Ethan Flare | The Pulse

.

[venus @RitOnchain]

The most profitable equation in a history of hedge funds:

Rt = R0 × (σt / σ̄)^γ

It makes money by ignoring noise.

Bookmark this and then read the article.

[Marcos Hernanz @MarcosHernanz]

After doing ~60B tokens, this is my full AGENTS.md

[Hugging Models @HuggingModels]

Ever wanted a massive AI model that fits on your laptop? Meet DeepSeek V4 Flash, a Mixture of Experts text generator, now in GGUF format. It's quantized down to 2-bit and 4-bit, making it super accessible. This is a game changer for local AI enthusiasts! #AI #DeepSeek

[Mind Matter @MindMatter28]

PRODUCTS THAT WILL MAKE YOU LOOK YOUNGER

1. PURE RETINOL

[Fran Pradas | Ingeniero de IA @franpradasAI]

🚨 ÚLTIMA HORA: China ha lanzado un empleado de IA que trabaja 24/7

Es completamente GRATIS.

Se llama DeerFlow.

- Investiga
- Escribe código
- Crea presentaciones de diapositivas
- Genera videos
- Ejecuta tareas completamente en tu computadora.

Te dejo el repo abajo: 👇

[Krihaa Media @KrihaaBoxOffice]

@OiFaela apply the moisturizer to the desired area and wait 10 to 15 minutes

then, apply the glycolic acid with a cotton pad, without rubbing

repeat the routine 2 to 3 times per week with consistent use, it's common to notice improvement in rougher or darker areas.

[Quant Science @quantscience_]

🚨 BREAKING: A new 33-page PDF demystifying how hedge funds create bias-free signals

This is what you need to know (Number 2 is the most important finding): 🧵

[Alok @analogalok]

If you thought the Gemma 4 31B (dense) model was fast, sit down. I just benched the updated Gemma 4 26B A4B MoE on a single RTX 4090 (24 GB VRAM)

9,200 t/s prefill. 160 t/s decode. 250,000 context window. All on a single consumer RTX 4090. The numbers are completely unhinged.

The 31B is a dense behemoth. But the 26B is a Mixture of Experts (MoE), specifically an Active 4 Billion (A4B). It holds 26B parameters of knowledge but only activates 4B per token. Because its inference memory footprint is so light, I didn’t even need KV cache quantization to hit a quarter million context.

Compiled the latest llama.cpp from source on Ubuntu 22 (CUDA 13). Fed it a 28k token prompt, and manually cranked the batch sizes (-b 2048 -ub 2048) to absolutely redline the Tensor Cores.

Here is the benchmarking breakdown:

# 1. The Baseline (No MTP)
Even without speculative decoding, the A4B architecture flies.

llama.cpp flags:
./build/bin/llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 250000 -ngl 99 -fa on -b 2048 -ub 2048 --port 8080 -v

Context Ceiling: 250,000 tokens (21.5 GB VRAM)
Prefill: 9,200 t/s (Absurd)
Decode: 124 t/s

# 2. The MTP Overdrive
Injected the new MTP draft model to enable Speculative Decoding.

llama.cpp flags:
./build/bin/llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp --spec-draft-model mtp-gemma-4-26B-A4B-it.gguf --spec-draft-n-max 4 --spec-draft-p-min 0.7 -c 250000 -ngl 99 -fa on -b 2048 -ub 2048 --port 8080 -v

Context Ceiling: 250,000 tokens (22.96 GB VRAM)
Prefill: 7,054 t/s (MTP draft overhead slightly caps prefill)
Decode: 156 t/s

# The Agentic Architecture Insight
Why does this matter? Because you can now build a killer local agentic loop on a consumer desktop.
Use the 31B dense model (from the previous post) as your heavy, deliberate Orchestrator / Verifier / Planner.
Pass the actual execution tasks to this 26B MoE.

At 160 t/s, this MoE can chew through code generation, tool calling, and massive RAG document retrieval over a 250k context window almost instantly, drastically speeding up your agentic loop.

If you own a single RTX 3090 or 4090 and haven't tried this specific stack yet, you need to pull these latest updates and run it. Local inference just leveled up.

Hugging Face links to the Unsloth 26B QAT quants and MTP drafters are in the replies. performance graphs also available in the replies.

[Ethan Flare | The Pulse @ethanFlare]

@DikenaClinton It is a Fallout 4 | Jet Chem | 3-D Printed Cosplay Prop
$19.00
https://t.co/VUgJVcd2yY

[Grimmer @0xGrimmer_]

A KID IN A BEDROOM BUILT 160GB OF SERVER MEMORY FROM OFFICE CASTOFFS. HIS PRICIEST PART WAS $150.

that clip is a guy walking you through the little rack in the corner of his room - a stack of second-hand mini-pcs and office desktops, blinking away like a company's server closet shrunk down to fit beside a router.

the whole build is the flex, and the numbers are almost funny. a lenovo thinkstation for $150. an hp elite desktop for $75. sixteen gigs of ddr5 for $20.

about six hundred dollars of used gear he picked up back in february. together it's roughly 160gb of ram - the kind of number a single new machine quotes you a small fortune for.

here's why it works. offices retire last-generation dells, lenovos and hps by the pallet, and "obsolete" on a corporate spreadsheet means "perfect" for a home server.

he's not buying performance he's missing out on. he's buying somebody else's depreciation, and paying pennies on the dollar for it.

and it isn't a trophy shelf. that pile runs a private cloud, backups, media, home automation, a place to host local ai models - real work, on hardware that already earned its price once in an office nobody misses.

the quiet part is the whole point. a corner of a bedroom now does what a small company used to rent a server room for, and it cost about as much as a phone.

no monthly rent, no new-hardware premium, no data leaving the house.

bookmark & watch today ↓

[Ruuj @RuujSs]

Quants don't usually talk about this.

