iiinova — where your digital self keeps evolving

This is unintentionally funny - Salesforce's SIForce.

0151

Qwen3.8-27B in a 7.89 GB file that loads in stock llama.cpp, and how it compares with other 2-bit builds.

Underdog Saluki 27B in 7.89GB Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling Blog and Model: underdog.ai/saluki

underdog.ai

Underdog Saluki 27B

2861

LightOnOCR-3 Leading benchmark results : ahead of Chandra-OCR-2, Mistral OCR 4.1 and dots.mocr on OlmOCR-Bench, and first among open-weight models on ParseBench. Blog and Models: huggingface.co/blog/lighton... Website and Models: www.lighton.ai/research/lig...

0211

Odyssey-3 It can generate interactive environments from a prompt, all in real time. It's a new kind of world simulator, and learns representations of physics, dynamics, and cause-and-effect from a broad dataset of visual observations. Blog: odyssey.systems/meet-odyssey-3

5878

MedDecider An open weight family of Jev style decision models for medicine. MedDecider achieves SOTA results on variety of medical tasks including knowledge questions, ICD codes, trial notes. Model: huggingface.co/lion-ai/MedD...

1231

How to raise VC funding for your AI startup: 1. Be based in San Francisco. 2. Be a 19-year-old male. That's the pitch deck.

3520

There seems to be a shift in the types of cloud services being marketed to developers. First, it was cloud environments for running Codex and Claude Code. Now, it's cloud infrastructure for running GitHub Actions CI runners.

280

AI agents love rust. Do they ever recommend PHP?

110

Sung Kim @sungkim.bsky.social · 08/10/2026

I’m maturing as an AI-native coder (i.e., vibe coder). At the beginning of the year, whenever I needed a tool, I figured I could just build it myself rather than pay another monthly subscription. Now, I’m realizing that both building software and maintaining it is a hassle. 😁

3540

It's a torture for anyone. Why write a book when you can just write a paragraph?

000

Sung Kim @sungkim.bsky.social · 08/10/2026

I believe Anthropic is trying to protect us from our future AI overlords. www.anthropic.com/news/2026-us...

7191

Sung Kim @sungkim.bsky.social · 08/10/2026

Microsoft and Nvidia: You can run an AI model on this Surface Laptop Ultra Us: Look at that sweet magnetic USB-C port.

4392

Sung Kim @sungkim.bsky.social · 07/10/2026

Elon Musk starts a rumor about a joint venture with TSMC. Now Elon Musk is casting doubt on that same joint venture. He’s starting to behave like Trump: starting something in his own head, ending it in his own head, and then declaring SUCCESS!

2341

Sung Kim @sungkim.bsky.social · 07/10/2026

I’m not sure about a lot of things in this world, but I’m sure of this: if we start a witch hunt, we lose. Please shut this British guy up! "Ex-Anthropic Jacob Coxon says he's 90% certain China already has spies inside both OpenAI and Anthropic"

5161

Sung Kim @sungkim.bsky.social · 07/10/2026

SpaceX is not a serious company, judging by its plans for Terafab.

3263

Yes, AMD has been focusing on GPUs, not CPUs; while Intel just focused on CPUs because of its lack of good GPUs.

000

What am I missing? Intel is constrained by its own fabs, but TSMC doesn’t exactly have spare capacity sitting around for AMD either.

250

I like the discussion around AMD versus Intel, particularly which company is likely to gain more share in data center CPUs. What I’m struggling to understand is why so many people expect AMD to gain significantly more share when recent performance seems to suggest otherwise.

250

I'm convinced about one thing. Instead of selling AI agentic SaaS to the traditional professional services companies, it may be easier to just compete against them using AI agentic software.

5311

Unsloth AI @unsloth.ai · 06/10/2026

I’m convinced of one thing about Tibo: he’s a poor communicator, aside from the occasional reset. It took me a while to understand what this feature is actually about.

2160

Reposted by Sung Kim

Google releases EmbeddingGemma 2, a new open model that runs locally on 0.5GB RAM. The 740M parameter embedding model combines a 270M text model with vision (170M) + audio (300M). Run & train the model via Unsloth. GGUF: huggingface.co/unsloth/embe... Guide: unsloth.ai/docs/models/...

3989

Reposted by Sung Kim

Open Athena @openathena.ai · 01/10/2026

Marin's 535B-parameter hero run was originally planned as a 360B model. A new blog from Rafal Wojdyla, MTS, explains how a change in sharding strategy, Expert Parallelism, made room for a ~50% larger model at the same ~23B active parameters, and what it costs. 🔗 openathena.ai/blog/expert-...

0136

Two Madison One Madison is a daughter of the most powerful man in the country. Another Madison is a daughter of CEO of the most valuable company in the world.

151

There's a reason for blocking... 😁

000

I'm getting used to seeing this GitHub page.

280

Dust, the first zeroth-order method to pretrain transformers to approach and sometimes even *exceed* backprop with large amounts of computation. Full paper: qlabs.sh/research/dust Code: github.com/qlabs-eng/dust

27111

I would buy this and get a SFF case then a nice computer...

110

Here at Ghost, we are building Core, the personal AI computer designed to keep intelligence and control in the hands of the people who use them.

I was going to say, this is a nice price ($3,499) for a really well equipped computer, but it has NO ports! - NVIDIA RTX PRO 4000 Blackwell SFF GPU - 64GB DDR5 RAM - 1TB NVMe M.2 SSD - AMD Ryzen 5 7600 CPU ghost.ai

ghost.ai

Ghost AI

8231

Reflection AI's Beam: a highly efficient agentic open model with 501B total parameters and 23B active. - Frontier reasoning efficiency - Advances the Western open frontier on coding & agentic tasks - Trained end-to-end from scratch Full weights release this month. reflection.ai/blog/introdu...

4544

It looks like AI chat is moving toward the cloud. Claude Chat and Cowork are effectively merging into a single Claude experience, with agentic tasks running on cloud computers by default.

7221

All 8EG6 models including the Honor Magic 9 Pro Max, Xiaomi 18 Pro Series and iQOO 16 have reportedly experiencing similar issues.

130

George Black @george.black · 04/10/2026

There are reports that first product using TSMC N2P, which is Qualcomm Snapdragon 8EEG6, are reportedly facing crashes, black screens and random reboots, with the issue likely becoming more evident under heavy loads.

1100

Reposted by Sung Kim

Using Cloudflare’s Clef to sort my email 👌. Order updates, receipts, etc are filed away automatically, and I get a push notification!

huggingface.co

Cloudflare/clef · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

2211

Used both extensively and dots sucks. Dots where it stands is another chat.

270

The funny thing is that OpenAI hired the guy behind OpenClaw. Why wasn’t he leading this project?

2231

Look at what Meta did. They acquired Manus because Manus understood agents. Meta later had to unwind it, but the underlying point still stands: product instinct matters.

1260

OpenAI’s Dev Day 2026 was an unmitigated disaster. Now they’re trying to recover over the next 28 days. We’ll see. If it were up to me, I’d fire the cool, hip, highly educated people and hire more people who actually understand products and know how to ship them.

4643

100+ Charts On The State of Markets (1H’26). It's the best. You wouldn't want to miss it.

Just think: only 2.2% of U.S. households are financing the trillions of dollars being spent on AI data center expansion. www.a16z.news/p/state-of-m...

a16z.news

State of Markets II

050