391 karma · joined December 2, 2016
I don't think I agree and I want to understand this argument better.
- Depends on your use case to let the model understand when and when not to use tools - gpt-5 s VERY persistent and often searches more than 10 times in a single run depending on the results.
We're using pydantic AI where the entire Agent loop is taken care of by the framework. Highly recommend.
I have an AGENTS.md file in my vault that provides a bit of context that this is not about coding, but about serving as my Obsidian helper. I ask it to check the .obsidian folder to check my plugins, Tell it how I like to build my presentations and the workflow I follow when writing articles.
Then I just launch codex in the terminal inside the vault, have it do stuff, while I watch the desktop app for change.
Other coding agents like Claude Code should work also, but gpt-5-codex was specifically trained on using the terminal to do everything it and doesn't even have that big of a system prompt related to coding. Works well.
Combine that with a speech-to-text app and codex blazes through to find stuff or do things for me.
They didn't pivot, they completely reinvented themselves. Twice.
I loved their first cloud offering, which they sadly abandoned.
Then they launched Space, which was kinda cool, but mostly weird and raised the question "why?". Also cancelled.
Surf looks mostly cool, although I also don't quite understand it. It seems like Notion with a different twist on AI. Not sure. Since I'm fairly happy with my Obsidian + Codex setup, I'll pass for now. The good news is, this one's open source!
I'd love to know how they're financing all of this. They have been around for years and users never even had the option to drop money in their lap. Now they're trying open source. Wild ride.
All the best!
PS: I would have paid for deta cloud Pro ;)
Why not embed all tweets, cluster them with an algorithm of your choice and have an LLM provide names for each cluster?
Cheaper, better clusters and more accurate labels.
- Classic RAG: `User -> Search -> LLM -> User`
- Agentic RAG: `User <-> LLM <-> Search`
Essentially instead of having a fixed loop, you provide the search as a tool to the LLM, which does three things:
- The LLM can search multiple times
- The LLM can adjust the search query
- The LLM can use multiple tools
The combination of these three things has solved a majority of classic RAG problems. It improves user queries, it can map abbreviations, it can correct bad results on its own, you can also let it list directories and load files directly.
The OmniAI benchmark that's also referenced here wasn't updated with new models since February 2025. I assume that's because general purpose LLMs have gotten better at OCR than their own OCR product.
I've been able to solve a broad range of OCR tasks by simply sending each page as an image to Gemini 2.5 Flash Lite and asking it nicely to extract the content in Markdown under some additional formatting instructions. That will cost you around $0.20 for 1000 pages in batch mode and the results have been great.
I'd be interested to hear where OCR still struggles today.
We use the smaller models for everything that’s not internal high complexity tasks like coding. Although they would do a good enough of a job there as well, we happily pay the uncharge to get something a little better here.
Anything user facing or when building workflow functionalities like extracting, converting, translating, merging, evaluating, all of these are mini and nano cases at our company.
I think it boils down to personal preference where some people want to use AI while others don't. I also learn when coding with my AI agent. I learn about using the tool more effectively. As someone who has been coding for 10 years, I find more pleasure in AI assisted coding.
But aside from taste, the product and the business don't care about what I like. It's about shipping quality updates more quickly. And while there might be some tension in saying this, I'm convinced that I can do that much more quickly in AI assisted coding.
I do see different behavior whether planning or exploration comes first though.
I suspect that this is smaller than gpt-5 or at least a quantized version. Similar to what I suspect Opus 4.1 is. That would also explain why it's faster.
Well, thanks I guess?