Which Nested Data Format Do LLMs Understand Best? JSON vs. YAML vs. XML vs. Markdown We tested LLMs from three different providers on nested data in JSON, YAML, XML, and Markdown formats. Two models performed best with YAML, while Markdown was the most token-efficient format.

Which Nested Data Format Do LLMs Understand Best? JSON vs. YAML vs. XML vs. Markdown www.improvingagents.com/blog/best-ne...

Reclaim The Em Dash Join the movement to reclaim the em dash from AI stigma.

www.reclaimtheemdash.org

Testing JSON, CSV, YAML and More: The Best Table Input Format for LLMs We benchmarked 11 data formats, including JSON, CSV, YAML, XML, and Markdown, to see which LLMs understand best. Results reveal surprising accuracy and token cost trade-offs.

How should you format tables of data you're passing to an LLM? www.improvingagents.com/blog/best-in...

AI assistance can significantly _improve_ the quality and security of the software we develop; it just depends how we use it. (And, over time, the tools are going to have more good practices built in.)

I agree it's not getting the attention it deserves.

Perhaps it'll take someone exploiting this at scale (hopefully in a fairly harmless way) to jolt us to our senses a bit.

Sebastian Siemiatkowski on X: "Yes, we did shut down Salesforce a year ago, as we have many SaaS providers—an internal estimate is about 1,200 SaaS shut down. No, I don't think it is the end of Salesforce; might be the opposite. Here is what actually happened and how/why we originally intended to NOT share" / X Yes, we did shut down Salesforce a year ago, as we have many SaaS providers—an internal estimate is about 1,200 SaaS shut down. No, I don't think it is the end of Salesforce; might be the opposite. Here is what actually happened and how/why we originally intended to NOT share

Some fascinating thoughts from Klarna's CEO here about their approach to enterprise IT.

Is AI tipping the balance away from best-in-class domain-specific SaaS towards 'all-in-one' SaaS?

And does 'opinionated' SaaS encourage good processes or shoehorn orgs into clumsy ones?

x.com/klarnaseb/st...

Mission-Critical Evals at Scale (Learnings from 100k medical decisions) YouTube video by AI Engineer

And for discussion of more sophisticated, at-scale stuff, I thought this talk was good: www.youtube.com/watch?v=cZ5Z...

I think a helpful early step is to put in place a way to run some very simple automated evals. That lowers the 'activation energy' to add more.

Like traditional tests, it takes discipline to add and maintain these evals but without them you don't know what you're breaking / making worse.

A robot software developer

How should software product development teams be using AI today?

What’s now a no-brainer?

And what’s not worth it? (Or too dangerous?)

My current thoughts: www.mattcollins.net/2025/01/how-...

If it wasn't open source, maybe another IDE's ecosystem would be dominant which would be worse for them.

GitHub - openai/openai-realtime-agents: This is a simple demonstration of more advanced, agentic patterns built on top of the Realtime API. This is a simple demonstration of more advanced, agentic patterns built on top of the Realtime API. - openai/openai-realtime-agents

Repo here: github.com/openai/opena...

How to Run and Customize Real-Time Voice Agents with OpenAI's Demo Code YouTube video by Matt Collins

OpenAI have published a handy demo repo showing how you can build advanced realtime voice agents using their API.

If you're a developer interested in experimenting with this stuff it's a nice, quick way to get started. (Fun, too!)

www.youtube.com/watch?v=gqaY...

GitHub - huggingface/smolagents: 🤗 smolagents: a barebones library for agents. Agents write python code to call tools and orchestrate other agents. 🤗 smolagents: a barebones library for agents. Agents write python code to call tools and orchestrate other agents. - huggingface/smolagents

If you're building AI agents, the smolagents library from Hugging Face (released a couple of weeks ago) is one to be aware of. It's intriguing to see that they've chosen to emphasise agents that write their actions in code (CodeAct-style) rather than as JSON-like snippets. github.com/huggingface/...

😺 New year, new AI? PLUS: Google's 9-hour prompt course in ~20 minutes (or less)...

My little writing productivity tool, Flowdrafter, gets a mention in today's edition of The Neuron (which apparently goes to 500,000+ people) as one of their 'Treats To Try'. 🙂 www.theneurondaily.com/p/new-year-n...

Based on an idea from Claude (AI), I created a little writing tool in a few hours using V0 (AI) and Cursor (AI). Rather surprisingly, it's currently #1 on Product Hunt! www.producthunt.com/posts/flowdr...

Based on seeing lots of companies, 98% of what people are calling AI agents are not what the AI labs would call agents. They are usually structured document retrieval systems with a prompt or two for summaries, there is very little control or decision-making given to the AI, very little AI planning

Best of luck with the new venture!

Wow, sorry to hear that - sounds rough. I'm glad to hear a bit of coding is helping. We live in good times for tinkering!

If you want to understand a manual process in a deeper way than you ever thought possible, try automating it. 😅 #AI

Current vacancy: Chief Technology Officer – The Labour Party Salary: Competitive Location: Head Office – London Duration: Permanent The Labour Party is looking to recruit a Chief Technology Officer. The post-holder will be responsible for developing and impleme...

An interesting CTO role for someone. labour.org.uk/about-us/wor...

6 AI tools I've been impressed by lately:

www.blinkshot.io - generate images in realtime
www.napkin.ai - get visuals from your text
www.flowvoice.ai - Mac app for fast voice input
notebooklm.google - docs to podcast episodes
chatgpt.com - advanced voice mode
v0.dev - generate web UI

How about you?