Track doesn’t mean store.

The word you are looking for is “collect”.

Saying you “track” implies you’re putting together a profile to be consumed in other services

It literally is token prediction with vector search.

Yes there’s an app layer in the chat product for convenience and parsing but the model is exposing methods like ‘complete’ that predict out word sequences.

When model temperature settings are not added in, you get the same exact response every time, just like ngram.

What you’re seeing with Gemini and ChatGPT is context caching to prevent getting sidetracked and response boilerplates with multi-modality so they can call out to image generation, a code completion service, etc. to assemble a complete response.

That’s why the mainstream providers seem so much better (or at least consistent in replies) - each prompt is actually requesting multiple models and doing a lot of application level work to assemble the response you see. That also why they take sooo long to respond.

The language model itself is quite literally a text completion machine, with settings.

The language model itself is a static data structure. On top of that is an inference engine. Currently all inference engines are autoregressive. Meaning they use the current output to produce more output. This isn't necessarily the only way to use a language model. Yann Lecun's JEPA is probably the most publicized project to work a different way.

Facepalm

LLMs are read only.

A developer would be the one using the LLM to evaluate page response text and scraping what they need.

LLMs cannot “crawl pages” or “scrape” or do anything - it’s an offline archive with a few convenience methods.

> LLMs cannot “crawl pages” or “scrape” or do anything - it’s an offline archive with a few convenience methods.

Being a bit needlessly pedantic here. I think it's well understood in these circles that such models are being driven by agents, which absolutely can crawl and scrape. The only thing a human needs to do is write rules for the agent to follow.

No, it allows people to dodge accountability.

It’s also not true.

An LLM cannot visit the internet at all.

There is no “agent” - that is a marketing term. It’s software and LLM that’s it.

There is no "software" - that is a marketing term. It's bits and that's it.

Not how words work, try calling it bits in any setting where you mean software.

Wouldn’t matter in court anyway or anywhere you put it to the test even if you called it bits.

Point is dodging accountability by saying some “AI did it” when it’s not reflective of reality.

LLMs are read only and have no access to the internet :)