Jason Milldrum retweeted
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
Jason Milldrum retweeted
Run Inkling 2 Mini, the strongest 26.4B model that runs on 16GB RAM.
Thinking Machines’ Apache 2.0 MoE model uses a looped transformer arch & scores 48.2% on DeepSWE v1.1, the best for size.
Run on Unsloth, llama.cpp
GGUF: huggingface.co/unsloth/Inkli…
Guide: unsloth.ai/docs/models/inkli…
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
Jason Milldrum retweeted
BREAKING: A mathematician at Anthropic claims to have used the company's AI to solve the Riemann hypothesis, the most notorious open problem in all of math
sciam.com/article/anthropic-…
Jason Milldrum retweeted
Haven't seen anyone talking about how sick the new Starship webcast intro is yet.
Remember seeing this circuit used as a super decoupling cap for SSDRA-era post-mixer amps. I always wondered if it was a bit of overkill.
Understanding the Transistor Capacitance Multiplier
When designing linear power supplies or sensitive analogue circuits—such as RF phase-locked loops, high-resolution data converters, or hi-fi preamplifiers - clean power rails are paramount.
However, achieving ultra-low ripple using massive electrolytic capacitors is often impractical due to board space and cost.
Enter the transistor capacitance multiplier.
By placing a smoothing capacitor on the base of an emitter-follower transistor, the effective capacitance seen by the circuit is multiplied by the transistor's current gain.
This allows you to achieve the smoothing performance of a massive capacitor using a fraction of the physical footprint.
However, the basic capacitance multiplier circuit has a notable limitation: with only a single series resistor feeding the base, the voltage drop across the collector-emitter junction is often too small, which can lead to unexpected ripple glitches as the capacitor charges to its peak.
Why the Potential Divider Version is Better
By incorporating a second resistor from the base to ground to form a resistive potential divider, circuit performance improves significantly:
* Optimized Voltage Drop: Lowering the base voltage slightly increases the voltage drop across the series pass transistor (typically maintaining a safe margin of at least 3V), ensuring the transistor operates correctly in its linear region.
* Enhanced Ripple Rejection: The increased headroom allows for vastly superior attenuation of power supply ripple and high-frequency noise.
* Stable Bias Across Loads: Sizing the potential divider current to be roughly 10 times the base current ensures that the emitter voltage remains stable across wide variations in output current.
Adding that second resistor to ground is a simple design tweak that yields a dramatic improvement in overall power rail integrity.
For more information check the link in the comments.
#ElectronicsDesign #AnalogCircuits #PowerSupply #HardwareEngineering #ElectronicsNotes
16
All that, plus a grapple if you're in a weedy, woody area like I am. The grapple is an amazing attachment.
If you’re going to be a homesteader..
BUY.
A.
TRACTOR.
Yes it’s the equivalent of buying a new car but for $750/Year and a $250 deductible, this entire tractor can be replaced with a new one should something go wrong and I can’t find insurance like that anywhere but for this thing.
Most people would finance a tractor but in my unique situation I’ve been cash heavy and credit poor for years.
So we bought it out right.
(Yes I know thank younger me for creating adult me’s problems)
On top of that I have 3 attachments that I use almost daily.
1: Bucket: $4500 attachment.
Grading ground, flattening, carrying almost anything, moving material from one place to another, emergency ladder, reaching high places, unloading soil, loading up other items, moving sick or injured animals, moving brush, moving firewood chopped and pre chopped, giving kids rides, light digging of top soil, turning over soil, doing driveways, snow plowing, you name it.
2: Backhoe. $7500 attachment.
Most expensive attachment so far but if we didn’t have this bad boy we could remove stumps which was a huge problem we have here in NH owning wooded land. It used to take us forever by hand with multiple guys what now can be done in an hour or less depending on stump size. The backhoe has also been fantastic for crushing items we don’t need and tearing down pens, fences and buildings we no longer have use for. It’s saved us 100’s of hours already.
3: Forks. $1500
I bought these on a whim but have used them at least once everyday since purchase. I really didn’t know how bad I needed this. I’ve used them for moving firewood bins, pallets, carrying fencing, moving small structures and tables the bucket couldn’t, moving the new makeshift water tower we built to stop hauling buckets, loading trailers, moving appliances, unloading freight, loading freight, and more.
The forks crossover from farm use to being used for our reselling business and it’s awesome.
Forks and bucket already paid for themselves and the backhoe is getting there.
I’m not sure if we’ll end up needing any other attachments beyond these 3 but this post in general was just a love letter to Kubota and Tractors.
I put off buying one for years and I shouldn’t have.
15
>The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
Wooo! 🎉
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery!
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
- $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%.
The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder.
Start building today! 🚀
- Blog: qwen.ai/blog?id=qwen4-max
- API: qwencloud.com/models/qwen4-m…
- Token Plan: qwencloud.com/pricing/token-…
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
36
I have like zero reach since I'm starting from zero again on this platform, please RT!
I'm pleased to announce that I have added an MCP server to QRPedia.com! Point your favorite AI tool at it and get design answers straight in your harness. More MCP-connected tools coming in about a week.
qrpedia.com/mcp-server.html
I'm super looking forward to Omarchy 4.6 with the Local AI functionality built-in. Sounds like it will have a DB of what the best, proven models are, which will be very useful.
20
I now have working local inference on my AI shed PC served on my Tailnet. Radeon AI Pro R9700 (32 GB VRAM) on a Ryzen 7 3700X PC. Hermes gateway running on the shed PC and Hermes Agent frontend on my laptop. Here are my impressions of the setup so far...
2
1
134
I also got Grok Bot set up to had all coding tasks off to it for now, and all it will do is check its work. That will be interesting to see how it goes, and if it saves me usage in Grok Bot.
1
7






























