Pinned Tweet
We are LIVE at @CyreneAI
0x9663985492af1b339d959a800592ed131090f264
cyreneai.com/projects/luna-n…
Building business infrastructure for solo operators and business owners who are scaling.
gonna be real with yall @liquidai's LFM3 is probably the iphone moment for local AI.
It's the first time I've seen a model on consumer tier hardware be legitimately useful.
the 1.7 B runs on a lot of things, and I think we are only gonna see more capability as time progresses.
AJ retweeted
solo wednesday 👋
this is microowners, a club for one-person businesses
built with LunaNineAi.com
It needs to be said @deepseek_ai has such a good reputation that people treat an outage as, ahh they are preparing for the new model to be served.
Only @Alibaba_Qwen, and @Zai_org have similarly good reputations, albeit in different parts of the community.
If open AI had a similar outage we would be roasting them, with deepseek we are cheering them on because they are chill.
DeepSeek went down today, and China is already sure why.
The status page logged it at 16:58 Beijing time. About 1 hour 44 minutes of partial outage on chat, the V4 Pro API and the V4.1 Flash API.
Chat and Flash came back first.
The V4 Pro API kept answering "Service is too busy".
Chinese AI accounts all read it the same way: compute is being moved for V4.1 Pro. One user even claims the name deepseek-v4.1-pro already answers through a Tencent service.
DeepSeek said nothing, as always.
V4.1 Flash went from 0910 to 1017 in about 5 weeks on the same base. 1017 already runs my agent sessions on 4x 3090.
If Pro gets the same treatment, it will be scary.
I am ready for it.
Simplify, add lightness.
wise words from Colin Chapman
we've spent the last few days simplifying things behind the scenes at LunaNine, removing unnecessary code from the site builder
it's easy to assume a better product means more features, more options, more things users can do
but a big part of building software is knowing what to remove, especially when you're building for people who didn't sign up to become software experts
they just want to get their website up, change it when their business changes, and get on with running the business
the goal is simple: you shouldn't need to understand code to run your website.
AJ retweeted
new week!
running a business alone means there will always be more to do
so choose what matters most.
AJ retweeted
Spooky Saturday!!
somewhere between building software and being mildly haunted by bugs, we found time to get dressed up
happy halloween from our little corner of the internet. 🎃
AJ retweeted
One of the best articles about AI website builders.
If you are a solo operator focused on sustainable business results, make sure to read this article, and definitely check LunaNine 🌙
That's a proper positioning statement in the world of AI slop.
AJ retweeted
We've just made a breakthrough: the first true omni model running 100% locally in a browser... fully end-to-end, with no separate ASR or TTS models needed.
With custom WebGPU kernels, we're able to run LFM3-Omni-2B at ~280 tok/s on my M4 Max, accepting visual, audio, or text and producing interleaved text and audio tokens. Just 90 ms to the first token and 140 ms to the first audio.
The future of on-device multimodal AI agents is going to be wild.
Question for the #keep4o community, which date/snapshot of 4o do you prefer from these three?
- gpt-4o-2024-05-13
- gpt-4o-2024-08-06
- gpt-4o-2024-11-20
Which one did you enjoy talking to more
This.
Difficulty used to be a heuristic for "is this complexity worth it"
Now we have to develop the skill to evaluate good design without the help of constraints like time or cost of engineering.
We all will know local AI is viable as an investment for the general public when private equity firms start buying out DGX sparks and mac studios.















