Francis

Hi

[Francis @slinky3d]

@nicksdot @PanNiewidzialny @MarcosHernanz I agree that less deps is good, but you don’t wanna spend hundreds of dollars on token to implement something that already exist. It’s a waste of money.

[Francis @slinky3d]

@ascetic_shadow You can just add a response to steer it away from artifact

[Astrid Wilde 🌞 @astridwilde1]

if anyone has suggestions for how to stop Claude and GPT from spending 60% of their time writing and running inane tests that accomplish nothing but waste tine and tokens and clog my codebase i would love to hear them

[Francis @slinky3d]

@annew234 @noonasplashlove Judge from hell.

[How To Prompt @HowToPrompt__]

Google Deepmind argues that LLMs can never make real scientific discoveries.

They published a paper breaking down Albert Einstein’s private view of scientific discovery.

In a famous letter to his friend Maurice Solovine, Einstein drew a diagram of how science actually happens.

It is a cyclical loop.

First, you experience raw sensory data. Then, through a mysterious, non-logical act of intuition, you make an intuitive "jump" to abstract axioms. Finally, you use strict logical deduction to derive consequences from those axioms.

Generative AI has completely mastered two-thirds of this loop.

• Induction: Statistical pattern matching across billions of tokens.
• Deduction: Formal proof generation, like AlphaProof solving complex math Olympiads.

AI can crunch data and it can prove theorems.

But it cannot make the jump.

The paper argues that AI completely lacks Abduction, the generation of novel explanatory hypotheses when observational data is scarce.

The prevailing tech myth says that "creativity is just data compression." That if you feed an LLM enough text, scientific breakthroughs will naturally pop out.

Einstein’s formulation of General Relativity proves that is a delusion.

When Einstein formulated relativity, the observational data didn't demand a new physics framework; classical mechanics was still massively successful. The breakthrough required a conceptual rupture. An intuitive leap from physical reality to a brand-new set of foundational axioms.

An LLM can execute the math once the axioms are given. But it is structurally incapable of formulating those premises on its own.

It can interpolate inside existing human thought, but it cannot transcend it.

The translation of physical reality into formal axioms remains the absolute, hard bottleneck of artificial scientific invention.

We can build models with trillions of parameters. We can scale compute into the stratosphere.

We can make the calculator infinitely fast.

But until we solve grounding, the machine can process all the data in the universe.

It still can't make the jump.

[Francis @slinky3d]

@DenoDisciple @ggg78g89 Humans are trained on data and guided to predict how to respond to situations by taking in data we see from our birth till present, and inherently we really use prediction in our lives without thinking about it. Llm just does it at a higher scale and compute.

[anne @rosyholp]

Enough with all those ugly directing and acting. Let me present the peak beauty of C-drama directing and acting

[Bharath @Bharath_uwu]

I just saw $1.5 trillion on my AWS bill and my soul left my body

[dilemma @actuallyimthe]

Argentina just beat Spain at the 2026 World Cup final, 3-2.

[0xMarioNawfal @RoundtableSpace]

Anthropic has just published a free course on Loop Engineering

If you use AI agents, bookmark it

https://t.co/r3uEmfCi0L

[NIK @ns123abc]

🚨OPENAI ALLEGEDLY STOLE APPLE’S ENTIRE PLAYBOOK

“Hundreds of billions of dollars, decades of work”

Hardware Engineering

- circuit designs, component architecture, power management
- unreleased products guarded by internal codenames
- AI/ML integration for hardware
- EMI engineering + testing methods

Manufacturing Secrets

- proprietary processes + custom machinery
- equipment Apple builds and installs IN supplier factories
- proprietary metal alloys + finishing techniques
- DFM (Design for Manufacturability) expertise

Component Technologies

- power chips, battery systems, displays, acoustics, touch
- identities of specialized sub-suppliers
- exact specs Apple demands from components

Testing Data

- failure analyses, lifecycle simulations
- “negative know-how” — everything Apple tried that DIDN’T work

Supply Chain

- supplier contracts, allocation strategies
- global logistics coordination
- systems-level integration

ALL GONE 💀

[Aparna Dhinakaran @aparnadhinak]

https://t.co/31rd1aRZNM

[Francis @slinky3d]

Cape verde vs Argentina is a good game! What a game !

[Francis @slinky3d]

@Ghougle2 @250YearsAgoLive He didn’t claim he was in such a group. He downplayed it, testifying he joined to cheat the Tories out of money, then consented to have his name sent to the British ship. It wasn’t just sedition either: mutiny, sedition & treacherous correspondence.

[rody @0x_rody]

https://t.co/0MD6kZbj4O

[Gipp 🦅 @gippp69]

A CHINESE BUILDER TURNED A MAC MINI INTO A PORTABLE AI WORKFLOW BOX THAT CAN RUN 18 HOURS OFF A BATTERY

01:14 he connects the power module to the mac mini and the whole setup stops being a desk computer. now the same tiny machine can run from a battery, drive a portable display, hold extra storage and stay online away from a normal desk.

that matters because most ai workflows do not need a giant server. they need a quiet machine that can keep 12-20 background jobs alive, watch folders, move files, clean notes, sync research and process inputs without babysitting.

paired with claude, this becomes a portable second brain machine. 3 long lectures can turn into notes, 25 saved articles can become summaries, research folders can get organized, and unfinished drafts can keep moving while the owner is away.

the edge is not raw power. it is uptime in a tiny form factor. one small box can run 500-800 micro-tasks a month, while the same workflow inside cloud tools can quietly eat $60-150/month across storage, vps and automation apps.

that is why this build is interesting. the chinese guy is not showing a random charger trick, he is showing what happens when a small computer gets enough ports, power and automation to behave like a pocket-sized home server.

bookmark this before every desk setup gets a battery.

