Talk:Large language model

Lead length and complexity

Hi 0xReflektor. I saw that you fixed the template "lead too short" in the article large language model; the lead was indeed too short (around half the size of a typical lead) but has now become overly long (around twice the size of a typical lead). We should probably condense it, or move some of its content to the rest of the article.

I also believe the lead is too difficult to understand for a Wikipedia audience (WP:TECHNICAL). It would be well-written for a research article, but many people come to Wikipedia to discover what the concept means, and so we should try to write the lead so that they understand most of it even if they lack many of the underlying concepts and jargon. Alenoach (talk) 14:27, 4 October 2025 (UTC)

I add this message just to indicate that I removed on October 11 one lead paragraph and integrated another into the body. The lead's length is more normal now. There is still the issue with the technical complexity though. Alenoach (talk) 18:42, 20 October 2025 (UTC)

I attempted to reduce the lead's technical complexity by simplifying language and moving some more technical sections to the body. I also removed terminology that the source papers did not use, such as "few-shot learning" and "hill climbing". Diff for posterity and small subsequent copy edit. SenshiSun (talk) 21:18, 26 March 2026 (UTC)

This is unclear

Moving beyond n-gram models, researchers started in 2000 to use neural networks to "learn" language models

Does this really mean "learn about" or perhaps "teach".

Its important: the reader cannot easily understand what's actually meant here. ~2026-24004-11 (talk) 13:35, 10 May 2026 (UTC)

You're right that there was an issue with this sentence, thanks for reporting. I replaced it with "to use neural networks as language models." Alenoach (talk) 15:09, 10 May 2026 (UTC)

"TrueFoundry" listed at Redirects for discussion

The redirect TrueFoundry has been listed at redirects for discussion to determine whether its use and function meets the redirect guidelines. Readers of this page are welcome to comment on this redirect at Wikipedia:Redirects for discussion/Log/2026 June 20 § TrueFoundry until a consensus is reached. Isla🏳️‍⚧ 20:51, 20 June 2026 (UTC)

Finetuning section sentence citation mismatch

Current text: Since humans typically prefer truthful, helpful and harmless answers, RLHF favors such answers.

Citation extract: In fact, ChatGPT’s breakthrough was only possible because the model has been taught to align with human values. An aligned model delivers responses that are helpful (the question is answered in an appropriate manner), honest (the answer can be trusted), and harmless (the answer is not biased nor toxic).

This has been possible because OpenAI incorporated a large volume of human feedback into AI models to reinforce good behaviors.

Problem: We do not know how the humans were instructed to perform their task; reviewers could have been asked to score/correct based on values chosen by OpenAI. We do not know whether said humans are a representative sample of "humans" generally. We do not know whether the choice/identification of alignment values emerged naturally from an unbiased RHLF process or came about during another stage of training altogether The source text does not specify whether the humans are the source of the 3 values or a necessary enabler to imbue the model with such values

In short: To say that a model has been trained by humans to align to 3 values is not the same as those 3 values being preferred by "humans". To illustrate by analogy: A scientist can train a mouse to associate pushing a pedal with a treat but that does not tell us anything about the scientist's motives or preferences.

Proposal: cut the sentence entirely.

~2026-38376-22 (talk) 00:36, 5 July 2026 (UTC)

I've cute the sentence pending further discussion. In simple terms, the source for these claims needs to be stronger, and would have to be summarized neutrally. Grayfell (talk) 01:20, 5 July 2026 (UTC)