I use Deep Research to explore topics. For example:
- Text To Speech Engines. Tortoise TTS leads the open source TTS.
- Open-Source HTTP Servers. Caddy wins.
- Public API-Based Data Storage Options. Supabase wins.
- etc.
But these reports are very long. With O3 and O4 Mini supporting thinking with search, we can do quick research, instead of deep research. One minute, not ten. One page, not ten.
For example, I asked O3 to automate my weekly tech learning process:
From the links at https://www.hntoplinks.com/week pick those that would be of interest to me.
Read the articles and explain the key insights — what’s non-obvious and how can I use it.
When sharing how I can use it, research alternate approaches and suggest which approach is best.
This brings two new capabilties I love.
Memory. I said, … that would be of interest to me
. Given my past chats and context, it really knows me. For example, the reasoning trace shows:
I’m narrowing down the list for Anand, focusing on his interests like generative AI, open-source tools, and minimal libraries.
That’s exactly right!
Long reasoning. It reads each site step by step, unsupervised. Here’s a brief reasoning trace:
Searched the web
Got the Unsure Calculator article. Now, onto the next one: the Hackable AI assistant with SQLite cron jobs. I’ll open it and dive into the details. I think this will be a great fit for Anand, considering the focus on AI tools and development. It should provide some interesting insights!
Searched the web
Now that we’ve moved through a couple of articles, it’s time to tackle the Gemini 2.5 Flash AI model. I think this one will be particularly interesting for exploring advancements in AI. Let me grab it and dive into the details so I can distill the insights into something relevant.
Searched the web
Next up: I’ll open the OpenAI blog article on GPT-4.1 in API. This could bring in some cool technical
…
The result is fantastic. While on a trouser shopping trip, I learnt more than when I browse Hacker News:
- “Unsure” is a tool to calculate uncertainty formulas, like the uncertainties Python package.
- Responses API adds features like linkable history, mid‑chat truncation, flexible reasoning detail, and more.
- Augment Code is an AI code editor that’s growing popular on Reddit.
- GPT 4.1’s 75% discounted prompt caching (instead of 50%) gives them an edge on repetitive tasks. OpenAI
- Nix flakes are a reliable alternative to DevContainers that don’t need Docker – but don’t work on Windows.
- TLS certificates will expire in 47 days from 15 Mar 2020. Automated domain renewals are a must. Digicert
- … and a bunch of other things.
Here are Hacker News summaries for a month or specific days:
- Mar 2025: Mistral OCR is a dedicated OCR, not an LLM. DuckDB UI has notebooks!
- 13 Apr 2025: Learnt about Signalbloom which automates Q1 summaries and this MCP Primer
- 15 Apr 2025: 12-factor agents shares agent lessons. Send JSX instead of JSON. AI as Normal Technology.
This enables:
- Personalized learning, i.e. it tells me what I don’t know, and how I can apply it. This is powerful.
- Learning on the go, e.g. via voice while cycling or walking.
- Learning from untapped sources. This includes: GitHub repos, research papers, open data registries, patent filings, earnings transcripts, subreddits, judgements or acts, open data repositories or collections, and many more.
Take GitHub as an untapped source. I asked O3:
Go through the OpenAI Codex CLI repo on GitHub.
Teach me innovative, new, and surprising techniques or approaches or libraries I might not know about.
For each, explain what is interesting, how I might use it, and how this approach contrasts with alternatives.
This was a treasure trove, too! I learn about:
- Ink: React for CLI!
- Zod: schema validation
- Creating firewalls
- Fuzzy path matching
- … and a bunch of other things
I tried O4 Mini High and saw similar results. I felt O3 still gave me more personalized suggestions.
Let’s see what we can learn from pull requests on Codex.
Go through all the pull requests that have been merged into OpenAI’s codecs repository on GitHub. Pick the ones that would be most interesting to me. You can group a few if they are very related and give me the top 10 most interesting PRs that would be relevant for me. Also explain why these are relevant to me, how I might use them, and any interesting details about the way in which the PRs were written.
Soon, we won’t just follow a lesson plan — we’ll have lessons built just for us. AI will track how we learn and adapt in real time. It’ll feel like having a personal coach in your back pocket. That future starts now.
This also opens a door to endless curiosity. There’s no limit to what we can explore. Curiosity is the competitive advantage, now.