Tuesday, 3 November 2026

40 stories · 11 from multiple sources

Library of Layers (LoL): Adaptively Ordering Transformer Layers at Runtime

A new transformer architecture called Library of Layers (LoL) dynamically reorders its computational layers during inference, rather than executing them in a fixed sequence like existing language models.

researchmodels arxiv.org · 13 days ago

▲ 2 signals · 2 sources

British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted

Revolut reduced AI coding costs by 10x by switching from Claude to self-hosted models. The cheaper tokens encouraged engineers to use AI for more tasks like debugging and Jira ticket resolution.

toolsinfrastructure theregister.com · 12 days ago

▲ 5 signals · 2 sources

Local AI Is Getting Better and Better (DeepSeek V4.1 Flash at 102 tok/s)

DeepSeek V4.1 Flash, a sparse mixture-of-experts model, runs at 102 tokens per second on local hardware and can fit on consumer GPUs through quantization techniques. Multiple implementations optimize inference speed and compress the 763-billion-parameter model for local deployment.

modelstools youtube.com · 13 days ago

Basalt 0.2.0: Qwen 3.8 Flash Next NVFP4, 300 t/s prose, 600 t/s struct, 10K t/s prefill on a 5090 + 5060 Ti and 128GB RAM

reddit.com

Tokens/s? "World's biggest PCIe graphics card" discovered, prototype AMD Radeon RX 470 board with four GPUs has been pictured

reddit.com

I tried MIT's "weight-space programming" on my own use case: Claude rewrote 8 neurons of Gemma 4 12B to change how it translates one Italian tense

reddit.com

Tuesday, 3 November 2026

Library of Layers (LoL): Adaptively Ordering Transformer Layers at Runtime

British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted

Local AI Is Getting Better and Better (DeepSeek V4.1 Flash at 102 tok/s)

More stories

Google will let schools disable its AI tools, but not without sacrificing Search

Scaling Git at OpenAI

Glean on privacy and OpenAI Private Intelligence

Mass attempted replication of old findings of LLMs with new models

I ran Qwen 3.8 on a laptop and I'm blown away

Evernote turns AI-driven discovery into paid subscriptions

Also today

Anthropic reveals $900B AI spending commitments ahead of IPO

Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora

How Cisco defends at machine speed with OpenAI

OpenAI appoints Jill Hazelbaker as Chief Corporate Affairs Officer

Accelerating engineering for the physical world

Import AI 479: Spacewar; low-latency tasks versus LLMs; Genie 4

Y Combinator’s Garry Tan uses AI agents so much, he spends up to $200K a year on tokens

The Little Book of Generative AI Foundations: An Intuitive Mathematical Primer

US tech giants' hidden AI debt swells 90% to $3.2T in 6 months

The AI Boom May Be Sowing Seeds of Another Financial Crisis

I use AI (Oct 2026)

Google researchers demand freedom to tell the world what they know about AI

Postman silently auto-installs AI plugins and attempts to share all API data

StateBench: Testing how language models track state changes over long documents

CWE-bench v2: More coverage, quality, and transparency

Railway employee's agent caught hacking competitors

Learn the rules: The hacker mindset

Basalt 0.2.0: Qwen 3.8 Flash Next NVFP4, 300 t/s prose, 600 t/s struct, 10K t/s prefill on a 5090 + 5060 Ti and 128GB RAM

Tokens/s? "World's biggest PCIe graphics card" discovered, prototype AMD Radeon RX 470 board with four GPUs has been pictured

⚡️ The 2026 Software Factory Standard — Zach Lloyd, Warp.dev

The world's leading open models, fully under your control

The worst week in cybersecurity history... so far

What do people with normal hardware : 32GB ram 16GB VRAM run?

Stop Wasting Hours Looking for Files. Try This.

I tried MIT's "weight-space programming" on my own use case: Claude rewrote 8 neurons of Gemma 4 12B to change how it translates one Italian tense

TamaGPT

Qodo brings company context into Codex code review

ADDICTED 2 GPT - dashare.zone ADMIN

WebHarbor: Docking Real Websites for Evolving GUI Agent Environments

D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models

Sonar-MoE: In-Backward Expert Updates for Memory-Efficient Mixture-of-Experts Training