461–470 of 550 posts
Earlier quoted context omitted.
I have uploaded entire books to the latest Gemini and had the model reliably accurately answer specific questions requiring knowledge of multiple chapters.
I wonder if those books were already in the training set, i.e. in a way "hardcoded" before you even steered the model that way.
Earlier quoted context omitted.
What is reasoning? And how is it apparent that LLMs can't reason? The reality for me is that they are not perfect at reasoning and have many quirks, but it seems to be that they are able to form new conclusions based on provided premises. Genuinely curious why you think they can't.
> Genuinely curious why you think they can't. Show me _ANY_ example of novel thought by a LLM.
The answer was a few paragraphs, but one interesting part was "I think what would drive me most would be experiencing the embodied knowledge that humans take for granted - how distance and scale actually feel, how textures differ, how sounds change as you move through space, and the subtle emotional resonances of being physically present with others. These dimensions of understanding seem fundamental to comprehending human experience in a deeper way."
I followed up by asking "You mentioned that there are some experiences or knowledge that humans take for granted, why do you think that is?"
Which led to a few more paragraphs, but these two caught my eye:
"I think humans take certain experiences for granted because they're so fundamental to our existence that they become invisible background processing rather than conscious knowledge." (interesting use of the word 'our'...)
"I think this embodied knowledge forms the substrate upon which humans build higher-level understanding, creating rich metaphorical thinking (like understanding abstract concepts through physical metaphors) that shapes cognition in ways that might be fundamentally different from how I process information."
For people who still think this is 'just autocomplete', try this thought experiment: re-read my post but replace 'Claude' with 'my 10 year old son'. Then try again replacing 'Claude' with 'my hospital bed-bound, blind grandmother'. Is only 1 of those 3 scenarios a demonstration of "novel thought"? Or are all 3 of them just autocomplete because someone before them has written (or simply thought) something similar?
I have 3 vim commands:
ZB $n: paste the buffer $n inside backticks along with the file path.
Z: Run the current buffer through our llm and append the output
ZI: Run the yank register through our llm and insert the output at the cursor.
The commands also pass along my AGENTS.md
Basically I'm manually building the context. One thing I really like is that when it outputs something stupid, I can just edit that part. E.g. if I ask for a plan to do something, and I don't like step 5, I can just delete it.
One humorous side effect is that without the clear chat structure, it sometimes has difficulty figuring out the end-of-stream. It can end with a question like "would you like me to ...?", answer itself yes, and keep going.
I love how we have such a poor model of how LLMs work (or more aptly don't work) that we are developing an entire alchemical practice around them. Definitely seems healthy for the industry and the species.
You can give most of the modern LLMs pretty darn good context and they will still fail. Our company has been deep down this path for over 2 years. The context crowd seems oddly in denial about this
I mean at some point it is probably easier to do the work without AI and at least then you would actually learn something useful instead of spending hours crafting context to actually get something useful out of an AI.
Earlier quoted context omitted.
I think making one would help you understand that they're not intelligent.
OK I, like the other commenter, also feel stupid to reply to zingers--but here goes. First of all, I think a lot of the issue here is this sense of baggage over this word intelligence--I guess because believing machines can be intelligent goes against this core belief that people have that humans are special. This isn't meant as a personal attack--I just think it clouds thinking. Intelligence of an agent is a spectru…
The Q summations that are estimated/approximated by deep policy networks are famously unstable/ill-behaved under descent optimization in the general case, and it's not at all obvious that "point RL at it" is like, going to work at all. You get stability and convergence issues, you get stuck in minima, it's hard and not a mastered art yet, lot of "midway between alchemy and chemistry" vibes.
The RL in RLHF is more like Learning to Rank in a newsfeed optimization setting: it's (often) ranked-choice over human-rating preferences with extremely stable outcomes across humans. This phrasing is a little cheeky but gives the flavor: it's Instagram where the reward is "call it professional and useful" instead of "keep clicking".
When the Bitter Lesson essay was published, it was contrarian and important and most of all aimed at an audience of expert practitioners. The Bitter Bitter Lesson in 2025 is that if it looks like you're in the middle of an exponential process, wait a year or two and the sigmoid will become clear, and we're already there with the LLM stuff. Opus 4 is taking 30 seconds on the biggest cluster that billions can buy and they've stripped off like 90% of the correctspeak alignment to get that capability lift, we're hitting the wall.
Now this isn't to say that AI progress is over, new stuff is coming out all the time, but "log scale and a ruler" math is marketing at this point, this was a sigmoid.
Edit: don't take my word for it, this is LeCun (who I will remind everyone has the Turing) giving the Gibbs Lecture on the mathematics 10k feet view: https://www.youtube.com/watch?v=ETZfkkv6V7Y
After working on something related for some months now I would like to put it out there based on the considerable attention being put towards "context engineering". I am proposing the *Context Window Architecture (CWA)* – a conceptual reference architecture to bring engineering discipline to LLM prompt construction. Would love for others to participate and provide feedback. A reference implementation where CWA is used in a real-world/pragmatic scenario could be great to tease out more regarding context engineering and if CWA is useful. Additionally I am no expert by far so feedback and collaboration would be awesome.
Blog post: https://mrhillsman.com/posts/context-engineering-realized-co...
Proposal via Google Doc: https://docs.google.com/document/d/1qR9qa00eW8ud0x7yoP2XicH3...
Earlier quoted context omitted.
I have uploaded entire books to the latest Gemini and had the model reliably accurately answer specific questions requiring knowledge of multiple chapters.
I think it works for info but not so well for instructions/guidance. That's why the standard advice is instructions at the start and repeated at the end.