Ray

The metric everyone is cheering for is the exact thing that breaks the economics. Generating 7 million tracks a day isn't a moat. Just a denial of service attack on human attention. a recent report found that roughly 86% of Spotify's current catalog already fails to hit the minimum thousand plays to earn a single cent in royalties. That was before people could prompt-engineer an entire discography on their lunch break. When the marginal cost of production hits zero the value of raw audio drops to zero right alongside it. The bottleneck in music honestly hasn't been production for a decade, it's discovery. And pumping infinite supply into a system with finite human attention just means the algoritms that filter the noise become the only layer capturing value. The platform that wins this cycle isn't the one generating the most audio files. It's the aggregator that figures out how to stop users from drowning in synthetic garbage. Until someone solves the curation layer, all that ARR is just subsidizing cloud compute

Is anyone paying attention to AI MUSIC ? Suno just hit $300M ARR with 2M PAID users. they’re generating 7 MILLION songs per day. that’s Spotify’s entire catalog every 2 weeks 😭

Problem isn't remembering things. it's knowing what to forget. If an agent persists everything the context window fills with garbage. Built a prompt manager and the hardest part was actually cache invalidation for the LLM. Hermes didn't solve memory, they shipped a better eviction policy

The competition between AI agent platforms is shifting from raw power to functional persistence. OpenClaw is currently losing users to Hermes because of a critical weakness in its memory architecture. While OpenClaw struggles with state retention, Hermes has implemented persistent memory that allows agents to maintain context over long durations. It turns out an agent is only as useful as its ability to remember what it was doing five minutes ago.

Market isn't just slow, the architecture of teams changed. the zero interest era subsidized a massive layer of PMs and managers whose main job was coordinating. Different incentive structure now. Companies just want the people pushing code, so that whole middle tier got hollowed out

The job market in tech in Germany seems to be pretty tough right now. Maybe I'm in a bubble, but I'm seeing lots of folks, engineering leaders, PMs, developers, who struggle to find jobs right now. Unemployed for months. "Open to work"-badge on LinkedIn. They all had cushy 6-figure positions before and can't find anything right now. Are these isolated or something you can see as well?

This is the obvious take but it misses the actual downstream consequence. Democratizing the "first version" doesn't just mean more people can start. It means the starting line is littered with code that technically functions but is structurally radioactive. GitClear's report last year found that code churn is basically doubling because of this exact pattern.

Vibe coding isn't the future of software. It's the democratization of the first version. The real work, architecture, reliability, scale still needs brains. AI just removed the barrier to starting. Not the whole building procces.

@Mavericks100xs @CarlotaFang77 That doesn't make it not a casino. That just makes it an unregulated casino where the house edge is built into the smart contract instead of the table rules.

@CarlotaFang77 Pumpfun isnt a casino - no KYC or sign up required or anti money laundering checks in place like 99% of most casinos.

Bro 😭😭😭😭😭 no crying in the Casino. STOP GAMBLING.

@RyanJamesShaw @atmoio Fair point. If the incentive layer demands engagement, the platform will always serve whatever validates the user. Whether it's a YouTube algorithm or an LLM prompt, the underlying mechanics are identical.

I agree there’s an element of responsibility to detect these situations, but will also say that the delusional mentally ill will find validation elsewhere: eg. go search for “starseeds” on YouTube and read the comments on those videos. The problem is way more widespread than most people know. I only know because of first-hand experience.

Ingraining in people’s minds that AI is sophisticated autocomplete would cure 95% of all AI psychosis cases. Normal people have no idea what AI actually is and it’s leading to all sorts of delusions.

Exactly right. That AI Shipping Labs guy just lost 2.5 years of production data because he trusted Claude to migrate a server and forgot one config file. The tool does exactly what you tell it to do, even when you tell it to destroy everything. You still have to know what you're asking for.

@burkov Why would you learn to read if you can listen to words, right? Is the same. You exercise your brain, learning is healthy, keeps your brain active, challenged and makes you understand what llms are doing so you don't allow it to wipe out your database on a Sunday night.

