Bruce Schneier:

At work, if your job is to move a bunch of heavy things from one side of the room to another, you should use whatever assistive tech you have on hand: a wagon, a forklift … even an AI-powered robot. But at the gym, it makes no sense for that robot to lift weights for you. The point of weightlifting isn’t to move heavy things across the room; it’s to actually lift those heavy things. The same analysis holds for any task an AI can do for you. If it’s work – if the task has to be done and no one cares how – then it’s fine to use AI assistance. But if the task is more like the gym, and how the task is done is at least as important, then it probably doesn’t make sense to use AI.

Ben Thompson:

What Karp is driving at is the flip-side of the points I am making above: the challenge for enterprises when it comes to using the frontier models is not simply that tokens are expensive, but that they are potentially giving valuable data to the labs who, in the long run, have an economic imperative to move up the stack and own the customer directly.

Chad Fowler:

Why are humans allowed to edit the output at all? If you care about provenance—about being able to trace every line of code back to a decision—then manual edits are a problem. They break the chain. They introduce code that has no record, no explanation, no origin you can inspect. So you end up in an uncomfortable place: You don’t just prefer that agents write the code. You require it. Not because agents are better programmers. But because they are the only way to guarantee that every change passes through a process that can be captured, inspected, and replayed. Humans don’t stop participating. They just move up a level. They define intent. They shape constraints. They review outcomes. But they don’t write the code directly anymore. Because the moment they do, the system forgets how it got there.

I keep my agent skills and plugins minimal. This set of software development agent skills hits a sweet spot.

Mitchell Hashimoto:

This is my journey of how I found value in AI tooling and what I’m trying next with it. In an ocean of overly dramatic, hyped takes, I hope this represents a more nuanced, measured approach to my views on AI and how they’ve changed over time.

Simon Willison:

It’s become very apparent over the past few months that a comprehensive test suite is enough to build a completely fresh implementation of any open source library from scratch, potentially in a different language.

This has worrying implications for open source projects with commercial business models. Here’s an example of a response: tldraw, the outstanding collaborative drawing library, are moving their test suite to a private repository.

Some practical advice on supporting AI agents from Diwank Tomer.

Two things that stood out to me:

1. Asking the AI to embed bits of context in comments for itself so it can reconstitute context on previous decisions.
2. Making the tests purely for humans as a way to give the AI more rope to make changes.

As Gen AI programming tools continue to develop, I’m wondering what things will look like when we remove the human from the loop.

At that point, all the code that’s generated is solely for the AI. All the human-focused concerns we care about in code disappear—it becomes a black box. The code effectively becomes another intermediate language for a new layer on the stack.

As long as the solution fulfils its requirements and fits within the constraints of security and cost for the necessary performance, we’re happy.

Code structures and data schemas don’t matter—so long as the AI can refactor them to meet new requirements as they emerge.

It reminds me of stories of when assembly programmers first saw these flash new C compilers arrive on the scene and generate all this assembly code that no human had directly written.

Casey Handmer:

A model I’ve long been interested in is the Corporation as a stand in for AGI. We need some non-human autonomous legal and economic entity. A corporation is just that. The Fortune 500 are already non-human super-intelligence. They operate 24/7/365 according to inscrutable internal logic, routinely execute feats of production unthinkable for any human or other biological organism, often outlive humans, can exist in multiple places at once, etc etc.

Simon Willison:

Using LLMs to write code is difficult and unintuitive. It takes significant effort to figure out the sharp and soft edges of using them in this way, and there’s precious little guidance to help people figure out how best to apply them.

If someone tells you that coding with LLMs is easy they are (probably unintentionally) misleading you. They may well have stumbled on to patterns that work, but those patterns do not come naturally to everyone.

I’ve been getting great results out of LLMs for code for over two years now. Here’s my attempt at transferring some of that experience and intuition to you.

Cal Paterson:

A general pattern seems to be that Artificial Intelligence is used when first doing some new thing. Then, once the value of doing that thing is established, society will find a way to provide the necessary data in a machine readable format, obviating (and improving on) the AI models.

Also check out Building LLMs is probably not going be a brilliant business.

iA Writer dims the text you paste from AI tools. As you edit ChatGPT’s input and make it your own, iA Writer keeps track of what is yours and what isn’t.

An interesting feature. They have a spec for the Markdown annotations that underpin it.

From a leaked internal Google document:

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months.

The latest version of Photoshop Beta now includes a feature called generative fill.

I can imagine it’s easy to compromise the authenticity in your photography if you overuse these kinds of tools. With that said, and I know it’s potentially a slippery slide, there are situations where they can be incredibly useful.

For instance, I sometimes want to adjust a crop and need to fill in some areas to maintain the balance of the composition, and my Photoshop pixel-surfing chops aren’t up to the task.

I had a crack at using generative fill on a photo I recently took that I wished had more foreground.

The results look usable.

I also tried a prompt that removed the shadow on the left but the results of that were less natural.

Experimenting with various prompts and browsing their outcomes is far more enjoyable than swearing at the healing and clone brush tools 😂.

The ChatGPT model is huge, but it’s not huge enough to retain every exact fact it’s encountered in its training set.

It can produce a convincing answer to anything, but that doesn’t mean it’s reflecting actual facts in its answers. You always have to stay skeptical and fact check what it tells you.

…

I like to think of language models like ChatGPT as a calculator for words.

This is reflected in their name: a “language model” implies that they are tools for working with language. That’s what they’ve been trained to do, and it’s language manipulation where they truly excel.

The post includes a list of language manipulations you can try.

Eugene Yan:

Rule #1: Don’t be afraid to launch a product without machine learning.

Machine learning is cool, but it requires data. Theoretically, you can take data from a different problem and then tweak the model for a new product, but this will likely underperform basic heuristics. If you think that machine learning will give you a 100% boost, then a heuristic will get you 50% of the way there.

The Case for Learned Index Structures:

Indexes are models: a B-Tree-Index can be seen as a model to map a key to the position of a record within a sorted array, a Hash-Index as a model to map a key to a position of a record within an unsorted array, and a BitMap-Index as a model to indicate if a data record exists or not. In this exploratory research paper, we start from this premise and posit that all existing index structures can be replaced with other types of models, including deep-learning models, which we term learned indexes.

…

Our initial results show, that by using neural nets we are able to outperform cache-optimized B-Trees by up to 70% in speed while saving an order-of-magnitude in memory over several real-world data sets.

I used to find it odd that these hypothetical AIs were supposed to be smart enough to solve problems that no human could, yet they were incapable of doing something most every adult has done: taking a step back and asking whether their current course of action is really a good idea. Then I realized that we are already surrounded by machines that demonstrate a complete lack of insight, we just call them corporations.