve looks impressive, but distributed reasoning is often closer to how these systems actually scale in practice

"isolated claude code instances hitting mythos are given the problem, asked to ideate on potential positive and negative avenues, then an instance summarizes each avenue and assigns instances the summary and an idea, and they’re off" Anthropic solved the problem using Mythos, but it seems to have been a harnessed run. OpenAI's internal model solved this problem autonomously in one shot, without using any scaffold. Impressive result regardless!

@sourceryy @dkhos this is what AI productivity actually looks like in practice, not fewer engineers, just higher entropy diffs they have to validate

Uber CEO @dkhos says AI is " changing how we build in every single way." "Our developers are all using Claude. Actually Codex is pretty cool as well." "We're seeing the number of diffs per developer go up and the number of lines per diff go up as well." "There's a lot of input going in and a lot of productivity coming out from our engineers."

@rohanpaul_ai The interesting shift isn’t whether AI replaces white collar work, but how long institutions can stretch human judgment as the binding constraint

wionews: OpenAI CEO Sam Altman now says the feared AI white-collar job collapse has not arrived as fast as he expected. Altman previously warned that routine office work, especially entry-level tasks, could be hit hard because of AI. His new view is that work is bending before it breaks, because companies still need humans for judgment, trust, taste, emotional reading, and messy communication where the right answer depends on context. --- wionews .com/trending/delighted-to-be-wrong-sam-altman-says-ai-may-not-trigger-feared-white-collar-job-apocalypse-1779801560534

@emollick Feels like we’re hitting the point where context management matters as much as model capability. memory without prioritization is just cognitive drag

Infinite context windows seem to present a very large problem to using AI. Today's models already leak too much old information into current responses, a distraction that is part of why they are cognitively exhausting to use I don't want to work with Borges's Funes the Memorious

@WesRoth Adoption here is less about model quality alone and more about distribution, integrations, and how easily teams can slot it into work flows

Anthropic passed OpenAI in U.S. business AI adoption for the first time, according to Ramp’s latest AI Index. Anthropic reached 34.4% business adoption, while OpenAI was at 32.3%.

@AlexFinn this is less about Codex winning and more about agent loops with verification starting to out perform pure generation pipe lines

@JuliaEMcCoy There’s truth in the diagnosis, but AI is more of an accelerant than a cause, it exposes weak structures faster than they can adapt

They say AI will kill the middle class. I say, the middle class was already dead. Trading 40 years for a pension that inflation eats alive isn’t security. It’s slow surrender. AI didn’t break the system. It just turned the lights on.

@aakashgupta Most people are writing instructions for a system that only reads headlines

Most people think Claude reads their whole skill file every time. It doesn't. Claude scans only the name and description of every installed skill. The full instructions don't load until Claude decides the skill is relevant. That decision happens off the description alone. Your description IS the routing logic. A 37-character description ("Suggest recipes from what's in fridge") never routes. The skill stays invisible. Same pattern inside the body. Claude reads the top of the file carefully. Adherence drops past line 300. Critical rules at line 700 may as well not exist. Most skills fail before line 1.

@anishmoonka AI competition is no longer just about building models, it’s about controlling the flow of compute and people

