Simi

@BowTiedAsset @FrenchOG3 And I’m sure women wait too for the one, instead of supplying what they need from the free market of dating apps. I can’t believe the solved the problem for everybody by selecting for men’s scarcity and desperation.

@FrenchOG3 This app is different. It doesn't give you choices. It gives you one match in a very long time, so you basically have to give each one a shot, then go back to waiting for months for the next one

From a 2:1 male to female to this is a 12x inversion. If women know about this dynamic, the market structure will make it much more easier for guys to meet with women as there is going to be less fatigue on the woman side trying to filter the hit and quitters. However the question will be: will that objective abundance to the male benefit make the guys change their original inclination to settle down into butterflying around because of the oversupply, or treat it seriously. This is where guys who are complaining of apps giving women the illusion of choices is more a byproduct of their lack of leverage, or actually a legitimate one where they would actually go for something serious given a more genuine opportunity. From a structural and relative standpoint this should give the average man more of a Chad status, now what is he going to do with it is the interesting point.

@CraigSpillard With this calculation the landlord looks like an intermediary for the state, not an independent economic agent

Landlord charges £12,000 rent a year and is charged £6,000 per year on the mortgage interest. Of that £6,000 profit the government gets £4,536 the landlord gets £1,464 Of that £1,464 landlord has to cover repair costs, etc Is it the landlord or the government inflating rents?

@GreatBritishTT The state probably makes 50p on every 1£ in at supermarkets.

The British public thinks supermarkets make 50p profit on every £1. They make 2-4p. "Crooked" energy companies clearing 57%? Actual: 5-15%. "Rip-off" rail on 45%? Actual: 1-5%. The NHS, which makes no profit at all, is reckoned at 34%. The IEA just asked. The answers were a fantasy.

@sudoingX subjectively the tradeoff for me is better than larger context with lower kv cache, keeps doing a good job after the first and second compaction, still not satisfactory quality with q4 kv cache and larger context. Most tasks I give it fit within 128k context.

@sudoingX decided for a similar setup on laptop 5090:60k context q4km, q8 cache and doing the compaction on a laptop 4090 with nemotron 4b at f16. I get slower speeds 20-30tok/s and large contexts take 5-9 min to be processed

this is what my setup looks like today. about to test qwen 3.6 27b dense q4 on a single rtx 3090 at ~41 tok/s gen, hermes agent driving. predecessor model qwen 3.5 dense q4 made it work in one iteration when i ran the same agentic build on the same card. i've been daily driving qwen 3.6 27b dense for weeks now, the model i keep coming back to. if 3.6 oneshots too, this becomes the best model that runs on a single rtx 3090. consumer tier king. firing the test now will report back soon.

@_Unknown_D_ It’s voting economy. It’s the amount their potential voter base would consider a “fair” amount.

My question in regards to the Green Party pledge to raise the minimum wage to £15 is why stop at £15? If it holds true that all we need to do to improve living standards is just increase the minimum wage then why not make it £20, £30 or £40. Whats the economics behind £15?

@LilaRest @huggingface @outsource_ What context size?

Introducing 𝐆𝐞𝐦𝐦𝐚 𝟒 𝟑𝟏𝐁 𝐓𝐮𝐫𝐛𝐨 ⚡️ It runs on a 𝘴𝘪𝘯𝘨𝘭𝘦 RTX 5090, at 51 tok/s (single) and 1244 tok/s (batched). And prefills up to 15359 tok/s. It's 𝟔𝟖% 𝐬𝐦𝐚𝐥𝐥𝐞𝐫 in GPU memory and ~𝟐.𝟓𝐱 𝐟𝐚𝐬𝐭𝐞𝐫 than the base model, and retains nearly 𝐢𝐝𝐞𝐧𝐭𝐢𝐜𝐚𝐥 𝐪𝐮𝐚𝐥𝐢𝐭𝐲 on benchmarks (1-3% loss). Turbo is a derivative of the NVFP4 quant that NVIDIA released a few days ago. It fully leverages NVIDIA Blackwell FP4 tensor cores for ~𝟐× 𝐡𝐢𝐠𝐡𝐞𝐫 𝐜𝐨𝐧𝐜𝐮𝐫𝐫𝐞𝐧𝐭 𝐭𝐡𝐫𝐨𝐮𝐠𝐡𝐩𝐮𝐭 𝐭𝐡𝐚𝐧 𝐨𝐭𝐡𝐞𝐫 𝐪𝐮𝐚𝐧𝐭𝐬. I'm using it for hard classification tasks — on internal benchmarks it showed 𝐒𝐨𝐧𝐧𝐞𝐭-𝟒.𝟓-𝐥𝐞𝐯𝐞𝐥 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 (scored well above Haiku 4.5), at a 600𝘵𝘩 of the cost. A single RTX 5090 scales up to 18 req/s at 1000in/20out 🥵. Model card and benchmark in comments 👇 I'd love to hear your use cases.

