Recent content by RyanMorrison

Love MCP and use it myself all the time But this wasn’t a test of desktop apps, it was the web interface and MCP doesn’t work on anything but desktop

I agree Claude is better for front end coding but if you haven’t tried Gemini 2 then you’d know it is much better than it was previously

That is why I use very specific prompts. The results should be largely the same in terms of quality of output and comparable to mine.

If you’ve ever crossed the street in London that is exactly how it happens

I decided this one purely on aesthetic. Grok more closely matched the more casual - practice - concept.

DALL-E, the image generator used by ChatGPT is ancient in AI terms and is one of the worst of the major models. I rarely use it in tests anymore because of how bad it is in comparison. Google Gemini uses Imagen 3 which is only a couple of months old and surprisingly good.

Google also now runs an AI query for pretty much every search, showing the results in its AI overview. This does exactly the same thing as ChatGPT Search but worse.

You can code a version of the Oregon Trail with AI tools like Cursor. I've also used it to create a version of the game Taipan.

I agree that there are different prompting techniques for different models but I try to write simple descriptive prompts that are relatively generally applicable. Also, this is closer to how most 'average' users might utilize AI so is a good test of how they perform. I will do some more...

I think most users of tools like Cursor will be developers working for big companies looking to automate the boring parts of the job, including running tests and repetitive code tasks. Engineers are OpenAI, Midjourney and ElevenLabs are already using Cursor. It is basically a fork of VSCode with...

Unfortunately most of them have a relatively small context window so can't take more than about 500 characters of text in as a prompt. Even then the longest single video any of them can create is about 1 minute and that is Sora which isn't publicly available yet. We are getting closer though.

This is an interesting point. On the final element about different kinds of AI. With the demos I've seen of GPT-4o it seems one model can be general across these states. It is capable of reacting and changing base don human input and the change is being natively multimodal in and out.

Apple's approach is interesting though as it is pretty much all in your hands. It uses primarily on-device models with your data.

Some nice prompting work. The images look good. I am still impressed when AI gets text on images right - I sugggest trying Ideogram. It’s very good at legible text on images.

Jumping on this much later but you can use --cref create your first character then for each subsequent character type --cref and the url of the image to the first character. Use --cw [number from 1-100] to set how closely you want it to reflect the first. 100 is identical.