Can you confidently say that the architure of the LLM doesn't include any a priori bias that might effect the integrity of this LLM?
That is, the architectures of today are chosen to yield the best results given the textual data around today and the problems we want to solve today.
I'd argue that this lack of bias would need to be researched (if it hasn't been already) before this kind of model has credence.
LLMs aren't my area of expertise but during my PhD we were able to encode a lot of a priori knowledge through the design of neural network architectures.
It's not akin to a psychopath telling you they're sorry. In the space of intelligent minds, if neurotypical and psychopath minds are two grains of sand next to each other on a beach then an artificially intelligent mind is more likely a piece of space dust on the other side of the galaxy.
Start with LLMs are not humans, but they’re obviously not ‘not intelligent’ in some sense and pick the wildest difference that comes to mind. Not OP but it makes perfect sense to me.
I think a good reminder for many users is that LLMs are not based on analyzing or copying human thought (#), but on analyzing human written text communication.
--
(#) Human thought is based on real world sensor data first of all. Human words have invisible depth behind them based on accumulated life experience of the person. So two people using the same words may have very different thoughts underneath them. Somebody having only text book knowledge and somebody having done a thing in practice for a long time may use the same words, but underneath there is a lot more going on for the latter person. We can see this expressed in the common bell curve meme -- https://www.hopefulmons.com/p/the-iq-bell-curve-meme -- While it seems to be about IQ, it really is about experience. Experience in turn is mostly physical, based on our physical sensors and physical actions. Even when we just "think", it is based on the underlying physical experiences. That is why many of our internal metaphors even for purely abstract ideas are still based on physical concepts, such as space.
Without any of the spatial and physical object perception you train from right after birth, see toddlers playing, or the underlying wired infrastructure we are born with to understand the physical world (there was an HN submission about that not long ago). Edit, found it: https://news.ucsc.edu/2025/11/sharf-preconfigured-brain/
They are not a physical model like humans. Ours is based on deep interactions with the space and the objects (reason why touching things is important for babies), plus mentioned preexisting wiring for this purpose.
Isn't it obvious that the way AI works and "thinks" is completely different from how humans think? Not sure what particular source could be given for that claim.
I wonder if it depends on the human and the thinking style? E.g. I am very inner monologue driven so to me it feels like I think very similarly as to how AI seems to think via text. I wonder if it also gives me advantage in working with the AI. I only recently discovered there are people who don't have inner monologue and there are people that think in images etc. This would be unimaginable for me, especially as I think I have sort of aphantasia too, so really I am ultimately text based next token predictor myself. I don't feel that whatever I do at least is much more special compared to an LLM.
Of course I have other systems such as reflexes, physical muscle coordinators, but these feel largely separate systems from the core brain, e.g. don't matter to my intelligence.
I am naturally weak at several things that I think are not so much related to text e.g. navigating in real world etc.
Interesting... I rarely form words in my inner thinking, instead I make a plan with abstract concepts (some of them have words associated, some don't). Maybe because I am multilingual?
English is not my native language, so I'm bilingual, but I don't see how this relates to that at all. I have monologue sometimes in English, sometimes in my native language. But yeah, I don't understand any other form of thinking. It's all just my inner monologue...
No source could be given because it’s total nonsense. What happened is not in any way akin to a psychopath doing anything. It is a machine learning function that has trained on a corpus of documents to optimise performance on two tasks - first a sentence completion task, then an instruction following task.
I think that's more or less what marmalade2413 was saying and I agree with that. AI is not comparable to humans, especially today's AI, but I think future actual AI won't be either.
An absolutely fantast book. I've read it cover to cover a couple of times during my PhD (which focused on neural networks and numerical solvers). Gives the right amount of depth to serve as a detailed introduction whilst covering a lot of the key areas in optimisation. I still use this book as a reference years later.
I recently picked up Gleam for a small medical imaging project and I can quite honestly say I've never had more fun with a language. Docs are great and the whole language is small enough to be learned in about an hour.
