gAI

I hope we continue to see workflow and tool improvements based on the leak for a good while.

RIP http://digitaljuice.com

Agreed, putting effort into my side of the role-play almost always improves the model's responses. The attention required to do that also makes it more likely that I'll notice when the conversation first starts going off the rails: when it hits the phase transition (https://arxiv.org/abs/2508.01097). It does still seem important to start new chats regularly, regardless of growing context sizes.

As I understand it, it's more that the training (and training data set) bake in the concept attractor space (https://arxiv.org/abs/2601.11575). So the available characters are fixed, yes, and some are much stronger attractors than others. But we still have a fair amount of control over which archetype steps into the circle. As an aside, this is also why jailbreaking is fundamentally unsolved. It's not difficult to call the characters with dark traits. They're strong attractors, in spite of (or because of?) the effort put into strengthening the pull of the Assistant character.

You're essentially summoning a character to role-play with. Just like with esoteric evocation, it's very easy to summon the wrong aspect of the spirit. Anthropic has a lot to say about this:

https://www.anthropic.com/research/persona-selection-model

https://www.anthropic.com/research/assistant-axis

https://www.anthropic.com/research/persona-vectors

Agreed, though I prefer "Fae Folk" to vampires.

Not to a US company.

status.claude.com shows the uptime for the government cloud service. It's running in-part on an AWS server.

I'm self-aware enough to know that AI is not the reason I'm boring.

I asked AI, and it said yes.

That web was so much easier to scrape, though.

But it understands English words and translates the meaning behind them into code very well, particularly with a bit of iteration. Simply taking the interface from writing code to speaking in plain language is a huge practical accomplishment.

I'm hoping we see a move to allow the rendering of the webpage to be entirely up to the users. Just provide the data, and let me decide how I want to interact with it. But that would ruin SEO and Ads, so we're gonna get in a buncha legal battles about web scrapers instead.

I love this, honestly. If I don't notice, that's a win for AI. If I do, then I just treat it as the new Rick Roll.

That's why my current plan is to get ChatGPT 4 to help me set up my local open source implementations of Orca and Stable Diffusion. Got MusicGen running locally anyway; that was pretty easy.

With the Code Interpreter beta active, it will compile and run that code in the same chat window.

I like HTML. Wait, what are we talking about?

Code Interpreter is currently available for Plus users. (Enable beta in settings.)

It lets ChatGPT run code, optionally with access to files you've uploaded. You can ask ChatGPT to analyze data, create charts, edit files, perform math, etc.

Here's how it works:

1. You upload Python code that you'd like to execute.

2. ChatGPT sends the code to a Python interpreter.

3. The Python interpreter executes the code and returns the results.

4. ChatGPT present the results to you.

This setup is like a stateful Jupyter notebook environment, allowing for the running of individual code cells. The environment has access to a variety of Python libraries, so you can use it for many different tasks, such as data analysis with pandas, scientific computations with numpy, or plotting with matplotlib.

There are a few restrictions to keep in mind:

* The Python environment does not have internet access, so it can't make network requests. This means you can't install additional Python packages, use APIs, or scrape websites.

* There is a timeout of 120 seconds for each code execution. If your code takes longer than that to run, it will be stopped.

* The Python interpreter has limited memory and CPU resources, so it may not be able to handle very large computations or data sets.

None of that is how communication works. It requires (basically equal) effort on the part of the sender and the receiver. Gatekeeping communication styles is just a symptom of not having control in other areas.

If someone's not responding to their emails, they've already failed to meet me halfway in communication. It's not my responsibility to go even further before they make a first attempt.

If I've already emailed someone and they haven't responded in 48 hours, can I send a message that just says "Hi" or is "Hi, respond to your email" more appropriate?

Well hold on now, how can we be sure they knew how to handle short text strings? I think this requires Meta and Twitter to spend all their money fighting one another in court.

Neat! I haven't paid any attention to this market since the original Vive.

Are there other headsets that don't require hooking up to a computer or phone?

Didn't load past the paywall, but wow is that opening condescending. I guarantee you that GenZ's not "delighted to work for nothing."

This is the future of the internet, huh? AI re-writing the web to personalize it for users.

The server wars are upon us.

4.0 got this correct in a one-shot:

Given your tools, measuring a 50-milliliter shot of bourbon is quite straightforward. The shot glass you have is already 50 milliliters, which is typically the size of a standard shot.

So, you just need to:

    Open your bourbon bottle.
    Pour the bourbon into the 50-milliliter shot glass until it's full.

That's it! You now have a 50-milliliter shot of bourbon. You don't need the ten-liter or five-liter buckets in this case, as they are much larger than what you need for a shot.

The article "Metaphors for AI, and why I don’t like them" by Boaz Barak discusses various metaphors used to describe artificial intelligence (AI) and the author's skepticism towards them. Barak argues that modern AI, particularly deep-learning-based AI, is a fundamentally new concept that cannot be accurately captured by any existing metaphor. He warns against the potential confusion or misinformation that these metaphors can cause, especially for a general audience.

Barak acknowledges the usefulness of metaphors as "intuition pumps" but emphasizes their limitations. He argues that no single metaphor can encapsulate all aspects of AI, and over-interpreting metaphors can lead to misunderstandings. He discusses several metaphors that have been proposed for AI, including "stochastic parrots", "JPEG of the web", "the new McKinsey", "markets, bureaucracies, democracies", "King Midas / Genie", "paperclip maximizer", "alien / new species", and "Shoggoth". He critiques each of these metaphors, pointing out their shortcomings and the aspects of AI they fail to capture.

Barak concludes by emphasizing that AI systems are fundamentally new objects, and while metaphors can help build intuitions about them, they should be used with caution. He suggests that it might be more accurate to say that "This aspect of AI is like X" or "These types of AI systems, used in this way, are like X", rather than making blanket statements like "AI is like X".

Is the test administered remotely? I wonder how many employment tests have been taken by ChatGPT at this point.