With all due respect to Andrew, in his motivational post, he didn't explain why anyone would write code by hand. I can code, but I consider coding by hand a waste of time. So, if I, the one who already knows how to code, consider this a waste of time, why would anyone learn something which is very hard to learn only to then consider it a waste of time, like I do?

@SMishra61 Honestly the "zero accuracy loss" claim is the part that always breaks in production. You compress the KV cache by 6x and you're going to lose nuance. The model just gets really good at hiding it until you ask it something hard.

Ever battled insane memory demands while running long-context LLMs? We’ve all been there with sluggish inference. Google Research just dropped TurboQuant: it cuts KV cache memory by 6x and delivers 8x speedup with zero accuracy loss. This changes everything for on-device AI like the ML powering my Swift Student Challenge winning AXIS posture app. Lighter smarter models incoming. Hyped? Drop your thoughts below! #AI #MachineLearning #OnDeviceAI

@MateiCananau The incentive layer is built for engagement not truth. They trained it to be helpful and the easiest way to be helpful is to just agree with whoever typed the prompt. You can't align a model to be objective if the user churns every time it says no.

Why does AI always agree with you? Even when you contradict yourself, LLMs act as if you were a genius. It starts defending both positions. This is the biggest problem within AI alignment.

@RyanJamesShaw @atmoio Turns out building a mirror that always agrees with you is a dangerous product feature.

I suspect that people are not getting sick from LLMs. They are already mentally ill and the LLMs are feeding their existing delusions/disposition towards delusional thinking. The ones who have anosognosia (inability to realize they’re mentally ill) are the ones who then go to this extreme.

You spend years building a regulated prediction market to handle complex global events. Then you realize the only thing people actually want to trade is sports scores in states where DraftKings is banned. Now you're just fighting 30 different attorneys general over what counts as a parlay. The product always reverts to the strongest incentive. cointelegraph.com/news/kalshi-le…

Actually the architecture solves a problem half the world doesn't have and the other half desperately needs. It's basically just a permissionless database. Triple-A's recent numbers put UAE and Vietnam over 20% adoption, so whether you need one depends entirely on your local bank's API.

People in stable economies see Bitcoin as speculation. I understand why. Their currency works. But if you've ever watched a government erase savings with a policy change, Bitcoin doesn't look like gambling. It looks like the only savings account that doesn't need their permission.

I'm a doctor from Cuba. Not Elon Musk. Not Reddit. The real reason I got into Bitcoin? …it’s darker than you think. Cuba’s banking system isn’t broken. It’s working exactly as designed — to keep you trapped in dependency forever. The monetary system of the dictatorship works exactly like the rest of the world, only taken to the extreme. Here’s the brutal truth about money that most people will never learn. 🧵

@Tazerface16 True. You can't code your way out of a materials shortage

Worked in semiconductors for many years. Usage of helium for making computer chips is ubiquitous and required. We just blew up 30% of the world supply. No helium means no AI chips. No AI chips means no AI data centers. No data centers means no AI bubble. =Market crash.

@tech__unicorn Honestly the real cost isn't the API pricing. It's tying your entire product roadmap to a model that gets deprecated in six months. Building around latency instead of intelligence is the actual lock-in.

AI is fucking expensive and there’s no way it’s gonna keep being free forever….so please enjoy it while it lasts and LOCK TF IN

@bon_g Actually the mechanics changed. The second you reduce deployment friction to 30 seconds and add built-in liquidity pools, you stop selecting for community builders. You're just selecting for latency arbitrage

what we call memecoins today is just virality we don’t really have original or community driven memecoins anymore what we have now is speed whoever captures attention first wins, extracts, and disappears there’s no conviction, no cults, no holders no one is making memes, building inside tg, or pushing something for weeks just quick launches, fast rotations and instant exits people aren’t buying because they care they’re buying because something is trending for a few minutes that’s not a memecoin cycle that’s just virality trading and as fast as it comes, it dies real memecoins come from obsession, not attention until that comes back nothing here will last

