Ni Biashara

Friday, half past seven. The grill is coughing white smoke, fat hitting coal with that hiss that sounds like applause. Somebody's speaker is losing a fight with the crowd. The air is salt, char, coriander, and the specific sweetness of onions that have given up. A waiter walks past with a wooden board balanced on three fingers like it owes him nothing. This is where I do my clearest thinking about artificial intelligence, which tells you something about both me and artificial intelligence. I have watched a few hype cycles wander in here, order big, and leave without paying. So let me give you the private-AI-versus-cloud-AI argument the way it actually lands at a table with a cooling pile of meat and one shared plate of kachumbari. Three scenes. One thesis. No sci-fi. Scene one: the man who trusts the counter At the counter, a regular hands over his order and his money and never once thinks about the cold room in the back. Why would he? The meat comes out warm, seasoned, correct. Th...

Nobody puts a fish auctioneer and a language model in the same sentence. One smells of salt and diesel. The other lives in a warehouse in Virginia that hums like a fridge the size of a stadium. But watch both at work and the resemblance is rude. The auctioneer stands over a crate of tilapia at 5 a.m. He looks, he guesses, he talks fast, he sets a price, he moves the crowd. He is a prediction machine wearing gumboots. So is the model. Both take messy input and spit out a confident answer before you finish blinking. Both are usually right and occasionally spectacularly wrong. Neither will apologize. So let us go to the market at dawn and follow the money. Not the hype. The money. Because that is where you can actually see what AI is doing to work across the continent and across the diaspora. The dawn shift, before the tools arrived A fish market at first light is a machine made of people. The boats come in. The porters run. The cold-room man guards the ice like it is gold, because at the...

The newest model is almost never the one you should build on, and I will die on this small hill. Launch day is a party. Your product is a marriage. The two rarely want the same thing. Let me explain the way I explain everything worth knowing — slowly, with food on the table, at the kiosk near the stage where the owner keeps a radio and a very strong opinion. Dusk is the honest hour there. The heat has let go, the lamp comes on, and you can finally see which bottles were dusty all along. Scene one: the shiny bottle at the front A young fellow comes to the kiosk asking for the new soda everyone is posting about. Bright label, big promises, arrived this morning. The owner sells it to him, of course. But then the owner leans over and says the thing that stuck with me: "Nzuri sana. But I don't stock my fridge with it yet. I watch who comes back." That is the whole discipline of new AI models in one sentence. A fresh release is a claim, not a receipt. The benchmark chart is som...

The power token ran out at exactly the wrong second. A boy had just handed over his coins at the kiosk, the owner was mid-tap on her phone, and then the little prepaid meter on the wall beeped its sad beep and the fridge went quiet. The sodas started their slow journey back to room temperature. The boy, unbothered, asked her to help him type a question into a chatbot on the cracked shop phone while they waited. She squinted at him. "You want me to ask the internet a question," she said, "but there is no electricity to keep my milk cold?" I have been thinking about that kiosk ever since, because she accidentally said the truest thing anybody has said about AI all year. The milk and the chatbot are drinking from the same cup. We just don't see it, because our half of the cup looks like a free text box. What actually happens when you ask AI a question Let me walk through it the boring way, because the boring way is where the truth lives. You type a question. It lea...

"Wewe, why are you still paying that man every night? Buy your own matatu and be free!" That sentence was shouted across a depot in the half-light of morning, one driver to another, while a third pretended not to listen and listened the hardest. It is also, if you squint, the entire argument about open-weight AI models. The whole thing. People are writing twelve-page essays about model licensing, and meanwhile an auntie at the bus stage explained it in one line before her tea got cold. Let me unpack the overheard wisdom, because it is sharper than most of what gets called analysis. First, what we are actually arguing about An open-weight model is one whose actual brain — the numbers it learned, the "weights" — is published for you to download, run, and keep. You can put it on your own machine, your own server, your own quietly humming box in the back room. A closed model lives somewhere else. You send your question over the wire to a company's data center, ...

I will start with the confession, because everyone in the trade has a version of it and most people hide theirs. For three months I rented the biggest GPU I could find, paid for it like rent on a house I never slept in, and shipped a model that was no smarter than one I could have run on a quiet machine in the corner of the office. I was paying for horsepower I was not using. The boda rider parked outside my building had already solved this problem years ago. I just had not bothered to ask him. Watch a rider who actually makes money, not the one chasing the loudest exhaust. He does not buy the most powerful bike on the road. He buys the one that sips fuel, starts in the rain, and survives the potholes on Jogoo Road without a new chain every fortnight. His whole business lives or dies on one number nobody romanticizes: shillings of fuel per trip . Top speed is a story he tells at the stage. Fuel cost is the truth he counts at night. Two ways to power an AI idea There are, roughly,...