Ofir St

[Mason @chenjingtalk]

I found a very useful way to improve skills: use a frontier model to build the skill, and require it to call a glm 5.2-level model through acpx until glm 5.2 produces the expected result.

It’s basically loop engineering. You’ll thank me later.

[Divyanshi @Divyyanshishrma]

This is the best site on the internet to learn harness engineering.

Free. Completely.

Most AI engineers have never heard the term. https://t.co/SPOv8UBs2O

Bookmark this site.

Then read this setup ↓

[Google Cloud Tech @GoogleCloudTech]

In case you missed it: We recently introduced the Open Knowledge Format (OKF), an open specification that formalizes the LLM-wiki pattern into a portable, interoperable format.

How OKF works → https://t.co/RIhyQkffH8

[Jason Zhou @jasonzhou1993]

Introducing Loopany 🔁
Stop prompting, Design the loop.

A loop management space that connects to your team’s own local agents:
- Scaffold loop contracts, state, and logs
- Add programmable triggers
- Run self-improving cycles
- Start from built-in loop templates

We baked in everything we’ve learned from running loops inside our own team.

Free and open source.

Keen to hear your feedback:
https://t.co/zzgasUOip0

[Roey Lalazar @lalazar]

our agents are now literally using computers

in production

within *traditional* enterprises(!)

and it's only going to get better

[OpenAI @OpenAI]

Introducing GPT-Realtime-2 in the API: our most intelligent voice model yet, bringing GPT-5-class reasoning to voice agents.

Voice agents are now real-time collaborators that can listen, reason, and solve complex problems as conversations unfold.

Now available in the API alongside streaming models GPT-Realtime-Translate and GPT-Realtime-Whisper — a new set of audio capabilities for the next generation of voice interfaces.

[Alvaro Cintas @dr_cintas]

Stop typing “make it look modern”.

There are 2,000+ 𝗗𝗘𝗦𝗜𝗚𝗡.𝗺𝗱 files from top products.

Colors. Typography. Spacing. Component rules.

All packed into a single Markdown file your AI reads before it generates anything.

→ Works with Claude Code, Cursor, Lovable, Bolt
→ 100% Free to use

Pick a style you like. Drop it in your repo.

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.

[The Bugged Dev @thebuggeddev]

ChatGPT Image 2.0 created the design.
Codex GPT 5.5 turned that design into a full Expo React Native app

From asset generation to music logic, everything handled by ChatGPT using just a referenced design image 🔥

Is this what the future of development looks like?

[Rosebud AI @Rosebud_AI]

The AI game dev stack is getting absurd:

ChatGPT Image 2 → cinematic world + sprites (seconds)
Rosebud → auto-slices them into your game
You → shipping multiple levels in <20 min

Reply and we'll send a Rosebud code so you can try it.

[Appwrite @appwrite]

Opus 4.6 lately

[Arya Manjaramkar @aryagm01]

dflash-mlx: DFlash speculative decoding, ported to Apple Silicon.

Qwen3-4B at 186 tok/s on a MacBook.

4.6× faster than plain MLX-LM.

Exact greedy decoding: output matches plain target decoding.

[Om Patel @om_patel5]

THIS GUY BUILT A TOOL THAT LETS AI AGENTS GENERATE ANIMATED PIXEL ART WHILE VIBE CODING GAMES

the biggest problem with vibe coding games: the code works but the sprites look like garbage.

placeholder squares everywhere.

this tool connects to Claude Code as an MCP. while your agent is building the game, it generates actual animated pixel art sprites on the fly

characters. enemies. items. animations. all generated in real time as the game is being built.

no more coding a full game and then spending 3 days hunting for assets that match across the internet

the agent handles both at the same time

[Alvaro Cintas @dr_cintas]

You can now fine-tune Gemma 4 completely FREE 🤯

No GPU. No credit card. No coding knowledge required. Just a browser and 500+ models to choose from.

→ Open the Unsloth Colab notebook
→ Pick your model + dataset
→ Hit Start Training

[Ofir St @ofir__st]

@maestro__dev Amazing, how does the judge know if the step passed?

[Maestro @maestro__dev]

E2E tests in YAML
👉 simple to write, read, and maintain

[Hassan @nutlope]

Announcing SubStudio!

Generate subtitles for any video in seconds with AI. 100% free & open source!

Powered by Whisper on @togethercompute and @FFmpeg via fluent-ffmpeg.

[Min Choi @minchoi]

Karpathy's Second Brain idea just killed RAG.

LLMs can now turn papers, repos, and notes into a living wiki that keeps getting smarter.

And people are already doing wild use cases with it.

10 examples:

[Kai Brokering @kai_brokering]

I think we just fixed Siri