@mr_james_c Another case of something brilliant a century ago, but let’s not do it anymore. This is not managed decline it’s managing the place into decline.

@sudosimi I suspect it originated with the tube in London, but like many things that work in Britain it was abandoned after WW2.

Why this wouldn't work in the UK: Buy land → build the train station → develop the area → capture the upside → be forced by statute to give away upside to some Leftwing client group rather than reinvest in more infrastructure

@sudoingX Laptop 5090 with 24G vram - close to 3090 in memory bandwidth.

what VRAM are you working with? i'm planning my next round of benchmarks and i want to test what matters for YOUR hardware. drop your exact GPU below. model, quant, what you're running. i'll tell you if there's something better for your setup.

@ATaylorFPGA @riscvprogram So debugging is conjuring fixes out of the LLM weights of experience and burnt boards?

@riscvprogram Oh it is hard, it's very hard at times.

Why FPGA engineers get paid more? FPGA engineers are rare. Not because it’s hard… But because most people quit too early. You need to understand: •digital logic •timing •hardware •debugging without printf

@0xStacking @0xSero On 24gb laptop 5090 with iq4 xs quantisation got it to code, but tool calling with opencode is inconsistent for me. Sometimes it works, other times it hallucinated that it wrote files and did tool calling. Probably opencode harness is not the best thing to use it with

@0xSero can it code though is the question

This is god tier and fits on 24gb + It crushes everything up to 6x its size and ties with Gemini deep think and Deepseek on Math huggingface.co/nvidia/Nemotro…

@UKPropertyTruth Do you have a breakdown on leaseholds versus freehold loss rates?

We analysed every property resale across London in 2025 using HM Land Registry data. Overall loss rate: 12.9% But some areas are 2-3x higher: E14 (Canary Wharf): 34.3% NW9 (Colindale): 30.6% E16 (Royal Docks): 28.9% SE10 (Greenwich): 28.7% W14 (West Kensington): 24.7% Which area do you want us to look at next?

@DellAnnaLuca After 3 days in Japan I started picking trash and was getting annoyed at other tourists.

Japan, which famously doesn’t have public bins, is also famously clean. It’s a habits problem.

@levelsio @TermiusHQ @AnthropicAI I sftp it and prompt to read

This is a big problem I have actually I wish @TermiusHQ and @AnthropicAI together would solve

@ATaylorFPGA @thinkdefence Will those be first to volunteer when conflict breaks out?

@thinkdefence Classic example of defence procurement. Jobs over capability or delivery

What a headline

@Bengal_DPT For adductor I did progression to get to weighed copenhagens. But for when I don’t have the time for isolations the closed chain squat is the best compound for me. Now leg day is progression on the Nordic curl and reverse nordic. Nordic curl provides plenty of adductor tension.

@sudosimi Are you doing narrow stance width squats? Really doing just adductor exercises and learning to create pressure with your foot contacts is all that's required to get correct adductor tension.

Good friend is mid 30’s, in shape, never been hurt, works out 5x/week, has a fancy downtown gym membership. Recently had knee pain, went to see a physical therapist. Has a $100 copay. Every visit they passed him off to tech who watched him do tib raises. Never got better. Modern rehab is broken.

@Bengal_DPT I'm developing a adductor squeezer that's comfortable to hold, the yoga ball or pilates ring slips easily and I got to the level I want to add progressive overload to the closed chain squats.

@sudosimi Adductor tension is very important for the lower body.

@rohanpaul_ai Are the tactile sensor design and making guide be open sourced too?

ORCA Dexterity just open-sourced 3 anthropomorphic robotic hands, tendon-driven with self-dislocating joints for reliability. 3D-print the free STL files, add off-the-shelf motors, assemble in under 8 hours for ~$2,200, and you have full hardware for dexterous manipulation research in any lab. One of their variants is the orcahand touch. It comes with up to 83 taxels per fingertip, 1mm resolution, and 0.1N force detectability.