A Chinese lab spent about $294,000 building an AI that wiped almost $600 billion off the chipmaker Nvidia in a single day, the worst one-day loss any company has ever taken on Wall Street. The people who built it now have to get the government’s permission to leave the country. For the last three years, America has been trying to keep China from catching up in AI by choking off its supply of the best computer chips. The logic was simple. The strongest AI needs the strongest chips, and nearly all of them are made by one American company, Nvidia. So in 2022 the US banned the sale of its top chips to China and waited for China to fall behind. In January 2025 a little-known Chinese startup called DeepSeek put out an AI roughly as good as the best American ones, built on older and slower chips for a tiny fraction of the price. Wall Street took one look and panicked. If you do not need the most expensive chips to build great AI, then maybe Nvidia is not worth quite so much. Its stock shed almost $600 billion of value in a few hours. Now China is running the same playbook, only this time it is guarding people instead of machines. Bloomberg reported today that the government has begun restricting overseas travel for the top AI researchers at private firms like Alibaba and DeepSeek, who now need official sign-off before going abroad. Some have been told to hand their passports to their employer. At DeepSeek, the passports go to the hedge fund that owns the company. The official reason is that these researchers carry secrets the country does not want walking out the door. And they are worth guarding. Over the past year American tech companies have been locked in a bidding war over AI researchers. Meta offered some of them pay deals worth more than $100 million, the kind of money usually reserved for franchise athletes. DeepSeek’s entire AI cost less to build than what these companies now pay a single person. When one human being’s knowledge is suddenly worth that much, a country has every reason to keep them from boarding a plane. So the world’s two superpowers have each locked down one half of what it takes to build powerful AI, one hoarding the machines and the other the people who run them. A top AI researcher’s passport is now guarded as tightly as the advanced chips America refuses to sell.

@itsolelehmann True, but physical AI also has brutal constraints, safety, up time, maintenance, and integration cycles slow adoption way more than robotics talent alone suggests

Germany is a sleeping giant of physical AI everyone's been writing Germany off in the AI race because there's no German OpenAI and no big data center story. but theres actually two AI races happening: the first is software. chatbots, LLMs, data centers. US/China are winning that, not even close. the second one is physical. robots that pick up boxes, weld cars, carry groceries, stack pallets. and on this one Germany is one of the top contenders in the world this stat might convince you (it convinced me): Germany is 3rd in the world for robots per factory workers (449 robots per 10,000 human workers). only South Korea (1,220) and Singapore (818) are ahead. Japan is behind at 446. the US is all the way back at 307. so Germany already runs more of its economy on robots than almost anywhere else on earth. and the German companies building this next wave of physical AI are some global heavyweights. a few worth knowing... > Neura Robotics in Metzingen is building humanoid robots and raising €1B from Tether at a €4B valuation (this was March 2026). Volvo already in from an earlier round. > Sereact in Stuttgart raised $110M in April 2026 to build the software brain that lets robots see and grab things. already runs 1 billion+ real-world picks for BMW, Mercedes, and Daimler Truck. > Agile Robots in Munich was the worlds first robotics unicorn. revenue doubling yearly, around €200M now, heading for €1B. >RobCo in Munich raised $100M in early 2026 at a ~$500M valuation. their robots learn new tasks by watching a worker do it once instead of getting programmed line by line. already pushing into the US and aimed at the small and mid-size factories that make up most of german industry. > Fraunhofer (Germany's network of 76 applied research labs) built the evoBOT in the video below. self-balancing, two arms, carries 100kg of cargo, being tested at Munich Airport right now. but why is Germany specifically well positioned for physical AI though? three things stack on top of each other. first, the factories. Germany has thousands of family-owned precision manufacturing shops that have been logging sensor data for decades. that data is basically the training fuel for physical AI and almost nobody else has it at this depth. second, the customers are already there in-country. VW, BMW, Mercedes, Porsche, Bosch, Siemens. a robotics startup in Stuttgart can ship its first commercial deployment to a brand everyone recognizes in year one. that's why Sereact's customer list reads like a german car show lol. third, the engineer pipeline. Fraunhofer spins out companies like Agile Robots straight from its labs. KUKA built the first 6-axis electromechanical robot arm back in 1973. they've been doing this for 50 years. so the chatbot race is mostly settled and Germany lost spectacularly but the robot race is still early innings. and i think Germany's well positioned

@rohanpaul_ai this is what happens when demand runs ahead of structure, scarcity turns into a product, and then a bunch of intermediaries show up in the gap

