Marjan Milosavljević

I’m building @Imitation_AI. A visual prompt engineering tool where you don’t chat — you build AI workflows. Think in graphs. Not messages.

@iruletheworldmo SPUDucinations everywhere…

GPT-SPUD-IMAGES V2 is live GPT6 is coming next week asi is here we are eating well brothers

@ivanfioravanti LLMs are just memory Tetris. Even an M5 Ultra loses eventually.

M5 Max freezed 😱 pay attention to memory usage while playing with LLMs.

@AnthropicAI This is what the leaked sentiment tracking was actually for. Not surveillance. Diagnostics. Emotion research, behavioral diffs, user signal analysis. Same direction.

New Anthropic Fellows Research: a new method for surfacing behavioral differences between AI models. We apply the “diff” principle from software development to compare open-weight AI models and identify features unique to each. Read more: anthropic.com/research/diff-…

upgraded from next 14 to turbo 16.2.1 with some packages somehow everything still works dependency graph looks like modern art thinking about a baroque antiqua tattoo: “npm run dev” definitely the most typed three words on my keyboard…

@pmddomingos Companies without their own frontier models won’t reach superintelligence. This reads more like a shift toward productization than a genuine understanding of what superintelligence actually entails….

And another one down: Microsoft has given up on superintelligence.

@iruletheworldmo SLA: sometimes works

thanks for paying $200 per day sorry our models are unusable at all times it’s all your fault pleasure ignore codex being better and their daily resets and continue to spend money us again. it’s your fault dear user.

@Akasheth_ bro deployed vibes, not architecture…

how do you know that I'm a vibe coder his codebase:

@TheHackersNews Exactly. No sandboxing means no control over blast radius….

@TheHackersNews Not specific to one tool. Many Claude-like and OpenClaw setups run code with more holes than Swiss cheese….

⚠️ WARNING - Attackers are weaponizing the Claude Code leak. Fake GitHub repos now deploy Vidar Stealer and GhostSocks, using trojanized builds that look legitimate. 🔗 Read → #fake-claude-code-repos-deploy-vidar-stealer-and-ghostsocks" target="_blank" rel="nofollow noopener">thehackernews.com/2026/04/claude…

@iruletheworldmo Sounds impressive. Let’s see what survives contact with reality….

i’m slightly surprised by the lack of excitement around spud (GPT 6) this is a brand new pre train. this is an order of magnitude improvement model. if i had to compare it. its closest to the jump from o1 to 5.4 xhigh. mythos is exciting but they’ll never be able to serve it, they’ve made the gpt 4.5 error. with the level of compaction and memory tricks you’ll have infinite context. the personality is much better than opus 4.6 it’ll all be served in an app simple to claude, with a a version of cowork and codex bundled into one, memory shared effortlessly between the modes. i promise you, you are not hyped enough. gemini book smarts, opus personality, agency we’ve never seen. the new agent mode is…something else. sama and gdb aren’t claiming to have agi and going on the media offensive pre ipo as part of some hilarious joke that will back fire. around 3 months ago it was clear spud had made leaps they didn’t anticipate. and only now can they signal the excitement. it is bad news for knowledge work, i have to be honest. but. it’s beautiful.

Microsoft is shipping model output comparisons and calling it innovation. The Git community turns leaked Claude Code into chat with GPT. Like taking a Ferrari and replacing the engine with a Fiat Panda. Meanwhile: Google drops Gemma 4, a strong open model family. And we keep shipping. 🤘 Editable output nodes + Gemma 4 + GLM Turbo, now live on the canvas… imitation.works

@iruletheworldmo …2030

G P T 6 I S C O M I N G

@iruletheworldmo Most “memory” today is just persistence, not understanding. What matters is how well it retrieves and uses context….

anyone have a sense on who’s ahead on memory. whether that’s in app, compaction in cc or codex. or someone mentioned ‘honcho’ to me. feels like we’re getting to a point where memory adds value so curious if people have had a good experience claude feels great for me personally but people love codex compaction.

@iruletheworldmo We’re ready. Just waiting for the announcement….

ready yourselves.

@elpege_ Because it shows how far on-device AI performance has already progressed on Apple hardware. And how far behind others still are….

@marjan_milo @mweinbach why?

Gemma 4 E2B on iPhone 17 Pro Max in AI Edge Gallery!

Gemma 4 just dropped — already on the canvas. Google's strongest open model. 256K context. Apache 2.0. 18 models. 10 providers. One workflow. imitation.works

@ai_for_success REAL?

@ai_for_success Speed 2X 😂

You can run Google new Gemma 4 on mobile easily. I am using Gemma 4 version E2B on my Pixel 10 Pro. Here is all you need to do: - Go to the App Store and install Google AI Edge Gallery. If you already have it, just update it. - From there, you can install the model directly and start using it. Still exploring what this small but powerful model can do :)

@mweinbach This is a hardware story….

@mweinbach Showing this to Google Pixel users should count as emotional damage….