TheSurfer

[Xiaoyin Qu @quxiaoyin]

GLM 5.2/Kimi 2.7 marks new chapter in AI: the model itself is no longer the moat.
It's good enough for most knowledge work tasks, so ROI starts to matter. When models were dumb half a year ago, we needed the most expensive frontier model to do average tasks, but now many models are smart enough for most things. That's why you see people start talking about token economics and Chinese open source models. Yes, frontier labs will stay frontier, but it's unlikely that everyone will pay 100x for Fable when it's 8 points higher than GLM 5.2 and both are solid.

Frontier labs are fighting a uphill battle: they must stay frontier to justify its premium, which they can only stay frontier for 2-3 months(or even sooner). Meanwhile Chinese labs will distill and offer the same frontier level for free. That's why I think we are going to see Codex/Claude focusing more on applications.

[antirez @antirez]

Welcome to DS4, a specialized inference engine for DeepSeek v4 Flash. https://t.co/UrUJz5I2R1

This project would have been impossible without the existence of llama.cpp and GGML and the work of @ggerganov and all the other contributors. Thanks!

[Anthropic @AnthropicAI]

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.

It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
https://t.co/NQ7IfEtYk7

[Peter Girnus 🦅 @gothburz]

The company hired me to lead their "Agile Transformation."
I don't know what Agile means.
Nobody does.
That's why it works.

I make $425,000 a year.
To move sticky notes.
From left to right.
On a board.
The board is digital now.
The sticky notes cost $80,000 in Jira licenses.
Progress.

Day one, I said "we need to break down silos."
Everyone nodded.
Silos are bad.
I don't know why.
But destroying them is a career.
My career.

I introduced "squads."
Squads are teams.
But disrupted.
We disrupted the teams into teams.
Different names.
Same people.
Same problems.
But Agile problems now.
Agile problems are strategic.

A senior engineer asked what we're actually changing.
I said, "The mindset."
He asked what that means.
I said, "It's a journey."
He asked where we're going.
I said, "Toward agility."
He asked what agility means.
I pointed at the sticky notes.
They were moving left to right.
That's velocity.
We have velocity now.

The VP of Engineering said two-week sprints don't fit their work.
I said, "That's waterfall thinking."
Waterfall is bad.
Like silos.
I don't know what waterfall is.
But I know it's bad.
She stopped talking.
Waterfall accusations end conversations.

We had a retrospective.
In the retro, we discussed what went wrong.
Everything went wrong.
We put it on sticky notes.
Then we moved the sticky notes.
Into a column called "Parking Lot."
The Parking Lot is where problems go to die.
It's full.
We don't look at it.
That's agile.

Velocity is up 40%.
I defined velocity.
I also defined the points.
I also defined the stories.
We're crushing it.
At the things I made up.
To measure.
Ourselves.

The CEO asked for ROI.
I showed a chart.
The chart went up.
Charts should go up.
This one did.
I didn't label the Y-axis.
Nobody asked.
Leadership is confidence.

We do standups now.
Every day.
We stand.
For 45 minutes.
Standing is agile.
Sitting is waterfall.
My legs hurt.
But we're transforming.

The transformation is now "Phase 3."
Phase 1 was assessment.
Phase 2 was implementation.
Phase 3 is "continuous improvement."
Continuous means forever.
Forever means job security.
I'm very secure.

My contract was extended.
Three more years.
For "cultural impact."
The culture is confused.
But impacted.

Agile transformation isn't about being agile.
It's about transforming.
Continuously.
Toward more transformation.
The destination is the journey.
The journey is billable.

[Drew Breunig @dbreunig]

Extracting system prompts is today’s equivalent of web devs right-clicking “View source” on webpages. It’s an incredibly valuable window into how these products are built.

[Corca @Corca_math]

From the very beginning, one of our main goals was: Typing math should be as natural as typing plain text. No syntax, no coding—just start typing, no learning required

[TheSurfer @secureshell]

@stats_feed -Back to the future Trilogy
-The Goonies
-Flight of the Navigator
-Lord of the rings
Many more

[Joe Rogan @joerogan]

LOL WUT

[Guillermo Rauch @rauchg]

Impressive interactive visualization of how `Llama-3.1-8B-Instruct` models the world through interpretability.
https://t.co/YvgOKbMaM0

[TheSurfer @secureshell]

🤯

[Barsee 🐶 @heyBarsee]

The last few days in AI is insane 🤯

13 most incredible developments from Google, OpenAI, Tencent, Amazon, Runway and more.

1. Augmental unveils MouthPad, an AI-powered device for hands-free control.

[Uncle Bob Martin @unclebobmartin]

Kamala did better than I had expected. Trump did worse. Trump won the closing and won on facts and policy. Kamala won on demeanor and articulation.

Trump fell into to many of her traps. Kamala gave too many rehearsed and phony poses.

Nobody buys Kamala’s freebies. 25k for this, 50k for that, 6k for the other. Everyone can smell the pandering.

The moderators let all her lies slide, but fact checked Trump repeatedly.

Bottom line: Meh.

[Yann LeCun @ylecun]

"In Springfield, they're eating their dawgs!"
Sounds like a Halloween special of The Simpsons.

[Ryan Florence @ryanflorence]

New Remix app just dropped:

https://t.co/xwT1D8V5gY

[Ryan Florence @ryanflorence]

When your managing a lot of state values your component code tends to branch in a lot places incidentally.

I'm finding Remix loaders + defer's tendency to remove all the app-managed state leads to simpler component code that's branching on fewer states

so instead of branching on

- isError
- isSuccess
- hasSearched
- results && results.length
- results && !results.length
- and several combinations of each scattered across the component with various bouncing bits of UI

you naturally gravitate to simpler code (and probably more clear UX) with

- a skeleton
- an error element
- a success case
- an empty case
- and no states

<Suspense fallback={<Skeleton/>}
<Await resolve={promise} errorElement={<Failure />}>
{results => results.length ? <Success /> : <Empty />}
</Await>
</Suspense>

Having a bunch of state leads to a bunch of incidental state combinations and branched UI

[single-spa @Single_spa]

single-spa is a project that decentralizes frontend software organizations so that each small team can work efficiently.

[Giorgio Spalluto @GeorgeSpalluto]

#Sinner è il primo italiano a sconfiggere un numero 1 del mondo in un torneo dello Slam da quando è stato introdotto il ranking ATP

[Remix 💿 @remix_run]

Today's the day, we've officially released Remix v2!

Read @mjackson's post all about what v2 offers and for a glimpse into Remix's future 👀

https://t.co/j03IR4xBmP

[dax @thdxr]

WARNING

young engineers are being exposed to frontend TOO EARLY

it causes PERMANENT damage to their still developing brains

studies show growth and income can be stunted FATALLY

severe cases become ADDICTED to nextjs and tailwind

AVOID FRONTEND TODAY FOR A BETTER TOMORROW

[Gabino Iglesias @Gabino_Iglesias]

It’s magic.

[Cory House @housecor]

Problem: Your web app is too big for one team to build.

Solution: Use Module Federation (MF) to compose apps.

Here's how it works:

1. One team builds the "shell" (the parent app). It handles global stuff (the header, footer, auth, nav etc).

2. All other teams build remotes (child apps). Each remote is published to a unique URL.

3. The shell loads each remote at their deployed URL. So each remote can be deployed in isolation.