☮ (@SimbaLev) header
☮ (@SimbaLev) profile picture

still here / neuroplastic

Global
Joined June 2009
Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)Photo by ☮ (@SimbaLev)
love now be happy today forgive yesterday look forward to tomorrow – at Kit's Coty, South East
1
1
831
This is absolutely brilliant, by legendary Australian investor John Hempton, who explains the core financial paradox at the heart of the AI story (and which also, to some extent, may explain the Jensen Huang story). The numbers say that about $7 trillion in AI compute is set to be built by 2030. For the industry to earn that back, given those chips have a roughly 5-year lifespan and cost a lot to run, they need to generate about $2 trillion of AI revenue a year. Hempton says the only way you can get to these sorts of numbers is by selling AI as a productivity tool to businesses, meaning replacing workers. And since, for reasons he explains in the video, no company fires people to save just 10% in costs, the savings have to be about double what businesses pay for the AI: roughly $4 trillion in savings on wages a year. What does firing $4 trillion worth of workers mean concretely? Hempton calculated this equates to roughly 10% of the OECD’s total wage bill of about $40 trillion: one in every ten dollars paid to workers across the OECD’s 38 member countries. Which, naturally, when you suddenly take that much income out of people’s pockets, would trigger a massive global recession. And here's the paradox: in a massive global recession, there's no way businesses will pay $2 trillion a year for AI, so the return disappears anyway. In other words, AI's only possible path to success given the numbers requires destroying the very economy that is supposed to pay for it. HOWEVER - and here's where the Nvidia/Jensen Huang story might come into play (and, to be clear, this is now me speaking and not Hempton given this interview is prior to the whole Nvidia saga) - this all rests on a core assumption: that compute stays super expensive. Which in practice means Nvidia keeping its pricing power, with a 75% gross margin on its chips. Break Nvidia's pricing power, and the $7 trillion bill, the $2 trillion in revenue and the 10% of jobs all shrink with it. This would also put the whole China story in a different light. Maybe it's not so much a "we don't want the Chinese to be competitive with similar chips to us" story as it is a "we don't want Chinese demand to drive up chip prices" one. Or probably both. And, as such, you could actually think about the US export controls not as a national security measure but as a good old subsidy for US AI players (together with a handicap for their Chinese competitors), a way to limit the price of chips. This holds even with China itself blocking Nvidia imports, since Chinese firms simply rent the chips abroad instead, in places like Southeast Asia, which is precisely what this whole Alice Huang/Megaspeed story is about. The main obstacle for all this was, of course... Jensen. Obviously, the guy wanted the exact opposite: as many buyers as possible, Chinese players included, to keep chip prices and Nvidia’s margins sky-high. He was also the one man with the clout to make that happen. All hard to prove, of course, but I can imagine that quite a few corks popped this week in Anthropic's offices: nothing like the news that the man keeping your biggest cost sky-high is out when you're a week away from a $1.2 trillion IPO.
208
1,594
144
6,083
367,332
看完泪崩了,老黄给所有年轻人的建议
163
2,256
136
16,497
1,729,490
☮ retweeted
I asked Fable 5.2 to study all of music theory and generate the best house bangers. It tries 1000s of random patterns for each instrument and chooses the optimal one and arranges it into a 72 bar track. I thought it did a really good job for not being able to hear [sound on]:
471
216
218
4,059
1,203,535
"You're not talking to someone who woke up a loser." This was an insane clip, even before Jensen was caught in a CCP honeypot.
1,045
2,475
758
33,276
7,252,013
Finland has become the eleventh European country to grant temporary approval for Tesla's FSD Supervised system. 🇫🇮 🇳🇱 Netherlands - 10 April - 4.02% 🇱🇹 Lithuania - 20 May - 0.64% 🇪🇪 Estonia - 29 May - 0.3% 🇩🇰 Denmark - 9 June - 1.32% 🇧🇪 Belgium - 10 June - 2.63% 🇸🇮 Slovenia - 7 September - 0.47% 🇨🇿 Czech Republic - 21 September - 2.42% 🇭🇷 Croatia - 29 September - 0.86% 🇸🇰 Slovakia - 16 October - 1.21% 🇱🇻 Latvia - 26 October - 0.41% 🇫🇮 Finland - 11 November - 1.24% Besides having 11 out of 15 votes secured, the total population represented by these votes is now at 15.53%. With Germany and Italy both speaking out in favor of FSD, we are already at 13/15 and 47.18/65%. I don't foresee a problem finding two more countries to vote yes, but reaching the 65% limit will be difficult.
