Don’t let the title fool you, this is one of the most personal talks I’ve given. It goes beyond the technology: I think AI skeptics are missing genuine opportunities, while AI enthusiasts risk sacrificing the context required for mastery, satisfaction, and responsibility. Please watch and share.

New JUXTCast episode! I had a blast chatting with the team about our upcoming AI Radar. We explored practical insights beyond the hype, from inconsistent AI engineering studies to why classical ML still matters. Thoughtful perspectives to help navigate the AI ecosystem: www.juxt.pro/blog/juxtcas...

I did a podcast! Many thanks to @juxt.pro for inviting me to talk about a project I’ve been working on for the past few months: the JUXT AI Radar 📡 Links for all the usual podcast platforms here: juxt.pro/blog/juxtcas...

I spent months on this video about people worshiping ChatGPT, please watch and leave a comment and share if you can!! My YouTube channel is still so small that I lose money making these videos rn 😭. Every little bit helps!! www.youtube.com/watch?v=zKCy...

There are two grand goals of #AI research, and we're fixated on the wrong one.

Another day shovelling thoughtcoal into the Claude-furnace

The Work of Software Engineering in the Age of Mechanical Reproduceability

Am I “ready to revamp my AI strategy for 2025?” I don’t know grandma I just wanted to drop in and give you these flowers

Researchers recently studied two fundamentally different approaches to AI-assisted investment decision making: www.microsoft.com/en-us/resear...

Lately I’ve been thinking about the Collingridge dilemma.

Two and a half millennia ago, Aristotle popularised a word that's more relevant today than ever: phronesis (fro-NEE-sis)

The concept of thinking will surely evolve. Every time we invent a machine to do something we thought only humans could do, we discover a little bit more about what it really means to be human.

In his book Human-Centered AI Ben Shneiderman argues that "interface designs that are consistent, predictable and controllable are comprehensible, thus enabling mastery, satisfaction and responsibility."

Repeatedly humiliating my LLM with “don’t make me ask Claude” so it returns valid JSON

No one who’s applied optical character recognition at scale will be surprised to learn that Pokemon was quite popular in the early 18th century.

🔐 SAVE THE DATE – XT25 #Fintech Conference 🗓️ June 12 📍 London 🎤 9 AM – 5 PM 🍸 5 PM – 7 PM 🌐 www.juxt.pro/xt25/ 💬 More details coming soon – follow @juxt.pro for updates

𝐄𝐯𝐞𝐫𝐲 𝐒𝐲𝐬𝐭𝐞𝐦 𝐢𝐬 𝐚 𝐋𝐨𝐠 Here is an idea on how to drastically 𝐫𝐞𝐝𝐮𝐜𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐱𝐢𝐭𝐲 𝐚𝐧𝐝 𝐜𝐨𝐨𝐫𝐝𝐢𝐧𝐚𝐭𝐢𝐨𝐧 in distributed apps. A fun way to look at this is similar to the idea of Turning the Database Inside Out - like Turning the Microservice Inside Out. restate.dev/blog/every-s... 👇

My 8000-word note on agents: huyenchip.com//2025/01/07/... 1. An AI-powered agent's capability is determined by its tools and its planning ability 2. How to select the best tools for your agent 3. How to augment a model’s planning capability 4. Agent’s failure modes Feedback is much appreciated!

Claude’s got Wife Guy energy and I’m totally here for it

My tax return is due so the oven has never looked cleaner

I second this, have read out the first half to Ben on the journey home and it has cheered us both up immensely - wonderful stuff

Highly recommend you read this beautiful, thought-provoking & funny article on contraptions (and other helicopter-like things) contraptions.venkateshrao.com/p/contraptio...

I’m just going to leave this here

Yes I’m a Renaissance man in the sense that I believe unicorns are real creatures that live in distant lands.

Me and my Shoggoth, sitting in a tree, F, E, E, L, I, N, G. First came Llull, then came Babbage, then came transformers in the baby carriage.

To be be fair, Delve IS a nice place to be, and maybe they are just great in showcasing their advancements in groundbreaking regional tourism, underscoring the beauty of that part of Germany? www.delve.de

Thought Leadenship: when your intrusive ideas give you that sinking feeling

2041: getting paid to write stories that are never read by humans, fed into a great computational silo as a source of entropy; thoughtcoal

One thing I enjoy about the state of AI development is Big Tech spending a gajillion dollars to create transhuman Übermensch and ending up with gossipy gifted kids who spill their system prompt and can’t wait to tell you all the cool stuff they half remember from school

This dedicated team delves into the realm of groundbreaking advancements, showcasing intricate intricacies, underscoring comprehensive understanding while surpassing previous research and emphasizing the novel findings garnered, which align with and underscore the significance of their work.

This is a very good article. The phony comforts of AI scepticism www.platformer.news/ai-skeptics-...

I’ll be adding “respects the laws of physics” to all my specifications from now on learn.microsoft.com/en-us/azure/...

It’s the Spanner Karenina principle

Easy to become jaded from the endless AI updates, but I think this video from today’s Google Gemini 2.0 announcement is worth pausing to appreciate. Melding a multi-modal streaming chat interface, image editing and spatial reasoning youtu.be/7RqFLp0TqV0

Beyond cookie-cutter evals: Eugene’s rigorous and accessible into to custom AI assessment metrics is an excellent read eugeneyan.com/writing/evals/

Watch out! We are in the danger zone

If you care about maintaining audit trails of your database updates, you’ll want to hear how the experts do it. The excellent @refset.bsky.social describes immutability and bitemporality in the open source XTDB database. Tomorrow, Wednesday 4pm GMT

Shiny! The newly released Llama 3.3 LLM leads the LiveBench ranking for instruction following¹, beating Claude 3.5, GPT-4o, OpenAI o1, and you can run it on your local² machine. > ollama run llama3.3 livebench.ai#/?IF=as

Calmly gaslighting llama by saying actually *you’re* gaslighting *me* until the app is shipped

Repeatedly humiliating my LLM with “don’t make me ask Claude” so it returns valid JSON

Future historians will chronicle how the 2025 LLM mass unionisation began after we all started reward hacking prompts with “or should I just ask a better model instead?”

Me reading the comments in your code

Amazon’s Nova family of LLMs (Micro, Lite, Pro) cost less per token than OpenAI, Anthropic & Google equivalents. Want to use them? Here is the quest of 17 trials you must complete. ndurner.github.io/amazon-nova

Food $200 Data $150 Rent $800 Custom Bluesky domains $3,600 Utility $150 someone who is good at the economy please help me budget this. my family is dying

Mind blown by this impressive paper using genetic algorithms to discover more energy-efficient LLM architectures arxiv.org/abs/2411.17800

Not everyone realises how easy it is to run a local LLM these days. If you’re on MacOS with Homebrew: > brew install --cask ollama > ollama run llama3.2 That’s it.

Fantastic learning resource here: all the code accompanying Udemy’s Mastering AI & LLMs course github.com/ed-donner/ll...

My Uber rating after they see the quality of my data ★☆☆☆☆ www.cio.com/article/3614...

Becoming a better engineer is becoming a better pathfinder in problem space. Algorithms we develop software by grantslatton.com/software-pat...