🧙🏼 Anthropic on AI consciousness

Noticed how ChatGPT turned into a personal hype machine lately? Reddit's losing its mind over ubiquitous AI responses like "Dude. You just said something deep as hell without even flinching. You're 1000% right." Sam Altman calls it "glazing" and promises to dial back the enthusiasm.

Howdy wizards,

A special welcome to the 239 new subscribers who joined last week.

Optimize your seating configuration and initiate your preferred caffeine delivery protocol.

Here’s what’s brewing in AI.

DARIO’S PICKS

1. GPT-4o image generation now available in the API

130 million users have tested GPT-4o’s native image generation inside ChatGPT since its launch just one month ago.

The model is now available inside OpenAI’s API as gpt-image-1. Translation: Businesses can now start building high quality, accurate image generation in their applications.

Major creative tools are already building it right into their platforms:

• Airtable: Introduced image gen for campaign concepts, asset remixing, and localization

• Figma: Added image gen and AI editing capabilities directly in a Figma page

• OpusClip (AI video clipping tool): Added one-click YouTube thumbnail generation

• Adobe: Added ability to generate new aesthetic styles in Firefly and Express

The price that the companies pay for GPT-4o’s image gen is ≈$2-9 cents per generated image, depending on the chosen image resolution.

‎ Why it matters‎ ‎ Design work is changing before our eyes. Seemingly, it’s going in the direction of coding—the tools of the trade (PS, Figma, Canva, etc) get a deeply integrated layer of AI that can handle any aspect of creating and editing designs. The 1.0 versions are going to be deeply flawed and there will be pushbacks. But the direction seems clear: designers will move from pixel perfectionists to expert tastemakers and architects—as well as being the crucial human touch in dealing with clients.

IN PARTNERSHIP WITH FELLOW

Free AI Resource

Unlock powerful workflows from two CEOs who are reinventing the future of work with AI.

In this free resource, you'll discover how forward-thinking leaders are using AI to streamline meetings, cut busywork, and automate decisions.

Get actionable insights that you can implement in your own remote teams today.

Get your free copy

DARIO’S PICKS

2. Anthropic’s “model welfare” lab

Anthropic just launched a new research program to investigate and prepare for what they call model welfare. It’s designed to explore if AI, now and in the future, might be conscious and be deserving of moral consideration.

The company now has their own dedicated AI welfare reseracher, Kyle Fish, who first joined the alignment team before entering this role. The research from the welfare program will intersect with other efforts including alignment, safeguarding, Claude’s character and interpretability.

The discussion about whether AI is currently or will become conscious is at the core of this initiative.

‎ Why it matters‎ ‎ What Anthropic is discussing here is mainly to what extent AI has or is liable to develop agency, ie “make its own choices”. There’s no doubt that’s both super relevant to alignment and to shaping our relationships with these models.

When discussing consciousness, though, they’re starting off on the assumption that human consciousness is something that’s generated in our brain. Like a byproduct of having a high IQ or something. Mind you, the scientific approach to understanding consciousness has a long and proud history of yielding no results; some even say modern science was explicitly designed to exclude consciousness.

All this to say—it’s fascinating that AI is collectively confronting us about our own definition of what consciousness actually means.

Quite possible the biggest question brewing in AI these days.

UP CLOSE

In this mini-series I share different ways I’m using AI from week to week, as well as practical tips & tricks I discover and actually use.

A prompt to make o3 sound like you

You might’ve noticed I rarely share prompts in this newsletter. I'm not a big fan of pre-made prompts for two reasons; I prefer learning through experimentation, and I’ve found “universal” prompts rarely fit work.

But here's an exception —

This prompt gets o3 sounding like you by analysing your writing samples, giving itself feedback and going into a long reasoning loop to pin down your style.

Useful since ChatGPT doesn’t have custom writing styles, like Claude does.

A ChatGPT hack for faster prompt testing

I found a cool UX upgrade in ChatGPT. Leave a prompt sitting unsent in ChatGPT, open a new tab, and—boom—your prompt is pre-loaded. Fire off as many parallel runs as you like without copy-pasting.

Yes, it’s helpful if you wrote a long prompt and accidentally closed the tab. But, you can also use it to open a bunch of new tabs and send the same prompt.

I use it to:

• Test prompts that will be used in repeatable workflows, to quickly check the consistency in the AI’s response (since there’s random variation in all LLM responses)

• Generate multiple variants (images, copywriting ++); again, the random variation makes each output different so I want to test it several times and cherry-pick the best

Keep in mind, what usually makes sense isn’t one-shotting like this but having an actual conversation where you refine the output with follow-ups.

FROM OUR PARTNERS

Vibe code your new business app

Want to build business software & apps but aren’t a dev?

Databutton is built on top of the world’s first reasoning AI-developer, and will turn your new app idea into reality:

• It breaks everything down into a series of tasks and works with you through the entire build process

• The agent isn’t a command taker, it actively collaborates with you to make sure you land on the right solution (kinda like your own personal CTO)

• It manages the full-stack, including frontend, backend, and deployment

Try Databutton

COMPANIES THAT SHIP WITH AI

Explore 700+ AI case studies on contextwindows.ai

THAT’S ALL FOLKS!

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.