This discussion revolves around the nature and capabilities of Large Language Models (LLMs), with a strong focus on the analogy of LLMs as a "lossy encyclopedia." Several key themes emerge:
The "Lossy Encyclopedia" Analogy and Its Limitations
A central theme is the utility and accuracy of comparing LLMs to a "lossy encyclopedia." Many participants find the analogy helpful for understanding LLMs' ability to store and retrieve vast amounts of information, but also their inherent imperfections.
- "I find calling something a 'questionable analogy' makes it a tiny bit less likely that people will pick it apart with thousands of reasons it's not an exact match for what it's describing." - simonw
- "LLMs are a lossy encyclopedia with a human-language answering interface. This has some benefits, mostly in terms of convenience." - thw_9a83c
- "The problem is that language doesn't produce itself. Re-checking, correcting error is not relevant. Error minimization is not the fount of survival, remaining variable for tasks is. The lossy encyclopedia is neither here nor there, it's a mistaken path:" - mallowdram
- "I think an LLM can be used as a kind of lossy encyclopedia, but equating it directly to one isn't entirely accurate. The human mind is also, in a sense, a lossy encyclopedia." - kgeist
- "My point is that I find the chosen term inadequate. The author made it up from combining two existing words, where one of them is a poor fit for what they’re aiming to convey." - latexr
Participants debated what "lossy" truly means in this context, with some arguing it implies missing information that is obvious, while others believe it can extend to making up information (hallucinations) or changing answers. The non-deterministic nature of LLMs also plays a role here.
- "A lossy encyclopaedia should be missing information and be obvious about it, not making it up without your knowledge and changing the answer every time." - latexr
- "The units themselves are meaningless without context. The point of existence, action, tasks is to solve the arbitrariness in language. Tasks refute language, not the rather than the way around. This may be incoherent as the explanation is scientific, based in the latest conceptualization of linguistics." - mallowdram
- "You never have a clear JPEG of a lamp, compress it, and get a clear image of the Milky Way, then reopen the image and get a clear image of a pile of dirt." - latexr
- "More like a fuzzy encyclopedia" - laser
- "Lossy compression does make things up. We call them compression artefacts." - gjm11
- "The problem is that language doesn't produce itself. Re-checking, correcting error is not relevant. Error minimization is not the fount of survival, remaining variable for tasks is. The lossy encyclopedia is neither here nor there, it's a mistaken path:" - mallowdram
- "It's a terrible analogy because LLMs are already for audio what LLMs are for audio. You can use LLMs to create new songs and sounds. Encyclopedias don't create new songs." - cush
- "I think we will start seeing stateful AI models within the next couple of years and that will be a major milestone that could shake up the space. LLM is merely a stepping stone." - lvl155
Hype vs. Realistic Expectations and Misuse
A significant part of the conversation focuses on user expectations, the role of marketing, and the resulting misuse or disappointment with LLMs. Some users feel that LLMs are oversold as "oracles" or "phd™ level intelligences," leading to frustration when they fail at simple tasks.
- "An oracle was expected because that's what everyone kept saying it was or would be. If LLMs were shown and demonstrated realistically people would think they were really neat and find ways to use them. Instead I'm told I have phd™ level intelligence in my pocket. So of course people are going to be mad when it gets stumped on problems my 4yo could figure out." - osn9363739
- "LLM detractors, for lack of a better word, expect an oracle and then when they find out it's just a lossily compressed blob of human knowledge with natural language as a query interface they say the tool is useless." - baq
- "I think the main value is that it has a unified interface rather than 5000 different websites that you need to learn how to navigate." - jacquesm
- "The first thing I tell the juniors under my supervision: any LLM is not a fact machine, even though it sometimes pretends to be. Double check everything!" - AndyPa32
- "The thing I always tell those who heavily trust its output is to ask it something you either already know the answer to or are something of an expert in; the flaws become much more evident." - bodge5000
- "My company went head first into AI integration into everything. I'm counting down the days until some important business decision is based on AI output that is wrong." - refurb
- "The problem is that in order to develop an intuition for questions that LLMs can answer, the user will at least need to know something about the topic beforehand. I believe that this lack of initial understanding of the user input is what can lead to taking LLM output as factual." - quincepie
- "My mom was looking up church times in the Philippines. Google AI was wrong pretty much every time. Why is an LLM unable to read a table of church times across a sampling of ~5 Filipino churches? Google LLM (Gemini??) was clearly finding the correct page. I just grabbed my mom's phone after another bad mass time and clicked on the hyperlink. The LLM was seemingly unable to parse the table at all." - dragontamer
The Agentic/Librarian Metaphor
Another recurring analogy views LLMs as more sophisticated assistants or librarians, capable of interacting with external tools and refining their output. This perspective emphasizes the potential for LLMs when augmented with search capabilities or structured thinking processes.