The Sharpe ratio has 5 fatal flaws that make most "alpha" strategies look better than they are: no significance testing, a false Normality assumption, weak test power, p-value confusion, and multiple testing blindness.

This paper builds the full correction toolkit: Probabilistic Sharpe, Deflated Sharpe, Bayesian FDR, MinTRL.

The kind of edge that separates real quants from spreadsheet warriors.

Read it before your next allocation, or your competitors will.

[Oliver Prompts @oliviscusAI]

GOOGLE JUST KILLED THE DOCUMENT EXTRACTION INDUSTRY.

it's called langextract.

▸ extracts structured data from unstructured text
▸ maps every entity back to its exact source location
▸ handles 100+ page documents with high recall
▸ generates interactive html for verification
▸ works with gemini, ollama, and local models

replaces regex pattern matching, custom ner pipelines, expensive extraction apis, and manual data entry.

define your task with a few examples, point it at any document, get structured verifiable results.

100% free.

[Tatara @0xTatara]

this paper is f*cking insane

a Columbia paper built a strictly causal Hidden Markov Model that adapts as market regimes change

the result: 2.18 Sharpe vs 1.18 for SPX buy & hold, while cutting max drawdown from -14.62% to -5.43%

during the 2025 selloff it automatically reduced equity exposure and rotated into defensive assets

the crazy part is it does this without looking into the future

bookmark before this thread gets buried

[Superman @thesupermanmx]

Someone open-source harness that lets Claude Code and Codex generate real 3D CAD models from plain English.

It’s called text-to-cad, an open source harness that turns your coding agent into a CAD engineer..

Building a robot used to mean stitching 6+ tools, CAD software, URDF exporters, kinematics solvers, viewers. Days of work. Broken pipelines. Files no agent can read.

text-to-cad kills all of it.

→ Exports STEP, STL, DXF, GLB, 3MF
→ Generates URDF robot descriptions
→ Slices straight to G-code
→ Sends jobs to your Bambu Lab printer
→ Runs fully offline, no backend

One demo generates a full 7-DoF robot arm, working kinematics, custom GUI, the whole thing, almost entirely through prompts.

100% open source.

[Superman @thesupermanmx]

https://t.co/pDKaT2Saj7

[Superman @thesupermanmx]

NVIDIA open-sourced a 600M model that transcribes 40 languages in real-time at 80ms latency and it costs $0.

that's faster than you can blink. across mandarin, arabic, hindi, portuguese, tagalog, whatever,from a SINGLE checkpoint.

→ 17x more concurrent streams than buffered ASR on the same H100.
→ punctuation + capitalization built-in. no post-processing.
→ runs on your own GPU. no API bill

100% Open Source.

[Atlas @crptAtlas]

GITHUB JUST KILLED THE WORST PART OF VIBE CODING

they shipped a free tool called Spec Kit and it already crossed 120,000 stars

the fix is stupidly simple

instead of tossing vague prompts at an agent and praying it doesn't wreck your project

Spec Kit makes the AI write a full structured spec before it touches a single line of code

it works through the problem first

figures out what you want to build

asks about the gaps

lays out the project

then it starts coding

you get fewer insane bugs, cleaner output and results you can predict

the flow looks like this:

/constitution for your rules and standards
/specify for what you want to build
/clarify for the open questions before you start
/plan for architecture and stack
/tasks for the ordered work
/implement to run it

it plugs into Claude Code, Cursor, Copilot, Codex, Gemini CLI and 25+ other agents

120,000 stars, 10,000 forks, open source, shipped by GitHub itself

learning to drive agents like this is most of what separates people getting hired as AI engineers from everyone still fighting their prompts

[The Eagle flies free @Fa21519230]

🚨🚨 Dosis del tamaño de un guisante de pasta de ivermectina para caballos todos los lunes y martes. Es simplemente como ajenjo y miel...
Una dosis del tamaño de un guisante de pasta de fenbendazol para caballos todos los jueves y viernes. Simplemente es como cáscaras de nuez negra, clavos de olor y miel...
Después de unas semanas, cuando la mayoría de los parásitos intestinales hayan sido eliminados, podrías tomarlo como prefieras...
La ivermectina mata a la lombriz madre, que emite una hormona que impide que se rompan los sacos de huevos que los médicos los llaman tumores...
Cuando se elimina la lombriz madre, los sacos se abren. El fenbendazol mata a las larvas y ayuda a sanar el cerebro y la columna vertebral...

[Dr Sarah E. Squire @DrSarahES]

• Ivermectin – 2 ml
• Mebendazole – 450 mg
• DMSO – ½ tsp
• C60 – 1 tsp (3 weeks on, 1 week off)
• Turmeric – 1,500 mg
• Resveratrol – 500 mg

Note down this medications because the have higher chances of eliminating cancer cells than chemotherapy

[Chen @ch3nweiii]

this is the Heston model closed-form solution

ψ_T(v) = e^(-rT) · φ_T(v-(iα+1)) · S_0^(iv+α+1) / ((α+iv)(α+iv+1))

if you can't read it, that's the point

this equation prices a single option under stochastic volatility
one option. one strike. one expiration

Citadel prices 10 million of these per day
Susquehanna does more

every options desk on Wall Street runs closed-form solutions like this in production, on GPUs, with sub-millisecond latency

retail buys a call because "the chart looks good"
the counterparty selling that call already solved a stochastic differential equation to figure out what price to quote you

that's not a metaphor. this exact math sits between your buy button and the fill you get

Steven Heston published it in 1993
Nobel-adjacent work. free on SSRN. every quant knows it by heart

you were never trading against other retail
you were trading against a PDE with a name

watch the clip