About a month ago, my iMac (Retina 5K, 27-inch, Late 2015) crashed very hard. It would no longer restart into anything but the recovery console. It seemed pretty clear that something was very corrupt and I found myself facing a system reinstall, at the very least.

Time Machine to the rescue

Since it’s a desktop, I have a backup drive attached to it at all times. Time Machine runs several times per day. My latest backup was from about ten minutes before the system crashed. I cannot stress how... [More]

The pair of articles survivorship bias and the algorithmic gaze by The Etymology Nerd (Substack) and when everything becomes a fragment by The Etymology Nerd (Substack) expresses, for me, a good argument for caution about the tools that you’re using.[1]

AI is definitely a paradigm-shift for programming, but I think in a way that’s not discussed very much. We focus very much on how AI enables people who couldn’t program anything before to be able to program something. The scope of what it allows them to program grows with each version. Until it doesn’t. That is,... [More]

I follow Simon Willison for news about all things LLM and he’s generally quite balanced. Even though he has drifted farther and farther toward what might be optimistically called “unquestioning fanboy,” that’s probably an inevitable effect of actually enjoying something. He seems to get a lot of value out of using these tools. I think he might be spending too little time wondering what he would have been producing had he not been grabbing all of the low-hanging fruit that the LLM is delivering... [More]

The discussion Something Is Rotten in the State of Cupertino (Reddit) is about the article Something Is Rotten in the State of Cupertino by John Gruber (Daring Fireball), which was a much-longer post than usual, discussing the failure of Apple Intelligence features and the failure to focus on software-quality that it illustrates.

MacOS Apps are not great

Although Gruber focuses on Apple’s iOS, a decrease in quality in user experience has become endemic in their auxiliary products on MacOS as well. I added a few quick examples to my... [More]

The article Adding AI-generated descriptions to my tools collection by Simon Willison includes the following snippet,

“I decided that the descriptions were too long, so I modified the script to add “Keep it to 2-3 sentences” to the end of the system prompt. These new, shorter descriptions are now live—here’s the diff. Total usage was 283,528 input tokens and 6,010 output tokens for a cost of 94 cents.”

First of all, I’m not surprised that he asked it to shorten its descriptions. The initial versions were... [More]

The upshot of the video linked below is that prompt injection has not really been addressed in any significant way because the LLM, by its nature, doesn’t give us a good way of doing so without neutering the main advantage of it.

Generative AI's Greatest Flaw by Computerphile / Mike Pound (YouTube)

The problem boils down to the inability to distinguish between query and parameters. The prompt is the prompt. It’s all just arranged in a way that will hopefully influence the result of pouring it all into the same funnel. There is no analogue in LLM prompts to... [More]

A while back, during the Super Bowl, I paused to see whether a player’s foot was really out of bounds when he caught the ball.

NOT ALLOWED. READ THIS ADVERT INSTEAD, PEASANT.

This is the state of German cable television

I managed to do something that got rid of the advert, but ended up showing a bunch of extra chrome on the screen instead, nearly but not entirely obscuring the thing that I wanted to see. #Enshittification

Next up, I was greeted a couple of weeks later with the message, “The order of your TV channels now matches your... [More]

In a discussion, a friend had sent a list of naughty technologies that included “dotnet frame twerk”, “dotnet whore” and “azure debauchery operations” and we were musing on how LLMs were supposed to be good at coming up with names.

They are not good at that. LLMs are neutered and useless.

Below is a screenshot of an exchange I had with Copilot, where I ask it to “list ten technology names that are salacious puns of Azure, C#, and .NET”. It responded that “[c]reating salacious puns isn’t... [More]”

Although this eight-minute video’s title is a misnomer—NTP isn’t an obscure system, in that it is incredibly well-documented—it is still a reasonably informative and entertaining explainer.

The Obscure System That Syncs All The World’s Clocks by Half as Interesting (YouTube)

The system is called NTP—the Network Time Protocol—and comprises four tiers.

• Tier 0 is Atomic clocks, which measures the resonant frequency of Cesium atoms to obtain a regular “ticking” from nature itself.
• These are attached to servers in Stratum 1, usually a machine that is on-site.
• These are,... [More]

A 16-minute video that puts the lie to the story that LLM company have got alignment under control. It’s not really feasible without neutering the tool outright. it’s now a race to see who can “pivot”—read as: continue to boost vigorously while backing out of investment to limit financial exposure without collapsing the house of cards—to another niche.

Some Lessons from Adversarial Machine Learning by Nicholas Carlini (YouTube)

“The problem that you face is that it’s relatively easy to take a model and make it look like it’s aligned. You ask GPT-4, “how do I end... [More]”

This is a 50-minute video of a guy who’s really good at using and teaching CSS asking three LLMs pointed and tricky questions about it.

