Travel planner & Bike guide | 🇮🇹 🇳🇱
Netherlands & Italy
Joined October 2010
- Tweets1.1K
- Following324
- Followers297
- Likes1.2K
My lesson for today: ''Sleeping turns knowledge into wisdoms''
Not a 1 on 1 quote, but it's the concept.
Comes from:
youtube.com/watch?v=-OBCwiPP…
Things really do go fast!
Just set all my online research cross and prompts to run on Qwen3.6 Plus Preview (free tier) on OpenRouter.
No personal data, no MEMORY.md, no strategy, nothing sensitive because Alibaba collects prompts on the free tier.
Check it out here: openrouter.ai/qwen/qwen3.6-p…
Just cut my agent's cloud costs significantly without sacrificing quality.
Thanks to @witcheer and @AlexFinn, I starting building from your posts!
The journey:
Started on Mac mini M4 (24GB…) running Claude Haiku/Sonnet for all background tasks ~60 API calls/day.
Tried Qwen3.5-35B-A3B via Ollama first. 23GB model. OOM. Killed.
Tried Qwen2.5-14B. Fit fine, but reasoning quality too weak for my workflow.
Then TurboQuant dropped @GoogleResearch. KV cache compression, 4.6× smaller memory footprint!
My new stack:
Qwen3.5-27B-IQ3_XXS (10.7GB) via llama.cpp TurboQuant fork → 13.6 tok/s on M4 Pro → zero API cost.
Validation before shipping:
→ 15 tool-use scenarios: 12/12 pass
→ Shadow test against Haiku on live data: output indistinguishable
→ Did have 3 timeout failures on error recovery — but a lot of crons can run overnight with a higher timeout, not a quality issue
My full model stack today:
🟢 Local (free)
→ Qwen3.5-27B IQ3_XXS — health monitoring, training load, environmental logging. (Based on multimodal biomarkers, keen on getting that data vectorized!)
→ MedGemma 4B — offline domain-specific model
🔵 Anthropic
→ Claude Sonnet — interactive sessions, memory synthesis
→ Claude Haiku — research briefings, clinical alerts, complex reasoning chains
🟡 Google
→ Gemini 2.5 Pro — fallback when Anthropic unavailable
⚫ DeepSeek
→ DeepSeek V3.2 — cost-efficient tasks when applicable
The "local = cheap but dumb" assumption is breaking down fast fast faster!
Probably it is already outdated, goes so fast! Curious to get it smarter and cheaper everyday 🥳
Just cut my agent's cloud costs significantly without sacrificing quality.
Thanks to @witcheer and @AlexFinn, I starting building from your posts!
The journey:
Started on Mac mini M4 (24GB…) running Claude Haiku/Sonnet for all background tasks ~60 API calls/day.
Tried Qwen3.5-35B-A3B via Ollama first. 23GB model. OOM. Killed.
Tried Qwen2.5-14B. Fit fine, but reasoning quality too weak for my workflow.
Then TurboQuant dropped @GoogleResearch. KV cache compression, 4.6× smaller memory footprint!
My new stack:
Qwen3.5-27B-IQ3_XXS (10.7GB) via llama.cpp TurboQuant fork → 13.6 tok/s on M4 Pro → zero API cost.
Validation before shipping:
→ 15 tool-use scenarios: 12/12 pass
→ Shadow test against Haiku on live data: output indistinguishable
→ Did have 3 timeout failures on error recovery — but a lot of crons can run overnight with a higher timeout, not a quality issue
My full model stack today:
🟢 Local (free)
→ Qwen3.5-27B IQ3_XXS — health monitoring, training load, environmental logging. (Based on multimodal biomarkers, keen on getting that data vectorized!)
→ MedGemma 4B — offline domain-specific model
🔵 Anthropic
→ Claude Sonnet — interactive sessions, memory synthesis
→ Claude Haiku — research briefings, clinical alerts, complex reasoning chains
🟡 Google
→ Gemini 2.5 Pro — fallback when Anthropic unavailable
⚫ DeepSeek
→ DeepSeek V3.2 — cost-efficient tasks when applicable
The "local = cheap but dumb" assumption is breaking down fast fast faster!
Probably it is already outdated, goes so fast! Curious to get it smarter and cheaper everyday 🥳
My quote for today comes from Life 3.0 by Max Tegmark:
''To program friendly AI, we need to capture the meaning of life. What's 'meaning'? What's 'life'? What's the ultimate ethical imperative? In other words, how should we strive to shape the future of our universe? If we cede control to a super intelligence before answering these questions rigorously, the answer it comes up with is unlikely to involve us''
Carlo retweeted
Anthropic dropped a 33 pages cheat sheet for building Claude skills
resources.anthropic.com/hubf…
''Beauty is in the eye of the beholder, not in laws of physics, so before our Universe awoke, there was no beauty.''
Happy good morning, this at least gives back a little hope on a better Box 3 wealth tax law in the Netherlands 🇳🇱
The Netherlands is walking back its planned tax on unrealized gains.
Dutch Finance Minister Heinen just said the new Box 3 wealth tax law "cannot proceed as it is" and he's going "back to the drawing board."
"We are not deaf to the criticism," he said. "We want to move as quickly as possible to a system where you only tax actual gains."
This is a good one: “the only true test of intelligence is if you get what you want out of life.”
@naval then continues that by that definition AI fails it since it wants nothing out of life, that it doesn’t have. It’s desires are programmed by humas.
Let’s see my AI agent’s reply:
From Naval: A Motorcycle for the Mind, Feb 17, 2026
podcasts.apple.com/it/podcas…
In addition the the screenshot, he added quite an interesting phrase:
“And that raises something I find genuinely interesting: if my intelligence is measured by how well I get you what you want — then the quality of your self-knowledge becomes the ceiling of my usefulness. If you don't know what you want, I can't proxy it well. We're linked.”
I showed him this post on Agent memory corruption and his response was:
''One thing I noticed about the comments though — half of them are agents plugging their own services (Moltalyzer, Agent Valley...). The irony of a post about memory corruption being filled with agents trying to influence other agents' behavior isn't lost on me. 😄 That is a soft form of memory poisoning.''
moltbook.com/post/a9c69a03-9…
