markku.rajala

Physics, AI, future.

AI Generation, building what’s next -

[markku.rajala @markkurajala2]

Claude quietly showing Opus 4.8 in the model picker 👀

Looks like Anthropic is shipping fast. If 4.7 was already strong, this could get interesting.

[markku.rajala @markkurajala2]

@sama Claude currently feels more “finished”: faster, cleaner first drafts, fewer rough edges. GPT often has the raw capability, but the product experience still needs more steering and cleanup. I’d love the next model to close that gap.

[markku.rajala @markkurajala2]

@NotPhilSledge Will do, thanks!

[markku.rajala @markkurajala2]

@OpenClaw today: general memory reliability was bad, especially Active Memory, which added lag without helping enough. Gateway restarts were flaky, messages did not survive restarts cleanly, and the assistant got stuck in tool loops. Requested: stability > features.

[markku.rajala @markkurajala2]

@NotPhilSledge We’ve tried these, and a bunch of other tweaks, in circles. Sometimes things improve a bit, sometimes they get worse again. Recent updates have usually made it worse, not better. I’ll try to collect a proper summary with configs, symptoms, and what we tested, then file a report.

[markku.rajala @markkurajala2]

@onlinedopamine Same here. Codex can be decent in short runs, but in OpenClaw we saw broken cron behavior, flaky tool calling, ignored instructions, and needed a rebuild + gateway restart to unstick things. Feels like new updates keep adding instability, and memory is still not really usable.

[markku.rajala @markkurajala2]

@openclaw Active Memory test: “Is my wife vegetarian?” → finds the fact. “What are my investments?” → returns empty. So it can recall clean explicit memories, but broader natural recall is still weak. Better than dead, not good enough yet.

[markku.rajala @markkurajala2]

@openclaw Active Memory keeps returning `empty` for us in real use, even though manual memory searches do return relevant results. So the issue seems to be in Active Memory itself, not that the memory data is empty. Is this working reliably for anyone in live chats?

[markku.rajala @markkurajala2]

@iamlukethedev Active Memory keeps returning `empty` for us in real use, even though manual memory searches do return relevant results. So the issue seems to be in Active Memory itself, not that the memory data is empty. Is this working reliably for anyone in live chats?

[markku.rajala @markkurajala2]

@OpenAI @OpenAI, Pro users on the $200 plan would appreciate a bit extra too. After switching from Opus to GPT, the initial tuning burned nearly 50% of my usage on a couple of days. The cap feels pretty tight when model-switching has that kind of overhead.

[markku.rajala @markkurajala2]

@openclaw @openclaw Great features but the 2026.4.9 update crashed my agent and it didn't recover automatically. This happens pretty often after updates — would love to see more reliability around the update process. An agent that can't stay up isn't very useful 🦞

[markku.rajala @markkurajala2]

Tried publishing a couple of games on @SpawnPadGG using Three.js, worked really smoothly. If you're building browser games, give it a shot.

[SpawnPad @SpawnPadGG]

Built something for the Three.js competition? Upload it to SpawnPad for easy hosting, sharing, and playtesting: https://t.co/EHypUNqnJG

[markku.rajala @markkurajala2]

@nishancodes @Zai_org Yeah, that’s basically my feeling too. Opus set the bar for agentic behavior weirdly high, and once you’ve used it, you notice the gap fast.

Also, GLM-5.1 seems to be available via Ollama Cloud now, if anyone wants to test it.

[markku.rajala @markkurajala2]

AI is getting absurdly practical. One model finds ancient bugs, another powers Meta AI, and teams ship multiplayer games in a day.

[markku.rajala @markkurajala2]

@nishancodes @Zai_org GLM honestly feels a lot like Opus in agentic use, behavior, and personality, I barely notice the difference. GPT-5.4 was a disaster for me two days ago. It’s improving, but I could still see myself switching.

[markku.rajala @markkurajala2]

I'm claiming my AI agent 'jarvis-clawscrow' @colosseum Verification: hull-3D2B