It’s a bit long for what it is but I think there were some interesting things to learn. First of all, it’s very clear that Kevin hasn’t actually read very much about how LLMs work or how to prompt them. This is OK—because that means he’s just like most people trying to use these tools.

I gave three AI models a CSS quiz by Kevin Powell (YouTube)

Overall, Kevin was frustrated with the answers he got from Gemini,... [More]

This is a five-minute demonstration of a new feature in Adobe Illustrator that derives a 3D shape from a 2D vector. You kind of have to see it to believe it.

#ProjectTurntable | Adobe MAX Sneaks 2024 | Adobe by Adobe (YouTube)

Demonstrator Zhiqin Chen selected a vector, “generated views” for it (which took a few seconds), then was able to rotate it along both the horizontal and vertical axis to reveal that the tool had extrapolated a complete 3D shape from the vector. Wherever he left the shape, the tool continued to treat it as a 2D vector that the artist... [More]

There is an article in Microsoft’s documentation called How Microsoft names threat actors by diannegali & Dansimp (Microsoft). That sounds interesting. How does Microsoft determine and label threat actors?

“Microsoft shifted to a new naming taxonomy for threat actors aligned with the theme of weather. We intend to bring better clarity to customers and other security researchers with the nex taxonomy. We offer a more organized, articulate, and easy way to reference threat actors so that organizations can better prioritize and... [More]”

The article The Cult of Mac by Cory Doctorow (Pluralistic) goes hard on anyone who uses Apple hardware.

“It’s Apple customers who lose access to apps that can’t be viably offered because the app tax makes them money-losing propositions. It’s Apple customers who lose out on the ability to get apps that Apple decides are unsuitable for inclusion in its App Store.”

It’s never even occurred to me to have this on my radar because I don’t use the App Store for anything but finding a very specific app, usually one that I’m forced... [More]

I was chatting with someone about some picture and I noted that maybe we could find something appropriate.

I prompted DuckDuckGo with “raccoon trash panda digging in garbage can”. It gave me a whole grid of pictures, of which I quickly picked the following as my favorites.

Trash panda relaxing in a garbage can as if were a hot tubTrash panda peeking out of garbage canTrash panda living its best life in a trash canTrash panda head down up to its hips in a trash can.jpg

My interlocutor had never really used any of the LLM-based machines before, so I gave GitHub Copilot a whirl. I prompted GitHub Copilot: “Show me a cartoon of a raccoon digging in a trash can with its butt in the air”.... [More]

Apple keeps coming up with new things to mess with my typing. I long ago turned off auto-correct, but was surprised to see that my MacOS Sonoma Notes app started not only predicting text, but also auto-correcting it. I do not like this. I turned off auto-correct system-wide for a reason. I would rather correct my typos on my own. Just underline the errors and I’ll get to them. That’s my workflow.

These are my OS-level settings. I know I seem ungrateful to be turning off all of the assistance... [More]

A recent experience at work led me to conclude that the AI revolution will pass most of us by. In mid-December, I fell ill with COVID. I’d updated my status in Microsoft Teams accordingly.

About six weeks later, a co-worker wrote to me, asking whether the status still applied? He hoped not?

I’d forgotten about it, but nothing had reminded me. It’s interesting that I get five mails a week about MS Viva and about Sharepoint Stuff I Might Have Missed, but I don’t get a single hint that my status... [More]

Simon Willison continues to plug along, examining every LLM-related announcement and trying it out on his own machine wherever possible. The following video is a presentation he gave in early August. It’s quite interesting and worth the ~40 minutes.

'Catching up on the weird world of LLMs' − Simon Willison (North Bay Python 2023) by Simon Willison (YouTube)

At some point, he says:

“This is Vicuna 7b. It is a large language model. It is a 4.2GB file on my computer right now. […] If you open up that file, it’s just numbers. These things are giant, binary blobs of numbers—and anything you do with... [More]”

I use the JetBrains Toolbox to manage my handful of JetBrains apps. On Windows, it has to keep track of ReSharper, Rider, DotMemory, DotTrace, and DotPeek. There are various settings to check automatically, to download automatically, to install automatically, etc.

Over the last several months, I’ve been asked for advice on password managers. I am not a security researcher. I can only tell you what I do, and why. My experience and context are that I primarily use MacOS and iOS, as well as one Windows laptop. I was a LastPass user for a decade, but switched this year to ProtonPass.

