Sanji Nai-Chien

[Sanji Nai-Chien @SanjiChien]

In case you missed it, I made a video on editing with Claude+CapCut

It shows a much simpler workflow for creating videos without getting lost in AfterEffects. Still one of my favorite editing stacks.
https://t.co/kFLnHdZAc1

[Sanji Nai-Chien @SanjiChien]

OpenAI expects to have AGI by the end of this year.

And Astra is already able to implement ideas, run experiments and complete work that once took human researchers a week.

The next step is persistent agents that keep working on their own and potentially discover new knowledge.

I’m so pumped.

[TIME @TIME]

TIME’s new cover: In 2026, OpenAI has seen key departures, rogue AI agents, major lawsuits, and has seen increased competition in the AI race. “We clearly had some missteps as a company,” OpenAI CEO Sam Altman tells TIME.

Inside the company’s plan for a reboot: https://t.co/1TDJ1zaFCS

[Sanji Nai-Chien @SanjiChien]

Anthropic may be about to move first: Fable 5.1 is reportedly being prepared for launch as soon as tomorrow!

[leo 🐾 @synthwavedd]

Well... looks like they [Anthropic] may not be waiting around for Astra after all. Preparations are ramping up for a launch as soon as tomorrow - and if they don't want to risk being embarrassed, they'll need it out latest early next week.

⏳

[Sanji Nai-Chien @SanjiChien]

Higgsfield just shipped Relight It’s basically post-capture lighting control.

You can reposition light and tune their color and intensity after the shot.

A pretty big deal for post-production tbh

[Higgsfield AI 🧩 @higgsfield_ai]

Meet Higgsfield Relight 🧩

Add professional lighting to any shot after you shoot it:

> Drag lights around your subject from any angle
> Set color and brightness for each one
> Change the mood of an entire scene

Now available on Higgsfield.

[Sanji Nai-Chien @SanjiChien]

GLM-5.3 Flash has 320B total parameters but only activates 18B, reportedly beating GLM-5.2 at one-teeth the price.

If the real-world performance matches the benchmarks this could be huge release.

[AiBattle @AiBattle_]

GLM-5.3-Flash 320B parameters 18B active

[Sanji Nai-Chien @SanjiChien]

OpenAI’s first custom AI chip “Jalapeño” is already running real models in the lab.

It delivered up to 4.1x higher performance, 3.6x lower latency and nearly twice as much work per watt in tests.

Deployment starts by the end of 2026, while Gen 2 and Gen 3 are already underway.

It feels like OpenAI is becoming a full-stack AI company👀

[OpenAI @OpenAI]

Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it.

The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without sacrificing efficiency.

[Sanji Nai-Chien @SanjiChien]

Finally an official announcement
OxAlpha is https://t.co/1mvIamsuek’s new GLM-5.3 Flash!

The company confirmed it to Bloomberg and the drop is TONIGHT.

[Bloomberg @business]

The new AI model that’s swept to the top of online usage charts with high performance at zero cost was created by China’s https://t.co/q5KG9bMPxw, also known as Zhipu https://t.co/mYlzIWdtqQ

[Sanji Nai-Chien @SanjiChien]

OpenAI’s first AI chip is already faster and more efficient than NVIDIA’s GB200 and GB300.

Jalapeño delivered up to 4.1x better performance and 3.6x lower delay while using less than 550W 👀

Maybe 750 tokens per second really will feel normal in a year or two

https://t.co/GMx97Es3Yf

[Max Weinbach @mweinbach]

Holy fuck OpenAI cooked https://t.co/E4uR54bXyz

[Sanji Nai-Chien @SanjiChien]

Do I really need it? No
Do I love it? Holy moly yesss

[Manish Kumar @Manixh02]

Found a fun way to generate QR codes that don’t look like QR codes.

- Enter a URL and it grows a 3D tree from the code itself
- The ground grid is the actual scannable QR
- Switch between Spring, Summer, and Autumn looks
- Runs fully in the browser, nothing uploaded
- Tap the tree to flatten it and scan

https://t.co/133ukElFea

[Sanji Nai-Chien @SanjiChien]

Cool concept of deep-sea robot from Kimi K3

[Dilum Sanjaya @DilumSanjaya]

I asked Kimi K3 to design a robot for deep sea research

Here's what it came up with

A jellyfish inspired robot with tentacles built from a hexagonal nanostructured lattice, designed to withstand extreme pressure.

[Sanji Nai-Chien @SanjiChien]

@ModelScope2022 Curious how it performs on a 96GB setup👀

[Sanji Nai-Chien @SanjiChien]

Grok Voice hit #1 on the speech-to-speech index!

It’s taking 15000+ Starlink calls a day and actually fixing problems.

I didn’t expect voice agents to get this good this fast tbh

[SpaceXAI @SpaceXAI]

Grok Voice Think Fast 2.0 is now #1 on the Artificial Analysis Speech-to-Speech Index.

This index measures whether voice agents can reason over the speech it hears, resolve real customer issues, and correctly complete tasks using agent tools.

[Sanji Nai-Chien @SanjiChien]

A new Mac mini could be just days away!

Apple has tested M5 and M6 versions as demand grows for running AI locally.

This might be the most exciting Mac mini

[Mark Gurman @markgurman]

NEW: Apple is gearing up to launch a new version of the Mac mini, as early as ahead of its September iPhone launch. https://t.co/k8Zz9g0SFI

[Sanji Nai-Chien @SanjiChien]

Ox Alpha keeps getting better right in front of us.

Same prompt gives drastically better output day after day.

I’m so excited, can’t wait to find out what model it is👀

[Tim Jayas @TimJayas]

Ox Alpha has clearly stated that user prompts and completions are retained but not used for training

so the only possible explanation is they swapped the model under the same codename OxAlpha

if my theory is true then OxAlpha is not just one model but multiple variations being tested under the same hood

[Sanji Nai-Chien @SanjiChien]

@nvidianewsroom Latency is basically gone

[Sanji Nai-Chien @SanjiChien]

medical AI can pass all tests and still struggles with an actual patient’s messy 10 year history and that’s the real problem IMHO

[Sanji Nai-Chien @SanjiChien]

GPT Astra is confirmed, new Claude models (Opus 5.1 + probably Haiku) are in testing, OxAlpha is probably GLM-5.3 Flash

And now Qwen 4 got spotted too

September is stacking fast👀

[White @White1637402]

Qwen's newest model is in the Arena under the name "paloma".

Its frontend coding capabilities feel like they might be on par with 5Opus. 👀🔥

#Qwen

[Sanji Nai-Chien @SanjiChien]

Bro really woke up to a $3M market cap 👀

[Sanji Nai-Chien @SanjiChien]

@antpalkin I’m curious to see what happens when bot actually goes live w real money

[Sanji Nai-Chien @SanjiChien]

@antpalkin I’m curious to see what happens when bot actually goes live w real money

[Sanji Nai-Chien @SanjiChien]

Humans are now the minority of AI users

Agents now consume 7.3T tokens and it’s nearly 5x human usage, up 14x since February.

One more proof that 2026 is the year of AI agents

[a16z @a16z]

Humans are the minority user of AI

Agents burn nearly 5x the tokens people do, up 14x since February

Charts of the Week: https://t.co/7YT2BXvuUu