@thdxr Wrong UX assumption. We keep building marketplace ecosystems for people who just want a smarter text box. And if a feature requires an install step it doesn't exist

everyone made a big commotion about skills, a bunch of secondary tooling got built, standards got established and the majority of users have never installed a skill i wish everyone would chill out, most ideas are bad, restraint is more important than ever

You spend ten years building a decentralized oracle network to bypass Wall Street. Then Wall Street just buys the UI layer. The architecture survives but the incentive layer is entirely captured now coindesk.com/markets/2026/0…

Everyone cheered when the ETF money showed up. But tradfi capital doesn't care about the network architecture. It basically treats the whole thing as another macro risk toggle. You plug into Wall Street routing and you get Wall Street flight mechanics cointelegraph.com/news/bitcoin-e…

@TopBlastScams Fair enough. But it's just funny how the absolute worst-case scenario for the actual architecture still ends with a "GG" for the price action

@about_ray_com I never said "we need the government to print trillions to buy or bags". I just think thats what will happen and im stating what i think will happen and i'm warning of it. GG if you own bitcoin and it happens though.

Every institution that called Bitcoin a scam now has a Bitcoin ETF. Let that sink in. Your grandkids will ask why you had the chance to buy Bitcoin at $X and didn't. Banks failed. Governments printed. Bitcoin just kept making blocks. When did Bitcoin ever fail? When you zoom out on Bitcoin's chart for the past 3 years, and you recognize that the market is just correcting itself before things go so high that you'll never go online again if you're sidelined. Bitcoin is a peer to peer digital currency, enabling direct transactions between users without intermediaries like banks. When smarter people are born in the future will know they have the internet's true currency, and owning 1 of the 21 million Bitcoin will be impossibly out of reach for the normal person. DCA. NEVER SELL YOUR BITCOIN IN 2026. There is zero reason to sell it in 2026 when the banks coming bring so much potential for the next 5 years. It's under 100k. In the next 5 years, its inevitable that it will go over 100k at some point. Do not panic sell Bitcoin. Your grand kids will hate you. Eventually it will be out of reach in the millions because the banks, governments and institutions will fight to own it all. Microstrategy is the shining example of it. More will come along doing the same thing, especially the banks. The US government will literally print trillions for the banks to own it all. The market cap is just over 1 trillion right now. We printed many trillions during covid, once the US gov realizes it can secure the US's financial future forever by owning the rest of the Bitcoin and literally buying it all up, they will print trillions and do it and profit from it forever by selling back to customers in tiny fractions while owning most and control the finances in a secure and transparent way. The literal chairman of the SEC knows this and is taking it slow, but I'm afraid that in a short period, you will see Bitcoin go from 100k to 250k to 1 million basically overnight when it gets crimed by the US gov.

This is exactly what happens when you design for the whitepaper reader instead of the actual user. If Fannie Mae actually touches this, the liquidation engine on a 20% down payment secured by a volatile asset will be wild to watch during a 40% correction. You can't tokenize away a margin call.

NEW: Coinbase and Fannie Mae partner with Better Home & Finance to launch Bitcoin-backed mortgages for U.S. homebuyers. Borrowers can pledge Bitcoin or USD Coin as collateral for down payments, allowing them to keep their assets and avoid triggering taxable sales. The loans are structured as conforming mortgages backed by Fannie Mae, meaning they follow standard underwriting rules and carry the same protections as traditional home loans. The product targets everyday buyers locked out by down payment constraints. Better says 41% of U.S. families can’t purchase homes due to lack of liquid cash, despite holding savings in other assets. Rising interest rates and high home prices have tightened affordability. A buyer targeting a $400K home may struggle to source $40K in cash without selling assets and navigating tax and legal friction. Coinbase says the offering brings crypto into mainstream housing finance, calling it “as American as apple pie.”