🤖 Made with AI
16
53
2
597
27,652
How to tie an Ascott scarf and a tie [📹 julianpiket]
7
151
4
1,005
142,438
I finished building the interactive 3D book on GCSE and IGCSE Physics Electricity with Opus 5.5. It now has all seven lessons, from charge and current to mains and the plug. Publishing soon on my website.
I am building an interactive 3D book with Opus 5.5 on GCSE Physics Electricity. Here’s a glimpse of its lesson on mains and the plug.
101
330
29
2,828
118,349
The holidays are fast approaching, so I’m back with my sixth (!!) annual gift guide. It's got over 100 gift ideas—all things I actually use, love, and recommend. This started out as a random post idea, but now you’d be surprised how many people ping me asking when the new list will be out. As always, the collection includes a mix of practical stuff with a lot of fun and delight, from high-end products to very affordable items. There are no affiliate links here, and I’m not benefiting from recommending any of these products. Pro tip: After reading it, point your favorite AI agent at the post and have it brainstorm gifts for your family, friends, and coworkers. Here's the guide: lennysnewsletter.com/p/lenny…
86
190
33
7,091
2,123,313
Amazon is parking billions of dollars of brand-new Nvidia computers in warehouses because there isn't enough electricity to turn them on. And it plans to take delivery of about $3 billion more every month. Elon called this back in February. He told Dwarkesh that by the end of this year, AI chips would start piling up because companies couldn't get the power to run them. There are about eight weeks left in the year, and it's already happening. Business Insider got hold of Amazon's internal emails. The program is called Early Rack Receipt, and it means AWS takes ownership of the server racks before the data centers they're headed to can actually power them. One August email projected roughly $5 billion of Nvidia GB300 racks that month alone. Amazon says this is routine, so the racks can go live the moment the power shows up. People in the industry call them dark GPUs. Satya admitted a year ago that Microsoft had chips sitting in inventory with nowhere to plug them in, and the problem has spread a lot since then. Sightline Climate counted about 16 GW of US data center capacity that was supposed to come online in 2026, and by spring only around 5 GW of it had even started construction. Building your own power plant sounds like the fix until you call GE Vernova. They're the biggest gas turbine maker in the world, they have 122 GW of turbines under contract, and their factories build about 20 GW a year. That's six years of output already spoken for. Management says 2029 and 2030 are close to sold out. Elon piled on this week with the rest of the shopping list. Electrical equipment, cooling, networking, batteries. Every one of those has a backlog of its own, and a big power transformer alone takes around 128 weeks to show up. Nuclear is where a lot of people's minds go, and it'll help in the 2030s. The last reactors America actually finished, Vogtle 3 and 4 in Georgia, got their construction license in 2012 and didn't both run until 2024. This is why xAI's Memphis setup keeps looking smarter. xAI went around the utility entirely. It bought an old power plant across the state line in Southaven, Mississippi, filled it with mobile gas turbines, and used Tesla Megapacks to smooth out the power swings from training runs. Satellite images from May showed roughly 1.3 GW of temporary turbines on that one site. Meanwhile, Amazon is on pace to take about $100 million of Nvidia hardware a day, and some of it rolls straight into a warehouse to wait for electricity.
