lol

For mistakes etc you could probably ask it to do a second pass. You might even want to try a new, fresh chat so that it doesn’t know what the original was. That’s a good idea that I hadn’t thought about!

In some tests I was doing with using Qwen 0.6B LLMs for classification I did ask it multiple times and basically give more weight the more tries something appears in. In your case you can probably ask two different models and take anything translated equally both times as “good enough” and use an(other) LLM to check the remainining things, although the longer the sentence/text/key the less such a system is likely to help and the more the raw LLM abilities will be necessary.

And as for asking the LLMs for mistakes I was curious because big LLMs should be able to catch some mistakes due to reflexion…

(We’re going to need an AI community lol instead of posting to genzedong all the time)

I had made c/Singularity for things like this but most of the talks about AI/LLMs are on c/technology as it’s big.

then that’s additional context the LLM can use to understand what it’s translating by reading the key property.

I was just thinking that it would be easy to remove the keys for reduced tokens but treating it as extra context for the LLM makes a lot of sense.

Nice job!

And two questions: Do you ask the LLM after it’s done if there were mistakes or if there is anything that can be improved as well? And as for books and longer texts do you have to break them up or do you keep to things that can be done one go?

I am of the opinion that AI controlled ones arent the way things will go.

Maybe with some machine learning for targetting assistance at most. But a fully software piloted one seems pretty far off to me.

There are already drones with AI (mostly for final targeting and perhaps targeting acquisition) and I really doubt there won’t be a massive push for full automation of drones and humanoid bots as soon as it becomes feasible.

In the future, qnything less than fully automated resource extraction, production and war fighting seems like defeat to me.

There is a developer on YouTube that got his hands on a $50k G1 (with working hands) and he has been showing how to make it do things. I found it very interesting.

The basic idea, from what I understand, is that these robots are just now getting the software needed to actually stand and move around without falling, and that’s basically all these robots are shipped with. Anything other than walking and running needs to be programmed to be done, but now that we have LLMs it’s just beginning to be possible to program the robots to automatically decide how to move around and how to interact with the environment.

So, because of that, money and effort are now increasingly being invested into developing these robots into fully automous, and consequently the the prices are now quickly falling because the production at scale is finally happening in anticipation of the software needed being developed soon.

That sounds very pro-colonists and pro-genocide.