I like how the landing page (and even this HN page until this point) completely miss any reference to Meta and Facebook. The landing page promises privacy but anyone who knows how FB used VPN software to spy on people, knows that as long as the current leadership is in place, we shouldn't assume they've all of a sudden became fans of our privacy.

Ollama isn’t connected to Meta besides offering Llama as one of the potential models you can run.

There is obviously some connection to Llama (the original models giving rise to llama.cpp which Ollama was built on) but the companies have no affiliation.

Could you be more specific about which book and chapter talked about this problem and how the solution was different or similar?

Of the top of my head I can think of Extreme Programming Explained, The Art of Agile Development, Software Teaming, The Pragmatic Programmer.

What's the ublock block thing?

IP is IP. LLM vendors are in legal battles with all sorts of IP owners. Putting it explicitly there (as awkward as it is) creates a legal obligation where the company or individual that fed it to "AI" cannot deny: "oh! I didn't know, therefore it's not stealing"

Please elaborate

Basically, there's a spectrum which has at its ends two extremes:

1. R&D 2. Maintenance

These aren't about the actual activities performed; R&D could also be something like finding product-market fit at a startup and maintenance could also be working in product/engineering on, say the MS Word team. Just two opposite ways of operating.

In R&D mode, you want to encourage diverging ideas and err on the side of just trying if something works, which typically means you hire small teams of people who think differently, don't anchor too much to the past and avoid excessive layers of management and bureaucracy.

This is because you want to EXPAND the scope of possible actions and outcomes.

In maintenance mode, you want to get things just right every time and minimize the downside. For example, a radical redesign of MS Word would be a horrible idea because the entire world is used to a design they've known for 20 years. Here, you want people who keep things more or less the same. The same is true for internal infrastructure like billing, APIs, etc. This requires people who are less creative, but extremely thorough.

This is because you want to CONTRACT the potential scope of actions and outcomes.

I don’t know. With the advent of AI-productivity tools (regardless of where we stand in how much the actual boost is), EMS should be able to offload part of their admin work and devs should be able to deliver more and faster. Long way to say that I'm skeptic if those numbers still apply in modern or near future environment.

That's a really interesting perspective that I hadn't considered!! Per the first link, a lot of these ideas are based on how well a human can manage in their head what another human knows/doesn't know/is doing/isn't doing/etc. We can only manage so much of that. I still don't think a human can manage more than 6/7 complexities, and so I don't know if I think 1 human would be able to work with say 30 AI agents as 1 team, for example. Could 1 human work with 1 agent who controls 1000s of AIs, I think so, is this the part you think changes?

The tutorial itself is next level with stunning interactive visual story telling.

All of Josh's tutorials are like this. Check out some of his others!

It's a common pitfall to measure what's easy. OTOH, the calculations can be quite arcade and hard to relate to. Many observability platforms have something in place but they leave more to be desired.