If you have Google AI Pro then you can use 400 compute units of GPUs on Google Colab:
- G4 GPU
- A100 GPU
- L4 GPU
- T4 GPU
- v6e-1 TPU

@sungkim.bsky.social
A business analyst at heart who enjoys delving into AI, ML, data engineering, data science, data analytics, and modeling. My views are my own. You can also find me at threads: @sung.kim.mw
Interesting... ~5,000 people are internal users of frontier AI models. ~20 million are advanced frontier AI users. ~1–2 billion are casual AI users. The rest of the world has yet to use AI.
Then it asks you to sign in to ChatGPT again. Seriously? You ask it to update a Space with new criteria. Another ChatGPT sign-in. And then you discover it's consuming your Codex usage to do the work. What a mess.
Wow. OpenAI's Dots is a complete cluster****. I'd fire the entire team. You're already in ChatGPT. You try Dots because it's agentic and free with your subscription. Since everything lives in one long conversation, you ask it to create Spaces to organize your work.
DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair. It even works when the images and the captions come from different datasets. Project page: dominik-schnaus.github.io/unpaired-rosetta ⬇️
At this point, I do not even question her competence - she is incompetent.
It's kind of funny that I'm voting for a democratic socialist for mayor of Los Angeles, hoping she'll cut spending and crack down on waste and abuse. Our current mayor has presided over a city government plagued by financial mismanagement, inadequate oversight, and documented fraud.
If you have Google AI Pro then you can use 400 compute units of GPUs on Google Colab:
- G4 GPU
- A100 GPU
- L4 GPU
- T4 GPU
- v6e-1 TPU
This is unintentionally funny - Salesforce's SIForce.
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
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...
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.
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.
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.
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.
AI agents love rust. Do they ever recommend PHP?
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. 😁
It's a torture for anyone. Why write a book when you can just write a paragraph?
I believe Anthropic is trying to protect us from our future AI overlords.
Microsoft and Nvidia: You can run an AI model on this Surface Laptop Ultra Us: Look at that sweet magnetic USB-C port.
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!