I would be remis if after reading this I didn't point people towards talk box ( https://github.com/rich-iannone/talk-box) from one of the creators of great tables.
Adding fuel to the fire of "this is over engineering" but this is overkill right?! I'm not in the web development field but my own site is just deployed with Emacs (specifically HTML generated from org-mode).
My Ops brain says "Taken in vacuum, yes" However, if you make other things that are not static, put them into a container and run said container on a server, keeping the CI/CD process consistent makes absolute sense.
We run static sites at my company in containers for same reason. We have Kubernetes cluster with all DNS Updating, Cert Grabbing and Prometheus monitoring so we run static sites from nginx container.
I was going to write a response but you've put what I would have said perfectly. The problem, at least in academia, is the pressure to publish. There is very little incentive to write maintainable code and finalise a project to be something accessible to an end user. The goal is to come up with something new, publish and move on or develop the idea further. This alone is not enough reason not to partake in practices such as unit tests, containerisation and versatile code but most academic code is written by temporary "employees". PhD's a in a department for 3-4 years, Post Doc's are there about the same amount of time.
For someone to shake these bad practices, they need to fight an uphill battle and ultimately sacrifice their research time so that others will have an easier time understanding and using their codes. Another battle that people trying to write "good" code would need to fight is that a lot of academics aren't interested in programming and see coding as simply as means to an end to solve a specific problem.
Also, another bad practice few bad practices to add to the list:
* Not writing documentation.
* Copying, cutting, pasting and commenting out lines of code in lieu of version control.
* Not understanding the programming language their using and spending time solving problems that the language has a built in solution for.
This is at least based on my own experience as a PhD student in numerical methods working with Engineers, Physicists, Biologists and Mathematicians.
Sometimes I don’t blame people for committing the ‘sin’ of leaving commented code; unless you know that code used to exist in a previous version, it may well have never existed.
It can be very warranted. For a client I'm working with now I'll routinely comment out big swaths of code as they change their mind back and forth every month or so on certain things. They won't even remember it used to exist.
The key elements that I consistently see missing when comparing languages is the cost/effort of maintenance and how easy is to write fast/slow code.
The former boils down to, there's little point saving time in the writing of new code when new versions of the language hinder the use of old codes. Julia, being a very young language, simply does not have the legacy of backwards compatibility that Fortran has and as a result libraries written today are unlikely to work in 10 years time. Partly because of Julia's age and partly because backwards compatibility is not valued as much as it is in Fortran. The GNU Fortran compiler can still compile code written nearly 50 years ago (https://gcc.gnu.org/fortran/).
The second point is that benchmarks like this seem to focus on tuned implementations, which is really important, but another important aspect is how easy is it write fast code. With Fortran, it is very easy to write very fast code as a lot is abstracted away from the programmer so they can focus on FORmula TRANslation. My experience with Julia has not been the same and thus for people don't have the time to perform such optimisations, Fortran may still be the better choice.
I do look forward to watching the development of Julia but for now I'll stick with using Fortran as the high performance numerical language in my tech stack.
I'll finish with something I've read but I can't remember from where:
Over the years there have been many examples of "X with replace Fortran" and the only thing that ends up getting replaced is X.
The choice of 1 based array indexing is due to Julia having it's roots in mathematics - which is a common choice for mathematics based programming languages (e.g. Fortran, Julia, MATLAB/GNU Octave, Mathematica, R and Wolfram). The reason being is that matrix notation starts at 1 and not 0.
From personal experience, I wouldn't let something as simple as starting index getting in the way of programming for a language as you'll soon start automatically switching to thinking in 1 based and 0 based indexing depending on the programming language.
That is, the architectures of today are chosen to yield the best results given the textual data around today and the problems we want to solve today.
I'd argue that this lack of bias would need to be researched (if it hasn't been already) before this kind of model has credence.
LLMs aren't my area of expertise but during my PhD we were able to encode a lot of a priori knowledge through the design of neural network architectures.