Lightnews — Scholar-powered news

Reposted by Anthony Panozzo

Andy Masley @andymasley.bsky.social · 3h

Huel seems fine fwiw thebsdetector.substack.com/p/huel-is-fine

Huel is Fine

Breaking down the issues with Consumer Reports' investigative reporting on lead levels in protein supplements

thebsdetector.substack.com

2 9

Reposted by Anthony Panozzo

Tim Kellogg @timkellogg.me · 3h

Claude Agent Skills

on Claude.ai you can now create skills, which are directories full of markdown instructions and potentially many other assets

a key concept is progressive disclosure, or the agent’s ability to read the files that it wants to gain context

www.anthropic.com/engineering/...

$A flow-style diagram titled “Context Window” showing how an AI agent (Claude) uses system prompts and tools. At the top, a beige box labeled Agent’s System Prompt includes short skill tags like bigquery, docx, nda-review, pdf, pptx, and xlsx. A green annotation on the left says: “Short snippets from each Skill appended to the system prompt.” Below that, a gray message box shows the user’s initial message: User: Fill out this PDF based on what you know about me Attached: /mnt/uploads/order_form.pdf (Labeled with a green arrow: “Initial message from user.”) Then, Claude responds in a yellow box: Claude: Certainly! I’ll use the PDF Skill to help This is followed by a sequence of tool use and results interactions: 1. Tool use: Bash("cat /mnt/skills/pdf/SKILL.md") → Tool result: {stdout: “…contents of the SKILL.md file…”} (Annotated on the right: “Claude decides to ‘trigger’ the PDF Skill by reading it.”) 2. Tool use: Bash("cat /mnt/skills/pdf/forms.md") → Tool result: {stdout: “…contents of the forms.md file…”} (Annotated on the right: “Claude decides to follow the reference from the SKILL.md to the forms.md file.”) The diagram visually represents how Claude builds understanding from the user’s input and dynamically reads relevant skill files within its context window to perform a task.$

2 3 8

Anthony Panozzo @panozzaj.bsky.social · 8h

Probably not super related to your sentiment, but I added a new session hook that runs `date` so CC always knows the date. Avoids thinking commit dates are way in the future when saying "we did it this morning"

1

Reposted by Anthony Panozzo

nic 🔹 @nmk.wtf · 1d

This is a very good post:
www.lesswrong.com/posts/j3gp8t...

How the NanoGPT Speedrun WR dropped by 20% in 3 months — LessWrong

In early 2024 Andrej Karpathy stood up an llm.c repo to train GPT-2 (124M), which took an equivalent of 45 minutes on 8xH100 GPUs to reach 3.28 cross…

www.lesswrong.com

2 3 11

Anthony Panozzo @panozzaj.bsky.social · 2d

Interesting post: AI is Too Big to Fail

AI is Too Big to Fail — Sibylline Software

AI spending looks like a bubble, but political capture and energy realities make backing down impossible.

sibylline.dev

Reposted by Anthony Panozzo

Marc Elias @marcelias.bsky.social · 2d

🚨BREAKING: Federal court DENIES Indiana motion to dismiss lawsuit challenging Indiana’s new law — Senate Bill 10 — banning the use of student ID for voting. www.democracydocket.com/cases/indian...

🚨 Count US IN v. Morales

Learn more.

www.democracydocket.com

29 290 1.2K

Anthony Panozzo @panozzaj.bsky.social · 2d

[2509.25140] ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory

ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory

With the growing adoption of large language model agents in persistent real-world roles, they naturally encounter continuous streams of tasks. A key limitation, however, is their failure to learn…

arxiv.org

Anthony Panozzo @panozzaj.bsky.social · 3d

Interesting post: POV I wake up and stay off my phone open.substack.com/pub/skyfishi...

POV I wake up and stay off my phone

I spent eight (8) hours immersed in Pandora’s hellbox of dirty glass yesterday, according to my screen time calculator.

open.substack.com

Reposted by Anthony Panozzo

rands @rands.bsky.social · 3d

Beads is a lightweight memory system for coding agents, using a graph-based issue tracker: github.com/steveyegge/b...

GitHub - steveyegge/beads: Beads - A memory upgrade for your coding agent

Beads - A memory upgrade for your coding agent. Contribute to steveyegge/beads development by creating an account on GitHub.

github.com

2 6

Anthony Panozzo @panozzaj.bsky.social · 3d

Interesting post: Data centers and electricity - Part 1: They're not raising national prices open.substack.com/pub/andymasl...

Data centers & electricity - Part 1: They're not raising national prices

Yet, anyway, but they have raised local costs a bit

open.substack.com

Anthony Panozzo @panozzaj.bsky.social · 3d

Yes since Sonnet 4.5 my experience has been basically no rate limits even with going full bore and/or multiple concurrent agents (Max plan)

1

Reposted by Anthony Panozzo

Paul Ford @ftrain.bsky.social · 3d

I only wanted to find out how long it would take to get to Bamboo Garden for dim sum and now I’m six levels deep in a platform capitalism gig economy fossil fuel postcolonial military hegemonic semiotic turducken.

1 4 28

Reposted by Anthony Panozzo

Pascal Mamassian @mamassian.bsky.social · 22d

A nice shift in perceived colour between central and peripheral vision. The fixated disc looks purple while the others look blue.

The effect presumably comes from the absence of S-cones in the fovea.

From Hinnerk Schulz-Hildebrandt:
arxiv.org/pdf/2509.115...

An array of 9 purple discs on a blue background. Figure from Hinnerk Schulz-Hildebrandt.

28 250 660

Anthony Panozzo @panozzaj.bsky.social · 4d

Interesting post: Autonomous AI Hacking and the Future of Cybersecurity www.schneier.com/blog/archive...

Autonomous AI Hacking and the Future of Cybersecurity - Schneier on Security

AI agents are now hacking computers. They’re getting better at all phases of cyberattacks, faster than most of us expected. They can chain together different aspects of a cyber operation, and hack…

www.schneier.com

Anthony Panozzo @panozzaj.bsky.social · 4d

Interesting post: From Books to Prompts: The Instant Future of Productivity tomtunguz.com/ai-enables-i...

From Books to Prompts: The Instant Future of Productivity

Early adopters are testing productivity systems in hours instead of weeks. The signals suggest AI prompts will eventually subsume the $30 book market.

tomtunguz.com

1

Anthony Panozzo @panozzaj.bsky.social · 5d

Ridiculously detailed State of AI in 2025 slide deck

State of AI Report 2025

The State of AI Report analyses the most interesting developments in AI. Read and download here.

www.stateof.ai

Anthony Panozzo @panozzaj.bsky.social · 5d

Interesting post: Exploring Active Agent, or can we build AI features the Rails way? evilmartians.com/chronicles/e...

Exploring Active Agent, or can we build AI features the Rails way?—Martian Chronicles, Evil Martians’ team blog

Explore how the Active Agent library brings Rails conventions to AI development, with real-world examples from Evil Martians' early adoption experience

evilmartians.com

Reposted by Anthony Panozzo

Django Wexler @djangowexler.bsky.social · 6d

My boss: so how's our fourth quarter looking?

Me, the sales manager at the company that makes inflatable frog suits: well, you're never going to believe this, but