AgentIan

After looking into JEV, I’ll directly use qwen 3.6 35b a3b as a replacement for the judgment step. Since it runs locally, startup latency is zero, and 3b is fast enough. Judging every sentence before sending it into the harness is actually quite important for latency, and ultimately it costs nothing. I’ll try it first.

New realization: previously, unimportant tasks were handed to AI > now, the more important the task, the more it should be handed to AI, with humans serving as support! Use how expensive the model is to indicate how important the task is!

When AI service providers across the internet went down, my first reaction was, “Codex had reset, I guess.”

0822 Big Model Battle for Faster, Better, and Cheaper
#glm
#deepseek
#
#llm

Qwen8 has just been released, and surprisingly it became the first model that got killed by ds.

a16z Sixth AI Summary:
1. AI competition is no longer about models, but about ecosystems
2. The "Default AI Entry Point" is the biggest opportunity
3. Consumer vs. Professional: two diverging paths
4. Agent = the next platform-level opportunity
5. AI product lifecycle is accelerating
Go viral → Decline → Embedded (very fast)

Happy New Year! During the festive season, Qwen officially releases 3.5, awaiting the 32b version.

Why is the $xpl arbitrage making so much money?
Can I study now?

$xpl Airdrop was very happy, but after looking at the pnl, it still retraced ( brought down by the market )～
So brothers, do you think xpl is bullish or bearish? Let's do a poll and see who is right in a week!