Digest #529 → Build with 20k+ others - Become a Pro member³⁷

Hey folks, guess who’s finally building¹⁴⁶ with AI coding tools? Me! I think this is the actual process⁶⁰⁷ non-technical folks need to use to build with these tools (not learn to code)

Also scheduled new workshops all around building:

• 11th Feb with bolt.new²⁰⁶ - Create full-stack apps and deploy them. (sign up¹⁵⁵)

• 14th Feb with builder.io¹⁶³ - Get working code from your Figma designs. (sign up⁷⁷)

• 19th Feb with create.xyz²²⁶ - Quickly bring your AI app ideas to life. (sign up⁷⁹)

more workshops for building with AI (Cursor, Replit) are coming soon.

TLDR; inside today’s newsletter

• o3-mini arrives and a new agent in ChatGPT

• Gemini gets smarter, jailbreaking Claude and Mistral Small

• conversations from our community

• 5 AI tools with a new music maestro

• what caught my eye from OpenAI’s Reddit AMA

• 5 interesting reads including the death of Wikipedia

• to-dos

Alright, let’s get to it!

 🔎 Trends & news

• OpenAI’s o3-mini is finally here⁸³. It comes in two flavours for users: o3-mini and o3-mini-high. It’s available to everyone in ChatGPT (including free users)

  • Plus & Team users have up to 150 messages per day with o3-mini and 50 messages per week on o3-mini-high. Pro users have unlimited access.

  • On benchmarks, o3-mini’s performance³⁹ is much stronger than o1-mini but it is still dumber than o1. o3-mini-high is supposed to be a bit (not much) smarter than o1, but it failed a real-life test¹⁰³ that only o1 has passed for us.

  • These models can’t understand images (o1 can), but they can search the web (o1 can’t). ik, it’s a mess. o3-mini for devs⁵⁸ has more options though.

• OpenAI also launched Deep Research¹⁰⁶, a new “agent” in ChatGPT that’s based on the full o3 model.

  • Similar to Gemini’s feature with the same name⁶⁹, it scans through multiple websites, reasons about them and adds an intelligent POV to compile comprehensive reports (vs Google’s implementation which mostly summarizes what it found).

  • To be fair, Gemini Deep Research is $20/month with unlimited reports. ChatGPT Deep research is only for Pro users rn ($200/month) and you get 100 queries a month and it’s still a great value for money in my opinion.

  • Dan Shipper took it for a run²¹⁴ and calls it a “bazooka for the curious mind”. Ethan Mollick created a 30-page report on table top games²⁹⁶ using Deep Research and shared his views^272† on it.

  • And we already have two open-source versions trying to do the same: Open Deep Researcher²⁰⁶ and GPT Researcher¹⁷².

• Gemini quietly got upgraded⁸⁶ to a stable version of 2.0 Flash and Imagen 3. The 2.0 Flash model for devs is still under the “experimental” label but I assume a stable version is coming soon.

• Anthropic has a new paper on their approach to defending Claude against universal jailbreaks¹²⁷. There’s a demo too⁸² that you can try to break.

• Perforator helps devs slash server costs by 20% by spotting inefficient code in real-time. Open-source, runs with zero slowdown, and works with C, Python, Rust & more. Free on GitHub—because optimizing infra shouldn’t cost billions. 🚀 Try it now¹¹¹*

• Mistral AI released a new model - Mistral Small 3⁸⁶. They call it small but it has 24b parameters. Pushing it, innit?
  *sponsored

💬 Inside the community this week

Become pro⁶⁹ to join us building and learning together.

• Jeff asked about voice transcription apps and SuperWhisper alternatives. (link¹⁷⁷)

• Amie’s accepting questions for her “tips for using bolt” office hours. (link⁴⁶)

• Several members shared their task-capturing workflows in a thread. (link¹⁰⁷)

• Este and Wyatt compared notes on using Gemini’s Deep Research vs ChatGPT’s Deep Research. (link¹¹⁷)

• I shared Sahil’s (Gumroad’s founder) process of building with AI tools. (link¹⁶⁶)

• Daniel asked for help with implementing payments to his project built with Lovable. (link³³)

• Keshav shared that Grok is improving fast and he uses it for real-time queries. (link⁶⁵)

• H. Chad shared a study that claims AI explanations can change people’s minds about conspiracy theories. (link⁴³)

Join the conversation, plus full access to courses and workshops by becoming a pro member⁶⁹ today!

⚙️ Top new tools

• Fizz by Riffusion²⁵⁰ - Personalized, full-length, high-quality music. Free for a limited time.

• Text-to-API³³⁷ - Turn any website into an API and get structured data with Firecrawl.

• Gemini in Google Sheets²⁵⁵ - Generate charts and valuable insights right in your spreadsheets.

• Image to ASCII Art¹⁸³ - Convert your images into ASCII art (built²⁷ with o3-mini-high)

• Founderpal²⁴¹ - AI marketing platform to grow your business.

More tools here →¹⁰⁸
Showcase your tool at the top of this list. Book an ad here^19†.

👀 What caught my eye… from OpenAI's Reddit AMA

OpenAI did an AMA on Reddit⁵¹ after o3-mini’s launch. I combined the key answers (esp. the ones about their near future plans) below. Direct statements from OpenAI members are “italicised inside quotes”.

• GPT-4o is not done. This model will get more improvements soon and a highly anticipated feature: image generation using 4o (and not DallE-3) is coming in a few months.

• Combining all the features of ChatGPT (file uploads, search, canvas, voice etc.) with the reasoning models (o1, o3-mini) is a high priority. “The next step is integrating all of this so you can speak to a model that reasons as it searches and produces a canvas that runs Python.”

• o3 when? “in more than a few weeks, less than a few months.” I am expecting March end.

• showing thinking tokens: “bunch more than what we do today - very very soon. TBD on all.”

• Q: I wish ChatGPT could analyze images in PDFs (Claude has this). A: OpenAI recently released that in the Enterprise version²⁹. Coming to Plus in the future.

• Sam on Deepseek R1: “it's a very good model! we will produce better models, but we will maintain less of a lead than we did in previous years” and “R1 updated us on this [showing thinking tokens]”

• Sam also mentioned that “they’ve been on the wrong side of history” when it comes to open source. They are considering some changes but he also said that “not everyone at OpenAI shares this view” and it’s not a priority.

• Updates for Advanced Voice Mode, Assistants API and long context window are under work but no timeline for when these updates will land.

 📜 Interesting posts

• OpenAI’s Noam Brown called Deep Research the potential end of Wikipedia²⁶⁸. Shawn (@sywx) made good counterpoints and the exchange is worth reading.

• a16z’s 2025 update for voice agents²⁹².

• How conversational AI will change entertainment²⁶⁴.

• When AI helps doctors, when it gets in the way and when doctors get in AI's way¹⁷⁰.

• ChatGPT’s paid subs grew to 15.5M⁴⁰ in 2024 (up from 5.8M in 2023).

📌 To-dos

That’s it for today. Feel free to hit reply and share your thoughts. 👋

Enjoy this newsletter? Please forward to a friend.

Want to join a community of AI-curious folks? Become a Ben’s Bites member³² and get full access to our Slack, workshops, and courses.

Want to advertise in this newsletter? Click here²⁶.