- "llm is a pretty good librarian who has read a ton of books (and doesn't have perfect memory)" - tosh
- "even more useful when allowed to think-aloud" - tosh
- "even more useful when allowed to write stuff down and check in library db" - tosh
- "If you're using an LLM you need to think of yourself as an architect guiding a Junior to Mid Level developer. Juniors can do amazing things, they can also goof up hard." - giancarlostoro
- "Modern Russian Roulette, using LLMs for dose calculations." - tuatoru
Chain-of-Thought and Reasoning Capabilities
The concept of "chain of thought" and LLMs' reasoning abilities is discussed as a method to extract deeper or more nuanced information, suggesting models have information beyond immediate recall.
- "Chain of thought seems to be an extraction algorithm for information buried deeper." - pornel
- "The models hold more information than they can immediately extract, but CoT can find a key to look it up or synthesise by applying some learned generalisations." - pornel
The Role of Training and Model Refinement
Several participants point to the importance of training data and methods in shaping LLM performance. They suggest that current limitations might be overcome with better training, rather than being fundamental architectural flaws.
- "A lot of the touted "fundamental limitations of LLMs" are less "fundamental" and more "you're training them wrong"." - ACCount37
- "So there are improvements version to version - from both increases in raw model capabilities and better training methods being used." - ACCount37
- "An LLM is a lossy compression before all else. Then after u can call it names." - larodi
- "With better reasoning training, the models mitigate more and more of that entirely by themselves. They 'diverge into a ditch' less, and 'converge towards the right answer' more. They are able to use more and more test-time compute effectively. They bring their own supply of 'wait'." - ACCount37
Human vs. LLM Reliability and Analogy
The comparison of LLM unreliability to human unreliability is a recurring point, with some arguing LLMs are worse and others finding the comparison useful for understanding their behavior.
- "It's nothing new. LLMs are unreliable, but in the same ways humans are." - ACCount37
- "But LLMs output is not being treated the same as human output, and that comparison is both tired and harmful. People are routinely acting like 'this is true because ChatGPT said so' while they wouldn’t do the same for any random human." - latexr
- "Give me an LLM that can do the work of a decent junior dev, and I'll take it." - chrismt
- "The older I've gotten the more I realize the vast majority of people talk absolute rubbish most of the time, exaggerate their knowledge, spout "truths" which are totally inaccurate, and fake it till they make it throughout most of their life." - saberience
- "Are LLMs really lossier than humans? I think it depends on the context. Given any particular example, LLMs might hallucinate more and a human might do a better job at accuracy. But overall LLMs will remember far more things than a human." - jbstack
Safety and Critical Evaluation
The discussion also touches on the potential dangers of LLMs, especially in sensitive areas like medicine, and the critical need for users to verify information and develop an intuition for when LLMs are likely to err.
- "The main challenge is LLMs aren't able to gauge confidence in its answers, so it can't adjust how confidently it communicates information back to you." - cj
- "I used ChatGPT 5 over the weekend to double check dosing guidelines for a specific medication. 'Provide dosage guidelines for medication [insert here]' It spit back dosing guidelines that were an order of magnitude wrong (suggested 100mcg instead of 1mg)." - cj
- "My doctor is looking it up on WebMD themselves." - yujzgzc
- "If you don't want to answer clarifying questions, then what use is the answer??? Put another way, if you don't care about details that change the answer, it directly implies you don't actually care about the answer." - kingstnap