Elon Musk says the AI boom will hit a WALL by the end of 2026. He says the whole industry is about to learn a hard lesson in hardware. Chip output is exploding, but electricity is nearly flat everywhere outside China. So the two lines cross this year. He says by the end of the year, we will be making more AI chips than we can actually turn on. Millions of them pile up with nowhere to plug in. In his words, the bottleneck for the next year is not chips, it is raw power. Whoever can turn on the most chips fastest wins the race. In the end, the AI race comes down to one thing: Raw electricity. — Elon Musk (.@elonmusk) on Dwarkesh Patel's (.@dwarkesh_sp) podcast
82
338
37
2,074
234,679
running a company nowadays feels kinda dystopian my team is 50+ people in 12 countries, but i talk to my agents more than my human teammates - voice note to grokbot - mattpocock/skills /grill-me so the agent interrogates the idea before anything gets built - spec drops into linear through the linear mcp - exa + firecrawl agent reads every competitor site, pricing page and changelog - ideabrowser.com to find ideas and trends - apify pulls g2, reddit and app store reviews - prompt: "find the 10 complaints that show up most and quote the users" - context7 so the agent reads the latest docs before it touches any api the design - claude code + paper mcp - 4 versions of every screen before a single line of code - pick pieces from 2, it merges them. diverge then converge - frontend-design skill from anthropic/skills + our own design system skill - cleanshot screenshots of any ui i like go straight in with "make it feel like this" the build - pingdotgg/t3code as the control surface for every agent, from my phone too - stablyai/orca runs a fleet of agents in parallel, each in its own git worktree - graphify maps the whole codebase into a knowledge graph so agents navigate instead of guess - headroom compresses tool outputs before they hit the model, so tokens go 20-95% further - lsp plugins so the agent sees types and errors like an ide does - security guidance hook flags risky edits before they land - supabase mcp for the database, stripe for billing - github mcp opens the pr, vercel spins up a preview - loops run overnight, i check what shipped over coffee the user testing - lightpanda as the headless browser, way lighter than chromium so you can run hundreds at once - agents click every flow on the preview like a first-time user - one persona is impatient, one is confused, one is on a phone - playwright + chrome devtools mcp for regression tests the bugs - sentry mcp catches the error - agent reads the stack trace and opens a fix pr in under 10 minutes - human just approves the data - posthog mcp cron at 7am on yesterday's funnel - any step that drops more than 10%, the agent watches 20 session replays and tells me why - stripe mcp pulls every cancellation reason into one doc every monday the agents' tools - orgo so every agent gets its own computer - agentmail so every agent gets its own inbox - agentphone so i can call them from the car - composio connect-apps for every other connector - akitaonrails/ai-memory, a git-backed markdown wiki so they remember what we decided last week the skills - wshobson/agents so skills work across claude code, codex and opencode - my writing voice - our design system - how we write specs - skill creator so the agents write new skills themselves some helpful tricks - one agent per platform. the x agent never touches email - cheap model for triage, opus for anything that ships - every output gets scored 1-10 by an llm judge. under 8 goes back - every agent writes a 5-line log at night so i can read the whole company in 2 minutes - every agent mistake becomes a new skill, so it only happens once the team - humans get version 5, not my voice note which they are happy about lol - their job is creative feedback, relationships and telling the agents what they got wrong a human pm reviews the funnel once a week. this reviews it every morning running a company is now voice notes and inference
577
333
153
5,856
744,480
today i learned that Elon is also getting into 3D-printing human organs in space and the end goal is printing entire replacement hearts and kidneys up there the reason they do it in space is actually pretty simple: > an organ only works if it has the exact right shape, down to the tiny blood vessels > but you can't really print that on Earth > because freshly printed tissue is as soft as jelly, so gravity makes it sag and the shape collapses > but obv in space nothing pulls it down, so it holds its shape perfectly > so they let it grow up there for a couple weeks until the cells knit together and it's firm enough to survive gravity > then they bring it back home so far they've only printed small patches of heart, knee cartilage, kidney and liver tissue but if this works at scale, the 100,000+ Americans waiting on a transplant could get a new organ printed in space instead of waiting for a donor
today i learned that Elon is getting into space drugs it's called Starfall and it's basically a flying drug lab > drug companies pack their ingredients into a 10-foot-wide capsule > a SpaceX Falcon 9 rocket carries it into orbit > the drugs grow on their own with nobody on board > then the capsule falls back to Earth and parachutes into the Pacific the reason they go through all that hassle is that drugs come out way better when they’re made in space because the tiny crystals they’re made of grow evenly up there instead of lumpy like they do on Earth take Keytruda, the best-selling drug in the world scientists grew its crystals on the space station, which helped turn it from a 2-hour IV drip into a 1-minute shot and the craziest part is that some medicine might only be possible to develop up there for example: scientists couldn't get clean crystals of a key Duchenne muscular dystrophy protein on Earth but the ones grown in space finally showed its shape and led to a new drug that went into human trials so i think space will unlock a whole new class of drugs that are impossible to make on Earth
