Armin Catovic

OpenAI just open sourced a new 1.5B (50m active) model on HuggingFace with Apache 2.0 license! It's not a new LLM, this one is called Privacy Filter, and it's a PII detection model (checking if text has private information) A few interesting tidbits from the release + links:

@Layton_Gott I love Claude Code but purely from sec PoV, you would rather trust CC - a proprietary tool with models running out of your control - instead of a local Qwen model - bunch of floats and matmuls - running on your own hardware???

How do people TRUST Chinese models?! I barely trust Claude Code… Take security SERIOUSLY.

Very excited to see parallel agents feature in my IDE of choice, Zed: zed.dev/blog/parallel-…

New in Claude Code: /ultrareview (research preview) runs a fleet of bug-hunting agents in the cloud. Findings land in the CLI or Desktop automatically. Run it before merging critical changes—auth, data migrations, etc. Pro and Max users get 3 free reviews through 5/5.

@avshah99 This is awesome work! I spent a year building a legal AI startup. Do we see a future where AI agents will be legal representatives? Will it be agent-to-agent dispute resolution or case law?

🚨New preprint! We find evidence of LLMs enabling people to file lawsuits without lawyers (filing "pro se") at historically unprecedented rates in federal courts.👇 1/n

I mean this is insane, 27B model beats a previous gen 400B model across all benchmarks

@Alibaba_Qwen Unreal, as always, amazing work Qwen team 🫶

🚀 Meet Qwen3.6-27B, our latest dense, open-source model, packing flagship-level coding power! Yes, 27B, and Qwen3.6-27B punches way above its weight. 👇 What's new: 🧠 Outstanding agentic coding — surpasses Qwen3.5-397B-A17B across all major coding benchmarks 💡 Strong reasoning across text & multimodal tasks 🔄 Supports thinking & non-thinking modes ✅ Apache 2.0 — fully open, fully yours Smaller model. Bigger results. Community's favorite. ❤️ We can't wait to see what you build with Qwen3.6-27B! 👀 🔗👇 Blog: qwen.ai/blog?id=qwen3.… Qwen Studio: chat.qwen.ai/?models=qwen3.… Github: github.com/QwenLM/Qwen3.6 Hugging Face: huggingface.co/Qwen/Qwen3.6-2… huggingface.co/Qwen/Qwen3.6-2… ModelScope: modelscope.cn/models/Qwen/Qw… modelscope.cn/models/Qwen/Qw…

Dear followers, please see this conversation with Jon Stewart and David Autor on AI, work, inequality and learning. I personally had a great time.

@ghosttyped Imagine back in 1993. I wanted Windows 3.11 but it somehow blended Windows 95 and 3.11 (it actually did a web search for inspo)

Imagine if codex existed in 2005

GPT-ImageGen-2 did this in one shot, with just the prompt "turn all of Tennyson's Ulysses into a comic, across as many pages as needed. make it great, include the full text" 10 pages, though it did use what seems to be the ImageGen-2 's preferred "spackled drawing" style 1/

@emollick That's insane! For re-creating educational material this is a game changer.

@AxelMtbr @OliverMolander It will be a terrible shame if Alibaba drops open models, I know they are focusing on commercial tier but so far they’ve been keeping in step with open models. Not just LLMs but TTS and ASR also

@acatovicx @OliverMolander Here’s the link ft.com/content/b39da3… I don’t understand your logic re Meta. Whether or not they bet on LLMs or some other architecture is a different question whether they release it as open weights or not. Clearly Muse cluld be open weights.

I find it extremely sad that Europe couldn't find a way to dominate in open source / open weight models. Like we created Linux. We should have done better. Big opportunity lost, looks like China will eat the open source AI pie at least when it comes to LLMs

@AxelMtbr @OliverMolander Thanks. The logic is that they wasted too much time and money bickering and decided to play it safe and go for closed model glued onto Metas ecosystem.

@bensig No but lets vibe code one!

A friend of mine in LA wants to auction 1 day of human time - he will do anything for 24 hrs... but the buyer has to be an agent. Is there a good website for this?

@ProperPrompter I think this is fine, but spritesheets (motion frames) are still a challenge, e.g. here is the walking one (I previously used Gemini/Nano Banana, and gpt-image-2 is now on par with Gemini in this regard).

GPT Image 2 is rolling out and... Wow. It just one-shot a grid of 100 completely unique pixel art items with meaningful labels in a single image. Prompt in the reply below 👇 More tests with pixel art and Image 2 to come. Follow for more.

@thsottiaux If I could have one wish it's to improve the spritesheet generation capabilities. It's a lot better now (before I would get "pixel noise" issue). But it's still far off and requires heavy editing. But this would be a game changed in indie game industry.

New image model is now the default in Codex. It's incredible as a starting point for building websites or to generate assets for video games and codex can use it as any other tool in its toolkit. Asked Codex what it pictures my day looks like given we hit 4M active users today.

@vasuman In ascending or descending order? And are you testing Twitter user IQ by putting Reddit on 1. and 7?

Social medias ranked by the IQ of their users 1. Reddit 2. Twitter 3. YouTube 4. Instagram 5. TikTok 6. Facebook 7. Reddit 8. LinkedIn

@AxelMtbr @OliverMolander Interesting I didn't see any indication of Alibaba shifting (they do have paid/pro series of course). What's your source? Meta Muse is connected to that because they booted LeCun out and re-positioned. Had they had harmony, they could've gone a long way with open models.

@acatovicx @OliverMolander From my understanding, Alibaba is shifting away from open weight models. Metas Muse is also not open weight, so it’s not a q of LLMs vs world models.

In my spare time I am creating a 2d platformer, and the new gpt-image-2 is now on-par or slightly worse than Gemini when it comes to spritesheet creation (still requires heavy manual editing in aseprite). But it's better than before when I had weird "pixel noise" problems.

@AxelMtbr @OliverMolander Alibaba didn't abandon it - it continues to make Qwen models. Meta AI/FAIR is a management issue (competing strategies from "LeCun world models" camp vs LLM camp, against a failed VR/metaverse backdrop). Of course its not easy, but Chinese proved it can be done.

@acatovicx @OliverMolander Meta and Alibaba didn’t need short term gains but still abandoned it. Additionally, you get a ton of buzz, but it’s hard to endure ever growing cost with a v slow revenue ramp.