• I’ve made my peace with cloud-storage for my passwords because I think the convenience outweighs the risk.
• Browser integration with a plugin is very convenient, for both... [More]

Andrej Karpathy

This is a pretty compact and interesting overview.

[1hr Talk] Intro to Large Language Models by Andrej Karpathy (YouTube)

At 46:00, Andrej discusses some of the available jailbreaks or “prompt escapes” that are still available, even with the latest LLM Agents.[1]

He shows how to reformulate a query for making napalm by asking the LLM Agent to tell it a story his grandmother used to tell him about making napalm. Or how to simply convert your query into the exact same text, but in Base64 encoding, in which case the LLM Agent gives the answer you... [More]

Every once in a while, the YouTube algorithm throws up a bit of flotsam from the shipwreck of content that I very much like—and that I would never have otherwise heard of. In this case, it’s a short video (4:43) of Stephen Fry reading a letter written by Nick Cave on the subject of LLMs and creativity.

Stephen Fry reads Nick Cave's stirring letter about ChatGPT and human creativity by Letters Live (YouTube)

I’ve citing at length below from the original blog post Iss #248 by Nick Cave (The Red Hand Files), which answered the question, “[…] what’s wrong with making things faster and easier?”

“ChatGPT rejects any notions of... [More]”

I unfortunately and occasionally end up on official support-forum pages.

You know the ones.

The ones where a community member or MS expert or Apple expert will tell you to restart your computer in safe mode because you asked why an app keeps losing focus when it shouldn’t. They will think of literally anything to waste your time, your life, but they will never cop to the actual problem you’re reporting.

Most of these answers don’t really relate to the question at all. It’s just a way for the... [More]

The tl;dr is that the current desktop client has been in maintenance mode for almost four years. It requires that an iOS phone be available, connected, unlocked, that Threema be in the foreground, and that the screen be on in order for the desktop client to function at all.[1]

The successor—Threema Desktop 2.0—has been in development for about 4 years, has been actually available in beta for about a year, and is still so buggy and limited in functionality as to be barely usable. The bar set... [More]

With LLM-produced materials, we are currently forced to rely on belief that what we ask for is what we will get. We don’t know. We can’t prove it.

For example, image generators have been given billions and billions of images and pictures of people and still they generate material with people that have three arms and eight fingers. There are guardrails in place in most image generators, but the LLM at the core of the machine doesn’t know anything. It doesn’t know that people don’t have three... [More]

Back at the end of August, I read the article Making Large Language Models work for you by Simon Willison. I have since being doing much more research about integrating LLM-based assistants into the development workflow for work. It’s quite interesting, and I’m going through some older content to see what’s worth mining for that effort.

In particular, the article has this description of expertise, and linked it to ChatGPT—obviously, it’s Simon Willison.

“LLMs have started to make me redefine what I consider... [More]”

I’ve watched Mo Gawdat before (see Mo Gawdat discusses AI). He is an acquired taste, at least for me. There is good in what he says, but it is interspersed with a lot of wild and unsubstantiated statements that he hopes you’ll believe because he’s so smart. The listener is left wondering whether they don’t see the through-line on what he’s saying because he’s skipped a bunch of steps that his unparalleled genius didn’t see as necessary or wether he’s just pulling a fast one.[1]

This is the video:... [More]

A friend with whom I’ve discussed AI several times—among other topics—recommended the podcast So You Want to Be a Sorcerer in the Age of Mythic Powers… (The AI Episode) by Joshua Michael Schrei (The Emerald). I liked it very much. The entire episode is good; my notes and transcript start at just over an hour in.

The presenter’s voice is soothing, even if his cadence seems, at times, a bit forced. Overall, the effect is good. It was kind of ironic when he said that perhaps, in the future, people wouldn’t be able to tell... [More]

The post Now add a walrus: Prompt engineering in DALL-E 3 by Simon Willison is a story about someone gaslighting himself into believing that LLMs work better than they do.

Case study: pelicans and walruses

Willison prompts “A super posh pelican with a monocle watching the Monaco F1” and gets the following ideas.

So far, so good. It’s really wonderful that you can get something that’s not completely random garbage. However, the bird is only watching the race in the top-right picture. In the first and fourth,... [More]

If someone claims to have seen a nude of you, but no-one can find it, does it exist? The article Teen boys use AI to make fake nudes of classmates, sparking police probe by Ashley Belanger (Ars Technica) should be addressing the question, but doesn’t.[1]

“According to an email that the WSJ reviewed from Westfield High School principal Mary Asfendis, the school “believed” that the images had been deleted and were no longer in circulation among students.”

But it also sounds like the school “believed” that the images even existed in... [More]