OP here. It took me an afternoon to try different methods and test the limits. But now we know how it works it’s very fast to create new ones:

1. Prompt to make SVG - review in browser, iterate.

2. Prompt to write image prompt - review in editor, refine

3. Send to Gemini, get image

So maybe 5-10 mins.

I don’t know how to use Blender.

Also this method can be done over WhatsApp/telegram which is another plus over Blender type approach.

I found a simple technique to get reliable text and numbers in AI generated images.

I’m surprised the image models aren’t already doing this, so wanted to share since I’m finding this so useful

In some ways, this is similar to use of a Control Net. I've been doing this same technique for a while, using only SVGs as the base image. Works well.

Isn’t this sort of just “chain of thought” (i.e. the seminal https://arxiv.org/abs/2201.11903 ) where the user is helping the model 1-shot or k-shot the solution instead of 0-shot? I’ve used a similar technique to great effect. I feel things are so new / moving so fast that it’s hard to have common lingo. So very helpful to have a blog / example! But I wonder if the phenomena has been seen / understood before and just in smaller circles / different name.

Very impressive, simple, and reliable. I'm sure it will be picked up by image generation labs soon.

TLDR: use SVG to outline image correctly first, then send that image with your text prompt to get Gemini 3.0 Pro to render with correct numbers and text

Robots struggle with syntax-in-syntax. Really easy to confuse them when asking it to write a SQL query that targets a JSON column but it must respond with a JSON envelope so the harness can parse the result. Lots of escaping that needs to happen. Deeply nested structures in JSON also end up with foibles like missing a ] or } in a string of }}]}]}. Aside from the prompt injection possibility, just the result being straight up broken and requiring another LLM call is tokens flushed.

It doesn't work. You can't trust LLMs to 100% reliably obey delimiters or structure in content. That's why prompt injection is a problem in the first place.

Well-established where and amongst who, exactly? Is it seriously a common belief that this prevents prompt injection?

That would be more than a little alarming.

Yep, I suspect this is the rationale/driver too.

The business driver I assume is not lock-in on UX (as some say here) but the additional signal Anthropic gets when using their harness vs a 3P one. It makes sense to discount the price if that signal helps you improve your models, but that discount makes no sense if the user is running your model in another harness and you just get regular API usage signal.

Nice! I made a similar thing but the html for the text editor fits in a data uri, so it can be a bookmark or new tab page for taking quick notes

https://gist.github.com/smcllns/8b727361ce4cf55cbc017faaefbb...

Re 19, I made this with an iOS Shortcut a few weeks ago

  > A minimal voice assistant for my Apple Watch. I have lots of questions that are too complicated for Siri but not for ChatGPT. The responses should just be a few words long.

Use Dictate Text action to take voice as input, pass the text to OpenAI API as the user message with this as the system prompt:

“CRITICAL: Your response will only be shown in an iOS push notification or on a watch screen, so answer concisely in <150 characters. Do not use markdown formatting - responses are rendered as plain text. Do use minimalist, stylish yet effective vocabulary and punctuation.

CRITICAL: The user can not respond so do not ask a question back. Answer the prompt in one shot and if necessary, declare assumptions about the users questions so you could answer it in one shot, while making it possible for the user user to repeat ask with more clarity if your assumptions were not right.”

It works well. The biggest annoyance is it takes about 5-20s to return a response, though I love that it’s nearly instantaneous to send my question (don’t need to wait for any apps to open etc)

YMMV but a few things that helped me debug myself:

1. I got tested as an adult for ADHD

2. I experimented with diet, and found that a low carb diet reduced my symptoms quite noticeably. I continue to be shocked by what an affect my diet has on my experience and behavior.

3. I experimented with sleep, and having consistent sleep and wake hours also helped.

4. I started to meditate, do yoga, lift weights, get traditional Thai massage, do cold plunges. This helped me connect more with my body and get out of my head. I found that I needed to practice connecting with my body and my emotions because my analytical/cognitive voice was so dominant I wasn’t so aware of another way of experiencing reality.

5. I’m currently reading about Internal Family Systems therapy which I’m pretty sure will help a great deal but I haven’t yet found a therapist and started applying it.

6. I realized most of my friends and the people I connect with best are neuro-atypical in some way, and I love them for being quirky and unusual. It made me feel less embarrassed or ashamed to have weird aspects to my character, and embrace my strengths and accept my weaknesses (and add some systematic mitigations).

Hey,

2: yeah diet is important. A long time a drunk coffee. I try to avoid coffee know.

Also bread have a negative effect. So far I can tell.

3: that’s not so easy with a child :P

4: I’m walking. That’s my meditation.

5: thats interesting I give it a try.

Thanks for helping.

> 2. I experimented with diet, and found that a low carb diet reduced my symptoms quite noticeably. I continue to be shocked by what an affect my diet has on my experience and behavior.

Same. Switching to Keto definitly impacted my ability to focus and think clearly. Not sustainable for the long term, but cut out as much of the refined carbs and sugars as possible. Dialed back the coffee, too -- big difference.

I agree. The implementation here will still enable us to learn about some aspects of UBI. The question of how benefits affect the motivation to gain re-employment will still need to be studied in further experiments.

I think it's worth celebrating that Iceland is doing controlled tests to explore this way. To really understand the best design for UBI, we need more of this.

It's Finland, not Iceland.