Lightnews — Scholar-powered news

Anthony Shaw @anthonypjshaw.bsky.social · 2d

This is the grownup version of

Anthony Shaw @anthonypjshaw.bsky.social · 6d

If you demo an AI agent, it must be a weather demo. If you demo a multiagent, it must be a travel booking demo. This is written in law. Do not question it.

1 6

Anthony Shaw @anthonypjshaw.bsky.social · 6d

Found this as well github.com/microsoft/tu...

GitHub - microsoft/tui-test: An end-to-end terminal testing framework for CLI and TUI experiences

An end-to-end terminal testing framework for CLI and TUI experiences - microsoft/tui-test

github.com

1

Anthony Shaw @anthonypjshaw.bsky.social · 6d

@willmcgugan.bsky.social what did you use for textual testing?

1

Anthony Shaw @anthonypjshaw.bsky.social · 7d

Hacktoberfest 2025 : Getting Claude to implement features like normal but it's someone else's credits and they get a free t-shirt.

3

Anthony Shaw @anthonypjshaw.bsky.social · 8d

How is everyone getting these!!

2

Anthony Shaw @anthonypjshaw.bsky.social · 8d

The kids are on school holidays, so I recruited our youngest to draw a leaf in Scratch, then showed her how to write it in JavaScript 2D canvas.

Anthony Shaw @anthonypjshaw.bsky.social · 8d

🎃It's October! New release of vscode-pets 1.34
- New Autumn (Fall) theme with falling leaves
- New squirrel and skeleton pets (thank you Chris)
I've added several up-for-grabs issues for #hacktoberfest to the backlog.

1 5

Anthony Shaw @anthonypjshaw.bsky.social · 14d

On a #PyConJP adventure with @tiangolo.com to Osaka and Himeji

A selection of photos to Osaka and Himeji

1 2 25

Anthony Shaw @anthonypjshaw.bsky.social · 14d

There is a free tier!

Anthony Shaw @anthonypjshaw.bsky.social · 14d

TFW you catch your little toe on coffee table but can’t say the f word because there are kids around

4

Anthony Shaw @anthonypjshaw.bsky.social · 15d

I really liked ProGet for private enterprise feeds. There’s a free tier license.

1

Anthony Shaw @anthonypjshaw.bsky.social · 15d

That is exactly what I am doing 🚄

1

Anthony Shaw @anthonypjshaw.bsky.social · 15d

Yeh same. Smelt like a scam

1

Anthony Shaw @anthonypjshaw.bsky.social · 15d

Choo-Choo. Obvs at 300km/h you don’t need seatbelts 🤷🏻‍♂️

1 6

Anthony Shaw @anthonypjshaw.bsky.social · 19d

I’m guessing you haven’t seen either of the two horror movies which immediately come to mind 🦋

1

Anthony Shaw @anthonypjshaw.bsky.social · 19d

Solar export $ is practically $0 here so I’m looking at doubling the system generation size then adding a 24kWh battery. Have 9 months of data in Home Assistant to do some modelling. Shame there isn’t a nice API to fetch it into Pandas. 🐼

1 1

Anthony Shaw @anthonypjshaw.bsky.social · 22d

I didn’t, no. I was talked out pretty quickly this year. Will try and find it on YouTube

1

Anthony Shaw @anthonypjshaw.bsky.social · 22d

Finally got some VSCode pets stickers on my laptop

Don’t code alone stickers with the dog and cat from VSCode pets

2 6

Anthony Shaw @anthonypjshaw.bsky.social · 22d

Wow, #hacktoberfest spam is starting early this year.

1 3

Anthony Shaw @anthonypjshaw.bsky.social · 24d

"The majority of staff who voluntarily go into the office say they are happy to travel to the office; therefore, we extrapolate that the people who work from home would be happier in the office"
🫠

18

Anthony Shaw @anthonypjshaw.bsky.social · 25d

There’s a blog post by @pamelafox.bsky.social on this topic last month blog.pamelafox.org/2025/08/red-...

Red-teaming a RAG app: gpt-4o-mini v. llama3.1 v. hermes3

When we develop user-facing applications that are powered by LLMs, we're taking on a big risk that the LLM may produce output that is unsafe...

blog.pamelafox.org

1 1

Anthony Shaw @anthonypjshaw.bsky.social · 25d

SafetyBench and HarmBench are two popular ones.
Yes, smaller models tend to perform worse in those however some particular models where safety is not a priory perform poorly regardless of size. Most system prompts have safety guidelines and smaller models are more likely to ignore them.

1 1

Anthony Shaw @anthonypjshaw.bsky.social · 25d

Thank you. I ran out of time, but regarding your questions on safety and bias…
All model training data has biases. Some labs (not all) put models through various phases of tuning and testing to check model safety and likelihood to produce harmful answers. There are a few benchmarks on this 🧵

1 1

Anthony Shaw @anthonypjshaw.bsky.social · Sep 8

Someone logged into mastodon

1 1