Hi, I'm Diego!

AI Engineer
hello@diegobit.com


WHERE TO FIND ME

I repost AI news
on X and LinkedIn

Open source stuff
on GitHub
and HuggingFace

Videogame videos (in Italian) with
@indiependenti

hello@diegobit.com

I'm an Italian programmer working on GenAI based in La Spezia.

Feel free to browse around! You'll find some thoughts on AI, the world of software, videogames, my dev tools and interesting links from around the web.


Recent updates

Crawling shouldn't be that hard

o3 has a powerful search – it can even keep searching the web for minutes until it finds the precise location where a picture was shot. Maybe my Opinion Search is already outdated...

Sometimes all I want is to download the documentation of a website and upload it to Claude or ChatGPT project documents. When I'm about to work on something for a few days where I know the required knowledge is newer than the LLMs' cutoff, I don't want to ask the llm to do a web search at the start of every chat to re-load all the context ...

--More--

An MCP server for my Aranet4 CO2 sensor

I recently bought an Aranet4 CO2 sensor to monitor the CO2 concentration in my home. There are studies showing that high CO2 concentration decreases cognitive performance significantly (I recommend to start your research with this video from DHH).

The Aranet4 seems to be (one of) the only devices with a dedicated co2 sensors, and is for sure the gold standard. Unfortunately, even though it is quite expensive (I spent 220 euros in italy), you cannot store historical data on the mobile app, which is limited to the 7 days stored in the device itself.

That's why I built these ...

--More--

Thoughts on ARC-AGI-2

The recent ARC-AGI-2 release makes me think about how we learn and puzzle videogames. Recently OpenAI recent o3 models reached 75.7% and 87.5% accuracy on the first version of the ARC-AGI benchmark. Hovewer, it had a per-task cost of 200$ for the high efficiency version, and much more for the low efficiency version. They don't mention the cost, they just say it requires 172x the compute. You can make an estimate yourself... I won't make an analysis of o3 and LLMs, for those I'll refer to the blog post and the YouTube interview with Machine Learning Street Talk. What matters ...

--More--

How I use LLMs

If you are not conviced, or you don't understand LLMs, ChatGPT etc, please watch this video. Karpathy has an exceptional ability to explain complex topics in simple non trivial ways. He recently started a series targeted at a general audience that explains all the fundamental building blocks and provides the correct mental models to understand ChatGPT, how it works and how it's trained.