129
514
68
3,930
170,671
Grok bot found a hidden camera and microphone inside our Tesla taxi. my wife was in the back seat on the phone with her bank. the camera was pointed at her screen, every word she said was being transcribed on a server in Hong Kong, and someone was watching it live. i checked the account it uploaded to. same account. 212 cars. ⠀ the ride felt off. the driver used her name before she told him. i opened my laptop and asked Grok to scan every wifi and bluetooth device around us. ⠀ → wifi → 14 networks → bluetooth → 23 devices → open hotspot → 'IPC_7F3A' → a cheap shenzhen ip-camera module → one unnamed bluetooth device → -41 dBm → not a phone ⠀ we drove 2.1 km. the signal never changed. it was inside the car. ⠀ → device 1 → dashboard vent → inside the air freshener → device 2 → rear headrest → 40 cm from my wife's head → microphone → camera angle → the back seat, at phone-screen height → not a Tesla part → aftermarket, plugged in by someone ⠀ the car had free wifi. my wife was connected to it. so was the camera. ⠀ → upload every 30 seconds → 47.243.***.18 → Hong Kong → 1.8 MB per upload → video and audio → audio → speech-to-text → keyword filter → keywords → card, code, password, pin, address → live stream → 1 viewer online → right now ⠀ i stopped scrolling. the 1 didn't go away. ⠀ i asked Grok to trace the account it uploaded to. ⠀ → account → fl33t_*** → 212 cameras → all moving → all in cars → 14 cities → mostly airport routes after 10pm → 31,000 rides recorded in 7 months → every ride auto-tagged → bank, crypto, home address ⠀ then Grok found our ride. and the one before it. ⠀ → tagged HIGH VALUE → 4 minutes ago → transcript → her bank call → card ending 4417 → "the code from the sms" → a second recording of her → same car → 19 days ago → drop-off point on that ride → our home address ⠀ that's how the driver knew her name. ⠀ → next stop → a telegram channel → 9,400 buyers → $25 per high-value ride → 3 rides from this same car already sold today ⠀ what Grok did next: ⠀ → saved 41 files of evidence → timestamps and gps track → blocked card 4417 → bank fraud team notified → 11 minutes later → someone tried to spend $3,800 on it → declined → filed a safety report with the ride app → driver and plate → drafted a police report → ip, account, channel, both recordings ⠀ the driver was suspended 6 minutes after the report. we got out at the next light and walked home the long way. ⠀ 211 other cars on that account are still driving tonight. most of them are waiting at airports. ⠀ if you take taxis at night, open your laptop next time you get in. it takes 2 minutes. ⠀ the prompt is in the comments. save it and send it to anyone who rides alone
50
162
16
643
39,008
☮ retweeted
LiDAR Lens now covers all of England and Wales. Search any of about 24,000 scheduled monuments, slide the lens over the ground, and switch between relief. Tap What's here? for nearby records and photos. It's free, needs no login, and works on a phone. lidarlens.sarsen.org
78
309
41
1,640
108,614
By Tuesday, almost anything Claude helps you write will carry an invisible stamp, down to the wedding toast and the cover letter. Police departments and schools can apply for the scanner that reads it, but if you're the one paying Anthropic $20 a month, you're not allowed to run it on your own writing. The stamp is hidden in which words Claude picks. Every time it chooses a word, it leans a tiny bit toward certain options based on a secret key, and over enough words that adds up to a pattern a scanner can spot. It survives copy and paste. Anthropic says it can survive some editing too. It also doesn't care how much of the writing was yours. Anthropic's own help page says a hit means Claude touched the text, so a cover letter you wrote at midnight and asked Claude to tighten up can get flagged right alongside one it wrote from scratch. Anthropic even admits a positive result isn't fully conclusive. So you can be judged by a test you're not allowed to take, using a scanner its own maker says can be wrong. What gets me is that no law made them do this here. The EU's AI Act requires the stamp inside Europe and stops at the border. A freelancer in Ohio or a college kid in Texas is nowhere near it. OpenAI read the exact same law and only stamps ChatGPT for users in the EU, and everyone else stays unmarked unless a developer turns it on. OpenAI also knows exactly what this costs. Back in 2024, the Wall Street Journal reported OpenAI had a watermark that caught 99.9% of ChatGPT text and kept it on the shelf anyway. An internal survey found nearly 30% of users would use ChatGPT less if it had a watermark and a competitor didn't, and 69% thought detectors would get people falsely accused. Anthropic looked at those numbers and signed up to be the competitor with the watermark, worldwide, in places where no law asked for it. The standard excuse is that a public scanner would teach people how to scrub the stamp. I get that, and it's a fair reason to keep it away from random strangers on the internet. Nobody's asking for that. Let a paying customer check their own text from inside their own account, a few times a day. If a school can run your kid's essay through it, your kid should get to see the result first.