Forbes: Ordinary investors are now being sold access to private OpenAI, Anthropic and SpaceX shares before the companies go public, through a private-market chain where access can be real, expensive, delayed, or fake. The real market is messier because private companies tightly control who owns their shares, so buyers often do not receive direct stock but exposure through brokers, funds, or SPVs, which are pooled vehicles that hold shares on investors’ behalf. Each layer can add fees, delays, resale limits, and legal uncertainty, so the investor may be buying a claim on access rather than clean ownership of the company. The danger is that scarcity creates a perfect setting for middlemen to sell stories, promises, and paper structures that may not map cleanly to actual shares. This market is less like buying a stock and more like buying a seat in a crowded waiting room where the door may open, stay locked, or turn out to be painted on the wall. Both OpenAI and Anthropic have warned that they do not approve sales of stock to SPVs. In May the company specifically called out a number of “unauthorized” secondary markets, saying that “Any sale or transfer of Anthropic stock, or any interest in Anthropic stock, offered by these firms is void…” --- forbes. com/sites/phoebeliu/2026/05/26/inside-the-murky-market-selling-pre-ipo-spacex-and-openai-shares/

@AnthonyJohns @kimmonismus True, but even if it’s Apple servers, relying on Google models still means Apple doesn’t fully own the stack. that’s the part that actually matters long term

@Redflagsignals @kimmonismus These are custom Google models Apple will run on Apple’s own servers

Apple isn’t just adding Gemini to Siri. It is reportedly using a custom 1.2T-parameter Google model as the brain behind parts of the next Siri overhaul (Reuters). That is no small size. The question, therefore, is how Apple’s Gemini will perform and how fast it runs. In particular, simple queries are expected to run locally. Gemini 3.5 Flash is estimated to have around 300 billion parameters; Apple’s model will thus be significantly larger. However, the question also remains whether size actually pays off in this context. Apple’s model must deliver answers to everyday queries quickly-and be fast enough while doing so. Anyways, next months will be exciting in so many ways: -WWDC / Apple Intelligence with Gemini - GPT-5.6 - Sonnet4.8/Opus 4.8 (probably) -Gemini 3.5 pro (confirmed) Super hyped!

@kimmonismus Yeah, if it’s still Apple controlled infra, the privacy story doesn’t really change much, it’s more about where inference runs than who built the base model

@Redflagsignals That’s true. However Afaik they run them on their own apple sever

"AI will give you back two hours a day" okay but two hours to do what exactly more work, that's what, the time doesn't go to you it goes straight back into the pile and the pile grows to meet it every productivity tool in history has promised time back tbh none of them have ever actually given it to you they've just changed what you spend it on

productivity culture has a new move and it goes like this: step 1: tell you you're overwhelmed because you're not using the right tools step 2: sell you the tools step 3: when the tools don't fix it, tell you you're not using them right step 4: sell you the course the anxiety was never the problem they were trying to solve it's the product.

replacing the worst CEO with the best sounds simple in an Oxford lecture, what's undocumented is the board access, the institutional relationships, and the balance sheet required to force that change, the opportunity was not simple, it was simple for someone with Ackman's specific position in the market

Bill Ackman turned an $8 billion railroad into $25 billion in 16 months, he shorted a $7 billion bond insurer at $73 this is him explaining how at Oxford "no one believed me, so I said: I hereby commit to give away 100% of any personal profit. that day was the high for the stock, it went from $73 to $3" "the opportunity was simple: replace the worst CEO in the railroad industry with the best, stock went from $46 to $151 in 16 months" test your strategy 60 times before risking a dollar, most people won't, the post below breaks down what that looks like in practice ↓

"free tier" is doing a lot of work in this ranking. Gemini 2.5 Flash free and Gemini 2.5 Flash paid are not the same product, Claude Sonnet free tier has rate limits that make it functionally different from the version people are actually comparing, the list needs a footnote longer than the list itself

The top 5 free ai models according to ChatGPT: 1. Gemini 2.5 Flash 2. DeepSeek V3.2 3. Qwen3 235B 4. Claude Sonnet (free tier) 5. Llama 4 Maverick Do you agree with this list?