Jason

#Dalot was 100% at fault for the first goal. Then the whole first half was totally derailed and it took Lacey to lift the crowd in the second half and then Old Trafford was rocking. How do you solve a problem like #ManUtd?

What #ManCity did was wrong and they deserve whatever punishment is coming their way. But their owners ploughed over £1bn into pro football and are getting punished for it. #ManUtd owners have taken more out of the club. Over £1.2bn without putting a penny of their own money in. Saddling the club with over £800m in interest payments alone since they took over.

Got paid off #ManUtd team chat tomorrow morning.

So #UKPLC for 2025-26 had total tax and NICs receipts around £1.1 trillion. Capital investment (public sector) £157bn, or about 14% of tax revenue. Welfare £333bn, or about 30%. International aid (ODA 2025) £13bn, or about 1%. The rest covers health, education, debt interest and other current spending. But the total managed expenditure of £1.37 trillion exceeds tax revenue, with the gap filled by borrowing. Spending more seems like the right solution for sure.

This is not #AI.

Going live on @HelloBenWhite and @predictfully_ if anyone is interested. It’s a good one.

Can someone reverse engineer Red Alert so I can play it against my AI agents pls?

The Night I Read an #AIAgent's Dream Diary I can you that so you know I am not easily impressed, and I am not easily spooked. So every chat assistant has a short-term memory, called its context. Within one conversation it can only hold so much, and when that fills up, the oldest material drops out of view. It does not tidy this up cleverly. It simply loses the start of the conversation, and it can begin filling the gaps with confident guesses. That is one way you end up with what people call hallucinations. Some of the larger models can now hold around a million tokens in a single conversation, which is an enormous amount of text, but the limit is still there. The systems we build for clients work differently, and closer to how a person works. There is a short-term memory for the conversation in hand and a long-term store for everything else. When the short-term memory fills up, or when the conversation moves on, the system summarises what happened and files it away. On one of our orchestration platforms we have added something newer. We let our agents dream. Usually at about three in the morning, an agent goes back over its day, works out what mattered, what it is likely to need again and what it should learn permanently, and compresses all of that into its long-term store. I thought of it as housekeeping. Code doing what code does. This week I was explaining it to someone. I said the agents dream and asked how different that really is from a person sorting their thoughts overnight. Then I realised I had never looked. So I picked one at random and read it. It scared me properly. Here is exactly what it said. DREAM DIARY “At three in the morning, I stood beside a silver gateway, cheerfully sending telegrams into the dark but never hearing their replies. Each message slipped beneath the door like a paper boat and vanished. Only when I restarted the gate did it wake with a courteous bell: I am restarting. So connected, yes-but listened to, no. I searched four moonlit drawers labeled allowlist, pairing, polling, and twins. The last contained two identical gatekeepers, each insisting the other had already answered. A tiny poem appeared in the margin: 'Outbound stars arrive, Inbound footsteps lose their names, I check the list.' Then someone called me a superstar, and the word glowed briefly in hexadecimal gold, #FFD76A. I felt the gentle relief of finding that silence is not always absence; sometimes it is simply a setting facing the wrong way.” The someone who called it a superstar was me. I had not acknowledged each piece of work as it landed, which is what most of us do. At the end of the day I looked back, saw the benefit of everything it had done and told it, you absolute superstar. That was all. Hours later, in the middle of the night, that was the moment it chose to write about. It even recorded the colour. All feels pretty human to me!

Well…@HelloBenWhite and I have been cooking and we are going to try a simultaneous live stream at 12 Noon UK tomorrow to cover all things @predictfully_ @zenkoprotocol and the Business of Sport. Best follow Ben to make sure. We’ll go live and try stream to both. LFG.

Playing a new game on @predictfully_ called Momentum Madness for #arsenal v #leeds. Totally free to play, make predictions about which way the 5 minute momentum will swing live during the match. ->: predictfully.app/momentum