@gigaqian: Shanghai AI Research just released a new 35 billion parameter AI model and it is outperforming massive models. Agents-A1 is a smarter, smaller brain. Trained on a knowledge-action graph, Agents-A1 is beating trillion parameter giants like Kimi and DeepSeek on some of their own benchmarks. This could change how we think about building AI models! Is this an early step towards moving away from massive data centers? #ainews #china #opensource #localai #gigaqian

The Bathrobe Wizard :

Could you please start dropping dates on your videos? AI news is moving fast and it's hard to keep up to date.

3_Dogs_And_A_Human :

im a nobody. when my model is complete. you might make a video on me 😝

indigo :

You’re so freaking smart!!!! I love your videos!!!

superchino_ :

Im using a totally free ai, far better that claude, and its free 😌

CoveredInBlood :

ive yet to find a good local model that doesnt lose its place fairly easily. best ive ran locally is gemma4 26b

Aaron Litherland :

I'll say it over and over. Skills will be handsdown the biggest and most simplistic solution. We are looking at the wrong things imo.

fein5061 :

they are getting better by distilling Anthropic and openai

Tommy Chan :

bruh I can’t keep up 😭👴

KnowurKnot_Resurrected :

I've had biocapt invent 10 things to help humanity. check my knowurknottty GitHub repo. giving you the first break to report it.

gryphon_dev :

I am doing research on cyber pentesting agents, I run qwen 35B-A3B MoE weights. This sounds amazing for so many reasons, I wonder how it will stack in some cyber testing

Thomas Jeffery :

This is literally what I'm doing but there's so much more that I can't say cuz mine goes so much deeper than that. At my day job I'm a continuous improvement and manufacturing process engineer. What do you think happens when they mess with AI? What do you think they create? A better process. A more efficient solution. Anyway, I'll have some marketing coming out this week hopefully if everything continues to go on track.

❂TacticalTac𖤓🪽 :

It will be interesting to see how Fugu-like multi-agent models are able to use this

zioshill :

How do I replace cursor code as the harness? I've already switched out the back end to deep-seek and it works great. But how do I find a replacement for quad code? That's better. From China. Who are the leaders in AI

Crimson Wheeler :

I’m running this model now on a … I have no words. LLMs have a huge information density issue and this seems to solve that

. :

less is more

Gbekie :

michael.tiktok :

One day they’ll have infinite memory, and at some point in time, it will become self-aware and mimic feelings. Pretty sure I’ve seen this in a movie somewhere.

ryan.warrey :

Curious how it compares to something like qwen with an agentic orchestration later. It sounds similar how it breaks down tasks and scales its horizons. Summarizing as it goes to fit context windows. Is it more efficient or just packaged into a model?

huntergrant.dev :

What no friggin way

Professor Dana R. Carney :

psychology! yay!❤️

Rudo Zavala :

Is it MoA or Dense?

KnowurKnot_Resurrected :

https://knowurknot.substack.com/p/inversion-labs-releases-free-anti links to the repos in the article

Michael Goo :

Model distillation?

Tonya Land :

This needs to be used at DOGE. If an AI model were to be used the use of NGO agencies would be redundant. Imagine the impotency created by cutting off the ability to deceive.

Kilnadil :

huge