[Yannick Monye @keennay]

- DeepSeek V4 Flash - Native Precision (FP4 + FP8)
- Fits on 2x RTX Pro 6000 GPUs + 256 GB DDR5 RAM
- Using KTransformers: KVCache-AI fork of SGLang for GPU/CPU memory inference

I have a somewhat obsession running applications on resource constrained systems to squeeze the maximum performance possible. Part of that comes from a past life working as a systems engineer, building & upgrading nationwide (USA) Video-On-Demand streaming backends, while navigating headless *nix servers around the time "cloud" was becoming a buzzword.

KTransformers gets less mention across the LLM inference-sphere despite being among the engines listed for many of the popular models on HuggingFace (alongside vLLM, SGLang, & llama.cpp). The KVCache-AI team is best known for providing a forked SGLang for hybrid GPU / CPU memory inference, benefitting MoE models. I expect these hybrid setups to gain in popularity, especially on the consumer side as hardware prices continue soaring.

"Necessity is the mother of invention" as they say, and local AI runners will continue finding more creative ways to run intelligence, whether that involves GPU/CPU memory offload, distributed training / inference, model weight / KV Cache quants, or REAPs.

Here I have DeepSeek V4 Flash running at a 1M context length on 2x RTX Pro 6000s GPUs, using its native mixed precision of FP4 + FP8. KTransformers allows you to reduce your GPU utilization by offloading experts per MoE layer onto GPU VRAM, with the remaining balanced across system RAM. KTransformers also has the ability to update GPU expert placement during inference from routing statistics collected during the prefill phase. There's also a lot of trial and error involved given the limited amount of kernel support for RTX Pro 6000s.

Two of the prompt load stress-test benchmarks I like to run are from the local-inference-lab/llm-inference-bench Github repo & AlienKevin/SWE-ZERO-12M-trajectories HuggingFace dataset.

Here are the main KTransformers SGLang optimized flags:

- Context Length: 1048576
- Total Number of Tokens: 1048576
- Chunked Prefill Size: 16384
- Max Prefill Tokens: 16384
- GPU Prefill Token Threshold: 1024
- GPU Memory Utilization: 87%
- Number of Experts per MoE Layer on GPU: 134 / 256
- Max Running Requests: 256
- CUDA Graph Max Batch Size: 256
- CUDA Graph Batch Sizes: 1 2 4 8 16 32 64 128 256
- Available GPU Memory: 20.81GB (anything less was too tight for agentic coding)

Below are the AlienKevin/SWE-ZERO-12M-trajectories benchmark results for 100 prompts with 10 concurrent, ~8k input tokens, & ~1k output tokens. Both Radix & Chunked Prefix Cache were disabled for the absolute worst-case scenario:

- Prefill Mean Batch Tokens: 35756.93 tok/sec
- Prefill Median Batch Tokens: 652.90 tok/sec
- TTFT Mean: 20.698s
- TTFT Median: 12.714s
- Decode Mean Batch Output Tokens: 27.39 tok/sec
- Decode Median Batch Output Tokens: 20.63 tok/sec
- Utilized CPU memory: ~200 GB

A more detailed write-up will follow, which'll include the methodology of calculating the number of experts per MoE layer on GPU, maximum number of tokens, and GPU memory utilization for a healthy balance for running tool calls & benchmarks in this hybrid setup.

Hopefully this'll be reproducible for you and on alternative GPUs, as well as current & future models. Let me know how it works for you! My future plans involve GPU/CPU memory inference tests for MiniMax M3, GLM-5.2, and Kimi K2.7-Code.

All links for all of the resources getting DeepSeek V4 Flash native mixed precision on 2x RTX Pro 6000 GPUs + 256 GB RAM can be found in the follow up post.

[starmex @starmexxx]

EVERYONE BUYING A $2,000 MAC MINI M4 PRO FOR LOCAL AI MISSED THAT A SERVER RESELLER IS UNLOADING DEAD HP PROLIANT DL380 GEN 10 BOXES WITH 5 YEAR WARRANTY, A $250 USED TESLA P40 BOLTED IN RUNS LLAMA 3.3 70B AT HOME

00:19 he stands behind a pallet, "these 10 HP Proliant DL380 Gen 10 servers, they're a few generations older, you guys will be surprised who these servers are going to"

a DL380 Gen 10 with dual Xeon Gold 6248 and 256GB of DDR4 ECC sells refurbished for $380-500 in 2026, the same chassis cost $14,000 new in 2018 and got written off the books by every fortune 500 by 2024

drop a $250 used Tesla P40 with 24GB VRAM into one PCIe slot, total bill $650, llama 3.3 70B quantized to 4 bit runs at 9 tokens per second locally, the article calls $2,000 mac mini m4 pro the sweet spot, this stack hits it for a third

the seller in the video gives a 5 year warranty on hardware that hyperscalers threw away, the same companies the article warns will jack subscription prices are selling you the rack they used to charge $200 a month to access

a 2018 enterprise chassis pulls 280 watts at idle and screams at 65 decibels, the mac mini pulls 10 watts and runs silent, the tradeoff is real but $1,350 cheaper buys a lot of electricity and earplugs over 4 years

bookmark this and read the article below

[Mnimiy @Mnilax]

https://t.co/LPd0lcUCYN

[Francis @slinky3d]

@J4cks_heart @Mappy6984 The autoignition temperature is 250-280, and some parts of the pipes can go past that, especially for a long ride, so its possible.