Thanks for writing this, I hope people here will actually read this and not assume this is some unfounded hit piece. I was involved a little bit in llama.cpp and knew most of what you wrote and it’s just disgusting how ollama founders behaved! For people looking for alternatives, I would also recommend llama-file, it’s a one file executable for any OS that includes your chosen model: https://github.com/mozilla-ai/llamafile?tab=readme-ov-file

It’s truly open source, backed by Mozilla, openly uses llama.cpp and was created by wizard Justine Tunney of CosmopolitanC fame.

I also thought llamafile deserves a mention. Once you have all model params and tunings done bakes 'em into a single portable binary!

> how people who work in education seem to be incapable of learning anything about education

The people who work in education don't have this issue; the people who work in tech and assume that gives them expertise in education do.

Might be harsh, but more I think about education more I think that the main job for many if not most kids or students in general is making them to learn. By various incentives be it punishments or rewards. Only really at highest levels is there any self-direction. And even there is plenty of external factors like not being paid anymore if you do not publish something.

Eh, working on rebuilding a Montessori school for a couple years showed me that kids are very intrinsically motivated to learn, just not always what the person who’s there to teach them wants them to learn that day. But if you enable that self drive, and gently steer by exposing them to new things at the right level, they can learn a tremendous amount.

Our kids elementary school (Germany) recently voiced concerns that pupils - especially boys - do not fulfill the self defined learning tasks in the expected time.. There may be no upper boundary to how much out of touch educators can be.

No. It would be like finding a memory unsafe caused bug in a Java application that is due to a bug in the JRE. That would absolutely warrant a title like “I found memory unsafe bug in my Java code” when everyone expects Java code to be memory safe, which is analogous to the article in question.

I do not think you are completely grasping what you are talking about (what is a 'memory unsafe bug'?). Even in the example you give, that title would be literally wrong, as there will be no bug in your Java code; there would be a bug in the execution due to a deviation in the runtime executing your program.

When the Lean runtime has bugs, all Lean applications using the Lean runtime also have those bugs. I can’t understand people trying to make a distinction here. Is your intent to have a bug free application or to just show the Lean proof kernel is solid?? The latter is only useful to Lean developers, end users should only care about the former!

The intent is to have a proof of some proposition. The Lean runtime crashing doesn't stop the lean-zip developers from formally modelling zlib and proving certain correctness statements under this model. On the other hand, the Lean kernel having a bug would mean we may discover entire classes of proofs that were just wrong; if those statements were used as corollaries/lemmas/etc. for other proofs, then we'd be in a lot of trouble.

When I see a title transitioning from "Lean said this proof is okay" to "I found a bug in Lean", I'm intuitively going to think the author just found a soundness (or consistency) issue in Lean.

There are no Lean applications other than Lean. This is an important point most of the comments are missing. Lean is for proving math. Yes, you can use it for other things; but no, no one is.

Still good to have found, but drawing conclusions past “someone could cheat at proving the continuum hypothesis” isn’t really warranted.

I have been interested in this topic for a long time and to be honest, there is no better book on the topic than The Mythical Man Month. Yes from the 70s I think, but still the best I have read.

And what is your theory? That it’s better to not save those 2 hours since they will just go to waste anyway? Or that there is diminishing returns to saving work as people will tend to just spend longer on other things they were already doing? How can you be sure those 2 hours will not actually be used by most to do very productive things that in the end look like +4 hours in return??

No. I am not saying that it is a bad idea to do this.

I am saying:

Given you have saved two hours per person per week

Then the value for the company is _not_ equal to two hourly salaries per week. The consequences are just not that simple.

The actor model for concurrency.

Who has that end goal?? Who is going to direct the AI if only the CEO is left in the organization? The CEO will never actually do it , and will always need someone who can and will do it. I just can’t see a grand plan to take humans out of the equation entirely.

that most definitely is a plan, make no mistake about it. but as mike tyson famously said, “every has a plan until they get punched in a mouth” :)

The US engages in preventive wars, generally. For example, the wars in Korea and Vietnam were ultimately fought to prevent the USSR from becoming more dominant than the USA and ultimately to prevent it from becoming so strong that in an eventual direct confrontation they would be able to cause a lot of destruction in the US. Iran is similar: they seem to want to prevent Iran from getting nukes which could then be used to destroy Israel, which the US considers its protectorate. But this is a super slippery slope. It’s essentially the same excuse Russia used in Georgia and now Ukraine: they are near neighbors geographically and culturally that must be stopped from joining the enemy alliance in order to prevent the enemy from attacking Russia in turn, which would be much easier should those countries be part of NATO. But where do you stop? Should Cuba be allowed to join Russia military alliance? Should Mexico be allowed to join BRICS? According to US foreign policy, the answer is always no, because of “national security”.

You haven’t actually done that have you. If you did, you would immediately understand the problems MCP solves on top of just trying to use an API directly:

- easy tool calling for the LLM rather than having to figure out how to call the API based on docs only. - authorization can be handled automatically by MCP clients. How are you going to give a token to your LLM otherwise?? And if you do, how do you ensure it does not leak the token? With MCP the token is only usable by the MCP client and the LLM does not need to see it. - lots more things MCP lets you do, like bundle resources and let the server request off band input from users which the LLM should not see.

> easy tool calling for the LLM rather than having to figure out how to call the API based on docs only

I think the best way to run an agent workflow with custom tools is to use a harness that allows you to just, like, write custom tools. Anthropic expects you to use the Agent SDK with its “in-process MCP server” if you want to register custom tools, which sounds like a huge waste of resources, particularly in workflows involving swarms of agents. This is abstraction for the sake of abstraction (or, rather, market share).

Getting the tool built in the first place is a matter of pointing your agent at the API you’d like to use and just have them write it. It’s an easy one-shot even for small OSS models. And then, you know exactly what that tool does. You don’t have to worry about some update introducing a breaking change in your provider’s MCP service, and you can control every single line of code. Meanwhile, every time you call a tool registered by an MCP server, you’re trusting that it does what it says.

> authorization can be handled automatically by MCP clients. How are you going to give a token to your LLM otherwise??

env vars or a key vault

> And if you do, how do you ensure it does not leak the token?

env vars or a key vault

An authnz aware egress proxy that also puts guard rails on MCP behavior?

Gee, that's starting to sound like a whole "bloated" framework...