Llm

Despite what tech CEOs might say, large language models are not smart in any recognizably human sense of the word.

Since yesterday, Claude Code has been included in the Pro subscription we’re signed up for. I’d been wanting to try it for a while, and now nothing was stopping me. Only yesterday (a curious coincidence), I read AI Changes Everything by Armin Ronacher, which gave me a glimpse of the potential and made my hands itch to try it. The initialization of Claude on the repository surprised me; the analysis (reported in CLAUDE.md) is thorough and reveals a good understanding of the project, including both the code and the general functioning, strategies, techniques, technologies and libraries employed. ...

Today’s Armin Ronacher’s AI Changes Everything strongly resonates with me1. I may not be using Claude Code as a daily driver as he now does, but I’ve slowly and steadily introduced large language models (LLMs) into my routine, and I’m reaping the benefits. It wasn’t the purpose of his article, but I wish Armin had gone into the details of how, why, and when he delegates tasks to Claude Code. My AI Skeptic Friends All Nuts, linked in Armin’s piece, is also worth reading. ↩︎ ...

Run Your Own AI by Anthony Lewis is a concise tutorial on how to run large language models on your laptop from the command line via llm-mlx. It focuses on Macs M-series, but it’s also suitable for other hardware. Saving it here for a friend.

It’s raining a lot these days, and during my early morning walk, I see many earthworms popping out of the ground and filling the sidewalks and the streets. I observed this behaviour many times in similarly wet conditions. So I asked Claude: Why do earthworms appear on wet roads? When it rains heavily, streets fill with earthworms for several reasons: Earthworms breathe through their skin, which must remain moist. During heavy rainfall, their underground tunnels flood, forcing them to surface to avoid drowning. ...

I just finished reading Paul Auster’s The New York Trilogy1. On this occasion, I discovered a new use for ChatGPT and LLMs. ChatGPT and I chatted about the themes, especially the correlations and connections between the three short novels that comprise the volume. It was an alienating and revealing experience. For the first time, I am reasoning about a book with a machine, not a person. Because it knows everything about the text and draws on the shared global knowledge, it can give more satisfaction than most people do (also, it’s not easy to find someone around with whom I can talk about all the books I read!) Yes, it is wordy and repetitive, but it can stimulate and enrich my analysis2. ...

Charity Majors1 has a good, long-form article on the Stack Overflow blog. The title is misleading as, while AI’s impact on software engineering and its hiring process (spoiler: you’ll still want to hire junior engineers) is at the heart of the article, there’s so much more in it. It gets exciting in the second part, where she dispenses much from-the-trenches advice on team management and building. Hiring engineers is about composing teams. The smallest unit of software ownership is not the individual, it’s the team. ...

Simon Willison has a new article explaining an important and often ununderstood aspect of LLMs. There’s a remarkable difference between chatting with an LLM, as we users do, and training it. Short version: ChatGPT and other similar tools do not directly learn from and memorize everything that you say to them. Every time you start a new chat conversation, you clear the slate. Each conversation is an entirely new sequence, carried out entirely independently of previous conversations from both yourself and other users. Understanding this is key to working effectively with these models. Every time you hit “new chat” you are effectively wiping the short-term memory of the model, starting again from scratch. This has a number of important consequences. ...

Open AI just released ChatGPT 4o. The launch demo is available on YouTube, and yes, it is impressive. They did not launch v5, though, and 4o is only incremental, not exponential, as v4 has been compared to its predecessor. It may mean we’re at the end of the “exponential growth” phase of LLM models. However, the most critical aspect of this release is not technical, as Ethan Mollick correctly pinpoints in his timely What Open AI Did post: ...

It’s very fast to build something that’s 90% of a solution. The problem is that the last 10% of building something is usually the hard part which really matters, and with a black box at the center of the product, it feels much more difficult to me to nail that remaining 10%. Closing that gap with gen AI feels much more fickle to me than a normal engineering problem. It could be that I’m unfamiliar with it, but I also wonder if some classes of generative AI based products are just doomed to mediocrity as a result. ...