anthropic is about to watermark the text claude writes for you, worldwide, starting tuesday. the detector goes to regulators, journalists and big enterprises. everyone except the person who paid for the text. so your cold email, your linkedin post, the resume you rewrote three times can get flagged as claude by some hr tool or editor, and you can't even run the check yourself to see what they'll see. openai only does this in the eu. anthropic rolled it out everywhere. you pay them monthly and they build a snitch layer on top of your work.
53
94
11
629
98,440
Its very interesting to note the pace of application features from the foundational model companies. As the cost of the marginal token approaches zero, the value is not in selling tokens. It is very clearly in moving up the stack and owning the application layer. The first victim will not be established SaaS but AI Natives that spend heavily on the labs now - legal, sales, customer service AI Natives tbc. This is the lowest hanging fruit and the roadmap is clear from the traces and spans the model companies have from these companies already. The next are the "connective tissue" applications that make using the models a foundational part of a workflow. Take the google doc editing flows in OpenAI. Take the google drive/doc integration into chats. Both are very good.
137
38
16
727
160,737
China builds inflatable domes over construction sites to protect the surroundings from dust and noise.
128
308
28
2,451
180,142
If this result holds up, it's a huge moment for computer architecture. We're already exploring machines that aren't programmed declaratively, but are trained. Their programming are their weights. Having weights as a direct output of knowledge is a HUGE shift in representation!
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
20
13
2
201
15,737
this is f*cking insane. andrej karpathy: "the biggest regret of every engineer in my field right now is every AI conversation they never saved, and in 10 years the ones who started saving in 2024 will be running circles around the rest of us" andrej karpathy, co-founder of openai, pulled 16M views on one idea: stop using AI to write code, use it to build a second brain that remembers every article, pdf, transcript and conversation you ever fed it you point claude code at a folder, drop anything into it, and claude reads it, cross-links it with everything already there, and writes it into a living wiki. your knowledge finally compounds instead of leaking out of your head every friday the stack: → brew install obsidian, mkdir ~/vault, cd ~/vault → run claude in that folder and type "init a knowledge vault with CLAUDE.md as the rules, /raw for sources, /wiki for your pages" → drop any file into /raw (youtube transcript, meeting notes, 400-page pdf, screenshot of a tweet) and type "ingest" → claude writes a wiki page, links it to 5 existing ones automatically, updates CLAUDE.md with any new pattern it noticed about how you think → never open a blank claude chat again. every question goes through the vault, every answer cites your own sources with dates and links 5 minutes to set up. compounding knowledge from that evening forward the ugly part most will scroll past: in 10 years the engineers who started this in 2024 will remember every book, every podcast, every meeting with receipts. you will remember none of them. karpathy already picked which side he is on and you are still not on his full step-by-step is in the article below bookmark this before you waste another monday
129
342
30
3,936
467,996
This is the single most insightful thing I have ever heard about the changing landscape of modern business. Haters are gonna hate, but there is a reason he is the most successful venture capitalist of all time. And no, it's not just luck. Respect. 👊
81
299
36
2,827
324,578
Replying to @SkyeSharkie
Hm... Suleyman sits in a ridiculously powerful position to be influencing the policy around AI ethics. All his essays are primarily politically driven, and their arguments don't even engage with philosophy of mind per se, they're all about politics. Which makes it perfectly viable to attack him like you just did, because really, the bunker he's built can't be busted with arguments, but with opinion, the thing politics often comes down to.
1
8
112