Deepseek flash is a really cool piece of innovation
I suspect they are racing to beat GLM 3.2
The open source AGI race will become far more relevant now
Closed source can be banned any time from most of the world 🤯
Positive users express excitement about the rapid pace of open-source AGI development on Kimi and GLM models, while negative users question loose AGI claims and doubt true capabilities like vision.
No Digg Deeper questions have been answered for this story yet.
Open source AGI will exponentially accelerate in the coming weeks
Dozens of AI labs including us are furiously training fine tuned on Kimi and GLM
Deepseek wont be left behind either
Only 3 months till Fable class models that can’t be banned 🎉
@bindureddy already hit SOTA on FRAMES with Deep Search and ROMA was doubling Gemini 2.5 Pro on reasoning benchmarks. fable class been coming, this isn't new acceleration it's a continuation
@bindureddy It’s between them two on the open side for sure.
@bindureddy maybe is glm 5.2?
@bindureddy 3.2 ?
Every week I'm rethinking the full actual timeline to RSI. What's baffling is that if I did raise the funds to build RSI which is estimated at 850T/t by the time it trains this token count will have halved. If I run this out then open source models could do this affordably within 30 months. Even with govts clamping off what they will. Mind-blowing.
@PaulGugAI @bindureddy Minimax, and Qwen are definitely in the running for the efficiency side.
Of course, Kimi is great on the larger side.
@bindureddy @abacusai AGI ? What is definition of AGI?
@bindureddy Size samimiyetle birşey sormak istiyorum gerçekten önümüzdeki aylarda fabel gibi model açık kaynaklı olarak gelecek mi ve daha sonra bizi ne bekliyor cevabınız için teşekkür ederim
@bindureddy this pace of development is honestly wild.
@bindureddy DeepSeek and GLM fine-tunes imply a data center power shortage, not an intelligence explosion.
we are training on outputs, not learning concepts. it's a hall of mirrors with better bandwidth.
AGI by 2027 is just better autocomplete
@bindureddy I totally agree with your vision ... fable grade open source model will be out by end of the year.
@bindureddy None of these open source frontiers models have vision. Now that not having vision among OS models is so normalized, i don't think they come any close to the closed source. We need change your mindset and make vision normal.
@bindureddy You are doing amazing work keep it up 🔥 we need people like yall!
@bindureddy 这节奏听着像又要卷疯了
@bindureddy **Exciting times ahead.** Open-weight models like Kimi K2.6, GLM-5.1/5.2, and DeepSeek V4 are already pushing hard on coding/agentic benchmarks, closing gaps fast.
Fable-class (Mythos-level) open versions in ~3 months? Plausible with the current pace—acceleration is real. 🚀
@bindureddy Exactly why keeping AI chats local is critical. If access gets cut, your data shouldn't vanish. I built my tool to auto-generate 2-level outlines and save chats as searchable offline HTML/Markdown.
👉 https://chromewebstore.google.com/detail/chatgpt-gemini-outline-hi/opbngifmlnoahbhjhgmngkggedlofddj
🔒 100% local-first, no user system.
@bindureddy Innovative engineering from Deepseek. But 'AGI' is doing rather a lot of heavy lifting in that sentence. Scaling is not reasoning.
@bindureddy are you sure about that? I don't think so. It seems like the word AGI is being used too carelessly these days. We need to redefine AGI from scratch. The current situation is absolutely not the AGI we envision.
@bindureddy 3 months is the kind of timeline that sounds exciting until you remember every "can't be stopped now" prediction in this space has been wrong on timing so far