šŸ§™šŸ¼ GPT-4.5 rolling out now!

Also: How I use AI on my phone

In partnership with

Howdy, wizards.

I use AI every day for work on my laptop, but itā€™s also a big helper in my everyday life while Iā€™m on the go. Wanted to share a quick screenshot of how the main screen of my phone looks like these days (spoiler: itā€™s like Hogwarts for apps), and how I use these tools.

The green icon is a legacy app for talking to humans. Something I donā€™t do anymore.

Hereā€™s how I use AI on my phone:

ChatGPT: I use GPT-4o as my go-to for easy, everyday questions. If itā€™s a complex question or it needs to do some calculation, I switch to the o1 model. If I run out of o1 credits (happens all the time), I switch to the second best reasoning model, o3-mini-high. I always use the dictation/microphone button to ask questions, not voice mode. This way I donā€™t need to type anything and the AI wonā€™t interrupt my thought process (like voice mode does). I only use voice mode if I need to use AI while my hands are busy, like driving or cooking, or if Iā€™m feeling lazy and donā€™t want to read the answer. I also use the Attach Photo or Take Photo features frequently to make sense of things in my environment, translating stuff, transcribing text, etc. Iā€™m using ChatGPT Plus at $20/mo.

Claude: Claude is the best at natural communication (though it just got some serious competition ā€” more on that later), which is why I use Sonnet 3.7 to help me write stuff. Like if I need to contact customer support somewhere about an issue, Iā€™ll use dictation and explain the situation to Claude, which gives me an email to send. Any text I want to draft and thatā€™s intended to be seen by another human being, Claudeā€™s my go-to. Iā€™m using Claude Pro at $18/mo.

Perplexity: I use Perplexity basically as a Google replacement. Itā€™s the fastest web-searching AI youā€™ll get, and it shows you images and sources in an intuitive way. Also the new deep research feature is amazing if you need to dig deep on a topic. It hallucinates a bit but Iā€™d say itā€™s the best free option for synthesising information from many different sources, including real-time info. Iā€™m using Perplexityā€™s free version as it includes enough usage for my occasional use.

The type of stuff I use AI for on my phone is quite different than what I do on my desktop. There, itā€™s more work-related things: document processing, strategy, copywriting, coding, image gen, data analysis, custom GPTsā€¦less dragons.

Was this helpful to you? Hit reply and let me know, and Iā€™ll share more personal insights like this.

Now, brace yourselves, hereā€™s whatā€™s brewing in AI.

IN PARTNERSHIP WITH FELLOW

Keep your meeting data safe

Worried about all the AI bots joining your meetings?

Fellow is the only AI meeting assistant built with security controls and permissions.

Thousands of organizations rely on Fellow to capture their meeting transcripts, summaries, and action items ā€“ all in one secure platform.

Stop sharing your meeting data with AI note-takers that lack proper security.

Claim 30 days of unlimited AI notes today.

DARIOā€™S PICKS

Source: OpenAI // DALL-E

OpenAI released GPT-4.5 last weekā€”itā€™s the biggest, fanciest, most wallet-draining model to date. It was only available on Pro until today, when I (still a mere Plus subscriber) saw this landing in my model picker:

Great galloping griffins, itā€™s testing time

Hereā€™s the deets on the newest member of the OpenAI model family:

  • Itā€™s not a reasoning model like o1 or the more recent o3-mini. Rather, itā€™s based on scaling unsupervised learningā€”a machine learning technique where unlabeled data is used to find patterns without prior knowledge of the outcomes. That makes it awesome at recognizing patterns and draw connectionsā€”which makes the conversations with it feel more natural and gives it greater ā€œemotional intelligenceā€.

  • The achievement of this model is less about crushing benchmarks and more about increasing understanding of human intent and being better at everyday tasks.

  • It has search, supports file/image uploads and can use Canvas; however it doesnā€™t have voice mode, video and screensharing at the moment.

  • Itā€™s only available on ChatGPT Pro right now, but Plus and Team users should be given access some time this week.

  • All this juicy natural conversation goodness comes at a high price tag for those who want to build apps with it. At $75/$150 per million input/output tokens, GPT-4.5 is about 15-30x more expensive to use than GPT-4o.

  • Since GPT-4.5 is based on pre-training and doesnā€™t think before respondingā€”like the ā€œo-seriesā€ of models doā€”itā€™s better as a general purpose model. Better models based on pre-training means a stronger foundation for reasoning approachesā€”and the two approaches will compliment each other to continue to scale even more advanced models.

On Twitter, Sam Altman asked GPT-4.5 about something interesting, namely the nature of consciousness, hereā€™s its conclusion (end of a longer answer):

Look at that, the search bar just went sentient*

*Reminder that Sam Altman is selling a product

ā€Ž Why it mattersā€Ž ā€Ž Iā€™ve had my hands on GPT-4.5 for just about an hour as I write this. Had time for just a few prompts back and forth. I like it so far, itā€™s got style. Claude might get a run for its vibes. Will report back once Iā€™ve taken it for a real spin.

I think directionally OpenAI is moving towards a more sticky product with these type of improvements. Weā€™re tired of dull, homogenized AI content.

SHAMELESS SELF-PROMOTION (FOR SOMETHING ACTUALLY USEFUL)

Before I goā€”I launched something last week that you absolutely need in your life: the most complete collection of 500+ real-world AI implementation case studies anywhere on the web at contextwindows.ai

  • Filter by industry, agent type, AI provider, and more

  • See what's actually working in your industry

  • Discover which AI strategies are gaining traction 6 months before your competitors even notice them

  • Dozens of new case studies from major AI labs added weekly

You might notice most of the content on the site is locked. Why? Because I want you to pay me so I can finally afford therapy after digging through so much AI news every week.

Iā€™m running an early wizard campaign for my subscribers and thereā€™s two ways you can get access:

āš«ļø Refer 6 people to this newsletter and get free lifetime access (letā€™s be honest, youā€™re doing your friends a solid favor here). You can check your referral count below.

āš«ļø First 100 users: Get lifetime access for $49.99 (regularly $100) with the code WBIA-EARLY-BIRD.

These deals will vanish quicker than Claude can code. Get em while you can.

THATā€™S ALL FOLKS!

Was this email forwarded to you? Sign up here.

Want to get in front of 13,000 AI enthusiasts? Work with me.

This newsletter is written & curated by Dario Chincha.

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 ā€“ it will make it possible for me to continue to do this.