AI helps me get more done than ever before, but I'm still running out of time
All I ever wanted was to work less and get everything done
33
one month on linux after 20 years on mac!
mac is good. nothing wrong with using one. but i want more control over my computer.
i like to change things. how the system works and how it looks. i don't want to fight apple just to turn something off or change something they decided for me.
on linux i can do that. it's my system.
I was affraid to buy non mac computer so i first tried on mac in double boot. then i decided to risk and it was right decision as i see now.
i'm using @omarchy and it feels like the ubuntu moment back in the 2000s, but now with ai.
and that makes a big difference. setting things up is easier. installing software is easier. i don't have to spend hours reading forums to figure everything out.
if an app doesn't exist i can build it myself with ai. much faster than before.
that's what i like about linux now. i can make the computer work how i want without spending all my time figuring out how.
1
101
this is some crazy thing!! just wow
my vocal synth models the vocal tract as a network of branching acoustic tubes, unlike vocaloid which stitches together and pitch-shifts phoneme recordings. still working on improving naturalness by tuning it against recordings of real singers, but I'm already impressed at how cool it sounds!
3
156
Eugene Pyvovarov retweeted
FULL RECAP: Apple Intelligence Group Lab, the short version for Local AI builders
Sharing highlights from this week's session hosted by @shshnkp (Apple AI/ML Evangelism Team) with @mgorbach and engineers from Foundation Models, Core AI, Siri and App Intents teams at Apple.
1) Apple’s app framework can use your open-weight model as its LLM. Apple’s new app framework can use Apple’s on-device model, Apple’s cloud model or a Core AI converted model, which is interchangeable with a one liner! And your model gets access to the same tool calling, structured output, profiling with Instruments, etc. as Apple’s models.
2) Don’t use Apple’s on-device model for coding, even according to Apple engineers! Use Core AI to bring in a model for coding. One engineer uses a setup where he can use MLX and Core AI models in different “coding harnesses” and reports good experiences.
3) Free local Chat Completions endpoint on any Apple silicon Mac with macOS 27! fm serve provides an OpenAI style Chat Completions endpoint backed by Apple’s on-device model on any Mac. Nothing to install! (shared by Shashank the day after the lab)
4) Apple’s new on-device model is 20B, but only has 8K context. AFM Core Advanced (20B) will be available on the latest iPhones, iPads and M3+ Macs. Can’t opt out of Apple Intelligence in 27, and the model will always be present. Tip: use typed units for fields and let constrained decoding handle it (instead of wasting tokens explaining in prompt).
5) Agentic Siri only works when online. Tool calls from Siri to apps are done using Private Cloud Compute. When offline, Siri will use the on-device model but won’t be able to call into apps.
6) Lessons Apple engineers follow when building agents apply to any agent:
- write down instructions to the model about how/when to use your tools in a markdown file (“skills” are the same idea as agent skills everywhere else!)
- skills are just context to the model, meaning no ordering guarantees. If something should always happen in an order, it should probably be a tool
- make tools accept lists of items to operate on to avoid agent having to call it in a loop
- unit test your tools (deterministic) before testing the whole agent (non-deterministic)
Curated Q&A relevant to Local AI builders in 🧵
My minmap app was featured along other cool mac apps! Check out the article of Clark!
medium.com/macoclock/10-phen…
having always on general agent is a great, in @omarchy 4.5 its done as a quake like console. it has nice appearance, but my god it's ux is horrible, you have to always kick it in / out to copy text or look around while it is thinking.
i have remapped it to open my agent in the corner and have dedicated key to dictate always to it.
@hancore_linux maybe we can work it out somehow?
do you have one claude or codex to general questions which appear day to day? I did have to. Now I just have a plugin with dedicated talk button, which fires chat with pi and paste whatever i say to it and i dont need to remember that general pi chat location. use same voxtype as for other push to talk requests.
46
Missed to charge my laptop with @omarchy couple of times, so changed the battery plugin to turn desktop red when its under 10% of charge, hope i will not miss it any more.
1
48
I think we should have a way to store personal widgets/apps/themes in some personal repository in @omarchy so it is versioned and can be installed / shared on other machines of ours or if we want to share with someone.
32
this is just the beginning, apps will silently(or with a short notice) install skills, plugins etc to your computer without asking.
that's why i develop Agent Guard, to keep your computer secure
github.com/postmanlabs/postm…
30
i guess @ivanfioravanti this one should be tested over by you. you have enough M power in house 🙂
i will try to test on my mac studio soon too.
Replying to @n13
Yes, you can run MLX based models on Omarchy and use the Apple GPU, the Apple Neural Engine, all the hardware.
1
4
444
ahhahaha @grok is going nuts today. been on some snow party for sure 😂
1
1
32
we use deepseek flash at one of the client project and it does the same job openai models did before for 1/10 of the price if not less
Revolut: Europe's most valuable startup ($200B) just showed the world its AI strategy.
Every company will end up copying it.
At the start of this year, Revolut leaned on the frontier labs. Its 2,000 engineers used Claude Code and Codex, and some burned $1,000 to $10,000 a month each. The bills kept climbing, and the returns weren't there.
So Revolut switched its default to an open-source model called GLM. All it needs is GPUs, which it rents from Together AI.
The results, by Revolut's own count:
- The bill is about a tenth of what it paid OpenAI and Anthropic
- Engineers get unlimited tokens instead of spending caps
- Coding time is down 60% and lead time is down 70%
That's the chart below.
Revolut still uses Claude and GPT, but only as "the exception," for the tasks they do best. Everything else runs on open source.
So here's the question I keep coming back to: how many use cases can companies really justify sending to frontier models?
My answer is very few. Plenty of companies still route everything to OpenAI and Anthropic today. For how long?
I think open source ends up handling the vast majority of corporate AI work, and frontier models get a short list of the hardest jobs. That's a problem for labs whose revenue forecasts assume unlimited growth.
If you appreciate unpopular takes, follow me @m0xt_ for more.
18
Eugene Pyvovarov retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Maybe I will keep my M1 Ultra just yet
Hate to break it to you, but the global hardware shortage has not even started yet.
Right now, only 1% of humanity is burning any real amount of AI tokens. The rest are just using free accounts once in a while to ask dumb questions like "Who would win in a fight between a baby seal and 300 hamsters?"
And like Elon said, most RAM will not even go into data centers. It will go into Physical AI, which is robotics.
That is because robotics is the only thing that can actually boost human productivity in a real way.
People on X freak out every time AI solves a hard math problem, while people in the real world just go on with their lives like nothing happened. Meanwhile, robotics is already taking over factories and changing the physical world.
Imagine when these robots start replacing our daily chores. How much more RAM and compute will we need then?
Like I have said many times, building a new fab to make RAM takes at least 5 years, and supply is already fully booked out until 2029.
Even if the so-called AI bubble pops, this massive RAM shortage is not going away.
Starting next year, even the new iPhone might be too expensive for most people to afford.
32
This is for sure some sort of scam from AI labs.
You pay for not guaranteed amount of product(tokens) and they change the amount, the models you use(quant models) after you paid already.
Some for sure do token max but it would not be possible if labs just give exact amount of tokens you get by paying in bulk(monthly).
Avoid any yearly prepaid LLM plan out there.
Labs are starting to reduce the quota to increase revenues.
Churning is your only weapon.
1
93























