DeFi Entrepreneur & Builder 🚀 Co-Founder and Managing Director at @panana_predict and @servrox let’s go 🍌
pump.apt
Joined April 2013
- Tweets1.4K
- Following1.8K
- Followers1.3K
- Likes9K
Pinned Tweet
🏆1️⃣🏆 @panana_predict
🏆 DeFi / RWA:
1st: Panana Predictions dorahacks.io/buidl/16813
2nd: Sentra Restaking Protocol dorahacks.io/buidl/17617
3rd: Legato dorahacks.io/buidl/16659/
4th: eVault dorahacks.io/buidl/16003
5th: MeowFi dorahacks.io/buidl/16628
GPT-6.1 Terra added to my model comparison.
Higher = better AA score; left = cheaper. Dots show effort settings (hollow = none).
Terra High is the cheapest setting reaching AA index 40 among these 22 options.
Artificial Analysis v5.0 · 28 Oct 2026.
Marcel Mayer retweeted
Ok, I just checked this, and its literally running a freaking TensorFlow Lite model on 960×540 images with INT8 quantization every couple of seconds in an attempt to figure out what game you are playing.
It's literally called "alex.tflite" and its literally AlexNet and seems to follow torchvision.models.alexnet. As far as I can tell this does some strange nearest neighbor feature comparisons against a DB and literally doesn't work 99% of the time, you love to see your CPU get wasted every other second by this thing though. It also just seems to dump its results into the void because the consent gate comes later.
Anyways, here's the tf model graph and some dumped captures.
live-tweeting on samsung findings were fun.
i'll let my agents give it a bit more time, but this document is already the most comprehensive breakdown of some key concerns of tizen i've seen run on a live device.
it takes like a good while for me just to scroll through it (there's about 44 writeups as of right now on individual functionality).
release is probably happening in a few days.
Marcel Mayer retweeted
the new jev image model can help you with one of the world's hardest questions, when is the toast done?
use a toast as each step of your Score.
Released a new skill to kickstart your @telegram #serverless development ➡️ stark-ai-de.github.io/agent-…
You now have early access to Telegram Serverless. It allows you to run your bots — including their code and databases — directly on Telegram's infrastructure, without using your own servers.
lets see how this can improve my ai agent telegram gateways #serverless
1/3 I put together two charts using Artificial Analysis data to compare GPT-6.1 Sol on performance and cost.
Sol XHigh matches Astra High’s benchmark index of 51 at ~77% lower cost: $0.39 vs $1.73 per benchmark task.
2/3 I’d use Luna for routine work, test Sol for demanding analysis and reasoning, and reserve Astra Max for cases where the extra performance justifies the cost.
Start with low reasoning effort. Increase it only when your own tasks benefit.
3/3 Data: Artificial Analysis
Sol: artificialanalysis.ai/models…
Astra: artificialanalysis.ai/models…
Snapshot: Sol 1 Oct 2026; other models 28 Sep 2026.
Jev Capability Advisor: 10.44x faster median skill selection than native GPT-6 Luna Low. 132-skill catalog, 48/48 correct.
Compared skills by @PaliwalMuskan19 @AISuperDomain @doodlestein @lomeshdutta + hussi9.
Results + caveats: stark-ai-de.github.io/agent-…
Going to release a nice skill based on Jev from @typesafeai for github.com/stark-ai-de/agent… this week 🚀
youtu.be/qkBx0gMGuhY?si=8hy5… #jev
and here we go ⚡ xitter.cf/servrox1337/status/210…
1/6
Fastest agent helper in town ↯
Our Jev Capability Advisor uses TypeSafe Jev to recommend suitable skills and MCP tools.
In our benchmark, skill selection took 0.49s vs 3.12s natively: 6.40× as fast. 🧵
5/6
🎉 For now, we’re proud to claim the title: Best skill-chooser out there 🏆
Now available as a standalone skill and in the stark AI Developer Plugin for Codex.
6/6
Requires a TypeSafe API key. Automatic use requires host integration.
Code, installation and interactive benchmarks:
stark-ai-de.github.io/agent-…
#AgentSkills #MCP #Codex