zbyforgotp

This is exactly the behaviour that I would expect from a human teleported into a project and given a similar task.

After some time humans would gather some background info needed to be more productive and we need to find out how to copy that.

Personally I would not hardcode the discovery process in code but just gave the llm tools to browse the code and find what it needs itself.

Or maybe it could alter the behaviour in a positve way: https://en.m.wikipedia.org/wiki/Parasites_Lost

I skeptical. Carbs are converted to glucose in the intestines and we eat orders of magnitude more glucose in carbs than in sweets so candida should not really care about sweets, but all these studies are about craving sweets not carbs in general.

Yeah - that’s what I thought. The original article sounds like motivated reasoning - solar farms don’t look like “green technology” so there is that great need to make them more palatable and project like this cater to that need instead of economic calculation.

LLMs are not that different from humans, in both cases you have some limited working memory and you need to fit the most relevant context into it. This means that if you have a new knowledge base for llms it should be useful for humans too. There should be a lot of cross pollination between these tools.

But we need a theory on the differences too. Now it is kind of random how we differentiate the tools. We need ergonomics for llms.

I’ve heard this many times, also from good sources, but is there any gears level argument why?

LangChain is great for finding out who has a clue.

Yeah that was a typo

I keep a special laptop for all financial stuff. It is not that expensive and the inconvenience is limited. And on it I have CubesOS, but I guess OpenBSD would also be a good choice.

The problem is that the llm don’t learn to play moves from a position, the internet archives contain only game records. They might be building something to represent position internationally but it will not be automatically activated with an encoded chess position.

You can make it available to the player and I suspect it wouldn’t change the outcomes.

Is test time the same thing as inference time?

I found out that I can register myself as a company at Microsoft and then I can buy the license to that new account.

I cannot use it - it requires a login from school or other institution.

From the article:

„The problem is not that Halassy used self-experimentation as such, but that publishing her results could encourage others to reject conventional treatment and try something similar, says Sherkow.”

How do you run code in Bedrock? I don’t use computer games at all, but my son plays Minecraft a lot and is into programming. I investigated options for programming in Minecraft, there is Tinker, an app that lets you program in scratch, but it is not very usable, and there are Minecraft servers where you can program in Java, but the learning process is rather steep for that one and also it works only for Java Edition Minecraft, while most of my sons friends use Bedrock. Supposedly there is also Education Edition, but I have no idea how to install it - it seems that it is only available for schools.

Bullshit jobs once again. I don’t know. These companies are complex systems.

He is writing as if the engineers all knew how to fix the systems, but were just powerless to do that. But I’ve also seen projects lead by engineers that only added to the overall complexity.

There is a paradox in this - the people who seem the most confident about fixing the systems usually only make things worse. Chesterton fences and stuff.

This article triggers me because everybody who reads it will always believe that they would fix the mess if only they got the power, but in practice when they get power they would only add new complexity to the whole mess.

Ok, but at least it suggests that this other thing might be more efficient in some ways.

What is the range that she is keeping?

This is guaranteed, function calling without it is not. The old way can work for you, but my experience is different, especially with complex schemas.

Tables are a hard case for RAG, even if you parse them perfectly into Markdown, the LLMs still tend to struggle with interpreting them.

Traditional does not equate original. The original technique was never used widely and cannot be called the traditional way.

What you describe sounds like Agetic RAG https://zzbbyy.substack.com/p/agentic-rag

TLDR persistent people knowingly take risks while stubborn ones don’t want to know that there are risks. But you have to take risks to be successful.

It is an interesting question why they don’t outcompete other businesses.

Rewriting from scratch is harder than it looks, but it all depends on how far you existing software is from your goal. Imagine that what you have is a word processor and you need a web server, surely you would start from scratch. You might say that I chose an absurd example here, but it shows that sometimes what you have is too far from what you need for evolutionary approach.

It is easy to check for literal copies of the training material. I expect the providers to just add a filter doing that.

But I don’t think it resolves all the problems. Newspapers are losing their business model. Maybe it is just creative destruction and we’ll have something better in their place. Maybe, but journalism is quite important in our societies.

This sounds like the famous system 1 and system 2 thinking with the prover doing the systematic system 2 thinking and the llm doing the intuitive system 1 thinking.

But will people use them?