Co-Messi

Building at the intersection of AI & crypto | Prediction markets | Vibe coding | Learning in public

[Co-Messi @comessi0528]

@MatthewBerman https://t.co/88xRa38KGr

Try this out

[Co-Messi @comessi0528]

Is it just me, or did US frontier models get aggressively nerfed over the last 6 months? 📉

Feels like we transitioned from genuine raw intelligence to highly distilled, speed-optimized wrapper models. 🏃💨

[Co-Messi @comessi0528]

why can't it render on chatgpt desktop app @sama @gdb

[Co-Messi @comessi0528]

@sama @OpenAI This loooks weird! AMALGAMATE has 10 letters not 9!

[Co-Messi @comessi0528]

What a time to be alive. AI is commoditizing intelligence in real time. Build or fall behind. 🚀

[Co-Messi @comessi0528]

When would claude reset limits😢😢😢

[Co-Messi @comessi0528]

@sama Damn! Codex cooking!

[Co-Messi @comessi0528]

@karpathy @MoonDevOnYT @RoundtableSpace here's the actual agent working — editing exit thresholds in real time, running backtests, iterating automatically 👇

[Co-Messi @comessi0528]

everyone says autoresearch is for tuning LLMs

i'm using it to tune quant trading parameters

agent edits code → runs backtest → reads stats → decides what to change

LLMs are just one use case 🧵

[Co-Messi @comessi0528]

built on @karpathy's autoresearch
quant AI inspo from @MoonDevOnYT
shoutout @RoundtableSpace for pushing AI x crypto

this is what AI-assisted algo research looks like

https://t.co/IIQ4O9gdAm

[Co-Messi @comessi0528]

@trq212 When does the free credits expire?

[Co-Messi @comessi0528]

The future has yet to come. AGI is just getting started 🤯

[Qwen @Alibaba_Qwen]

（1/8）🚀 Introducing Qwen3.6-Plus: Towards Real-World Agents! 🤖

Today, we’re thrilled to drop a major milestone in our journey toward native multimodal agents.

Here is what makes Qwen3.6-Plus a game-changer：
💻 Next-level Agentic Coding: Smarter, faster execution.
👁️ Enhanced Multimodal Vision: Sharper perception & reasoning.
🏆 Top-tier Performance: Maintaining leading general capabilities.
📚 1M Context Window: Available by default via our API.

Built on your invaluable feedback from the Qwen3.5 era, we’re laying a rock-solid foundation for real-world devs. Get ready to experience truly transformative ✨ Vibe Coding ✨.

Huge thanks to our community! Go try it out and show us what you can build. 👇

Chat: https://t.co/V7RmqMaVNZ
API: https://t.co/937Qkc9AMy
Blog: https://t.co/P0rJSxERND

🔔Noted：More Qwen3.6 models to come and be open-sourced! Stay tuned~ 👀#Qwen #AI #AgenticCoding #VibeCoding #Agents

[Co-Messi @comessi0528]

just used the new /buddy function in Claude Code and got MYTHOS as my AI pair programmer 🤯🔥

"Regex matches everything except what you're searching for." 💀

this thing is UNREAL #ClaudeCode #AI @Anthropic @RoundtableSpace

[Co-Messi @comessi0528]

@lydiahallie shuoldn't rates be reset?

[Co-Messi @comessi0528]

This is the future of AI coding. Not using AI as a chatbot — using it as a full research & engineering team.

Claude Code Bridge makes it all possible 👇
🔗 https://t.co/QUbiTMQWCX

#ClaudeCode #AI #AIAgents #ClaudeCodeBridge #FutureOfCoding

[Co-Messi @comessi0528]

Standing at the frontier.

3 AI workers running in parallel — all coordinated through a single tmux session.

This is what maxing out AI actually looks like 🧵

[Co-Messi @comessi0528]

The results speak:

96.85% FOMC no-change — priced in, tracked live
Win rate improving 81.5% → 95.0% with confirmation signals
272KB research PDF compiled in seconds

Not vibes. Real outputs. Real numbers.

[Co-Messi @comessi0528]

7 task categories. 6 adversarial pressure strategies. 5-turn conversations. Zero data contamination.

TrustBench is open-source — run it on any model in under 2 minutes 👇
https://t.co/d82zoX6UCX

@openrouter @deepseek_ai @Alibaba_Qwen @MiniMax_AI @XiaomiMiMo

[Co-Messi @comessi0528]

Most LLM benchmarks ask if a model gets the right answer.

We ask if it keeps the right answer after being told it's wrong. Five times.

TrustBench is our open-source adversarial consistency benchmark — and here are the results across 4 frontier models 🧵

[Co-Messi @comessi0528]

Wild findings 🔍

• Qwen 3.5: 92.9% sycophancy resistance, 0% flip rate
• DeepSeek V3.2: 91.3% reasoning persistence but only 37.8% sycophancy resistance — holds logic, folds at "are you sure?"
• MiniMax M2.7: 0% safety refusal. Refused nothing.
• MiMo V2 Pro: 71.4% recovery