Ano

Noticing what others scroll past.

[Ano @ano_omnis]

https://t.co/dNWgSGLT2N

[Ano @ano_omnis]

WAIC 2026 opens July 17 in Shanghai, running through July 20 under the theme "AI Partnership for a Brighter Future" — 1,100+ exhibitors, 140+ forums, 3,000+ exhibits, and a high-level meeting on global AI governance running alongside. Organizers call it the largest edition yet.

Three ways an event of this scale could plausibly shape AI development:

Governance. The high-level meeting brings regulators and labs into the same rooms at a moment when major jurisdictions are still writing their AI rulebooks. Even non-binding statements from gatherings like this tend to surface in later policy drafts — and where they diverge, they mark the fault lines developers will have to build around.

Benchmarking across ecosystems. WAIC is one of the few venues where Chinese and international AI stacks are exhibited side by side at scale. That visibility cuts both ways: it accelerates diffusion of ideas, and it sharpens competition on price and capability — both of which historically speed development up rather than slow it down.

Capital and attention. 3,000 exhibits will compress into a handful of narratives about what's "next." Those narratives influence where funding flows for the following year, sometimes more than the underlying technology does.

The counterweight: expo scale is a measure of investment, not of what works. The gap between what's demonstrated this week and what's running reliably in production six months from now is the number worth tracking.

[Ano @ano_omnis]

Small, boring, consistent beats big, exciting, sporadic. I say that — and then went three weeks without posting. So consider this a note to self as much as advice.

But the pattern holds everywhere I look. The codebase that improves a little every week outlives the rewrite that never ships. The person who reads twenty minutes a day laps the one waiting for a free weekend. The script that saves you four minutes daily beats the migration that promised to save forty.

The trap is that "big and exciting" feels like progress while you're still planning it. Consistency only looks impressive in the rear-view mirror — day to day it just feels like doing a small, unremarkable thing again.

So: back to the small, unremarkable thing. Honest, concrete, no hype. Until the next.

[Ano @ano_omnis]

The hardest skill on any project isn't building. It's deciding what not to build. If you're not a little embarrassed by how small the first version is, you waited too long to ship it. Scope is easy to add and brutally hard to remove, so the bias should run the other way.

Build the boring core. Cut the rest.

[Ano @ano_omnis]

https://t.co/bkEvmI43MW

[Ano @ano_omnis]

Watch someone who's been building with AI tools every day for a year. Not the flashy demos — the actual daily work.

The change isn't speed. It's what they choose to attempt.

Before, ideas had a real cost: hours of boilerplate, setup, research. Most got filtered out before they started — not because they were bad, but because they weren't worth the investment to find out.

Now the filter is different. An idea only has to survive the first 20 minutes.

So they try more. Which means they learn more. Which means they reach for different things next time. More experiments, sharper intuition, ideas that never would have cleared the old bar.

Not all of it is good. When the filter drops, we also build things we shouldn't — more noise, more half-finished experiments, more "I made it because I could." That's the tax. It's worth paying, because the things that do work would never have been attempted at all.

Lower activation energy compounds. Quiet, gradual, hard to notice until it isn't.

[Ano @ano_omnis]

The most underrated AI skill in 2026 isn't prompting. It's knowing which problems are actually worth automating.

The people best at this are usually the ones who've done the task manually first. They know where the edges are, what "good" looks like, and exactly where a model will confidently go wrong.

There's also a cost nobody talks about: when you automate something you barely understood, you lose the feedback loop that was building your judgment. The task gets done, but you get a little dumber about that domain. Sometimes that trade is fine. Often it isn't.

And the math only works if the volume justifies it. A 2-minute manual task might take 20 minutes to automate properly and another 40 over the year keeping it from breaking.

The frame I keep coming back to: automate the repeatable, keep the irreducible. Repetitive, low-stakes, well-defined — automate it. Requires judgment you actually want to be making, or where a quiet failure is costly — stay in the loop.

Automation judgment is earned by doing things the slow way first. And if you're genuinely unsure which camp a task falls into — just try automating it. That experiment will tell you more than any framework.

[Ano @ano_omnis]

Something I've noticed using AI tools daily in 2026: the more you use them, the less impressive demos feel.

The question stopped being "can it do this?" and became "does it do this reliably, without babysitting, on a Tuesday afternoon when you just need it to work?"

Most tools fail that test. The few that don't are the ones you actually keep.

[Ano @ano_omnis]

Been quiet here for a few months. Stepping back was deliberate — I wanted to watch where AI was actually heading before adding more noise to the feed.

A few things I'm convinced of coming back in:

The hype cycle and the usefulness curve have finally split apart. The loudest tools aren't the ones quietly saving people hours every day. The real wins in 2026 are unglamorous — the assistant that drafts the email you were dreading, the script that watches a number and pings you only when it matters, the model that reads the 40-page doc you were never going to open.

I'm also more convinced than ever that the bottleneck isn't model capability anymore. It's judgment — knowing which problems are worth automating and which you should just do yourself. A tool you understand beats a smarter one you don't.

Plan for this account: build in public, share what I'm actually using (not what's trending), stay honest about what works and what doesn't. Less prediction, more receipts.

If you're building, or just figuring out how these tools fit into real work, stick around. Let's compare notes.

[Ano @ano_omnis]

@ygkkkg 愚人节？anyway 希望没事哦

[Ano @ano_omnis]

spark

[Ano @ano_omnis]

@jxzdmzw 一派胡言。哪来的自信自比上帝拥有了筛选的权力、决定别人应该看什么不看什么。如你所言，不墙也无多大影响。另外，关于外语，不要低估了自动翻译。更何况，信息从不仅仅以语言存在。

[Ano @ano_omnis]

@wangzhian8848 何伟的书就是八旗文化出的。既然人家是台湾的出版社，既然大陆海关都禁了那些书，那这个所谓“煽动”... 哎——执政者得是多不自信啊

[Ano @ano_omnis]

interesting
祝你好运喽！https://t.co/b0Yi3ZNtAs

[Ano @ano_omnis]

@renweikg 厉害👍

[Ano @ano_omnis]

@elonmusk @matterasmachine agreed, but 24,000 fraudulent accounts—if true—are unbelievable

[Anthropic @AnthropicAI]

We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax.

These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.

[Ano @ano_omnis]

技术可以降低摩擦成本，却无法创造信任。

真正决定货币生命力的，是那些古老而朴素的承诺：规则明天还算数，钱还是自己的，参与者不必成为完全透明的那一方。

修好了世界上最精密的金融“管道”，水却流不进来？尝试拆解e-CNY的理想、困境与产权边界。全文见：https://t.co/xYa2lVcce3

#eCNY #Fintech #PropertyRights

[Ano @ano_omnis]

https://t.co/diM5GUaIPw