hasteg

I agree that using AI as a "tutor" to help teach yourself things is very useful and works great, I've been doing that myself quite a bit. How do you stop students from being able to just use it as a second brain? I assume there's some sort of guardrails you have so it won't just give the answer? How do you ensure that there's an adequate amount of "struggle" that the kids are going through before just explaining the answer? I feel that that would be the challenging piece of the puzzle.

This is also how I pretty much filter LLM generated text in my head.

Using AI to actually learn is indeed very helpful I've found... I just don't think most people use it for that. I recently wanted to build and train my own (albeit, small and simple 10M param model) and used AI extensively to explain concepts for me, explain lines of code, and generate in depth visuals I can use to get a visual intuitive understanding of what is happening under the hood. I think people who have a natural curiosity to understand the why and how of things benefit immensely from it, but I do admit it is very easy to just offload all thinking to it instead of asking it why something is happening. I could have just asked it to implement the entire model using PyTorch and just ran the CLI command to begin training on some auto-generated dataset, but intentionally struggled through it. Actually learning from it requires intentionality for sure.

I just recently watched some (not all) of this video "coding a machine learning library in c from scratch" and seems like he's going through a similar process in this blog as this video. I would recommend watching the video to get an idea of what the fundamentals of a ML library look like. From someone who has recently been getting interested in actually writing ML code and trying to make sense of it myself (from the perspective of just a typical backend engineer) it was very interesting to see. Previously my experience with ML libs (PyTorch specific) was writing my own Mini-GPT and training it on a small dataset using my own GPU (5090). Cool to see the behind the scenes and took away some o the handwaveyness... https://www.youtube.com/watch?v=hL_n_GljC0I

Yeah, clear example of survivorship bias... not to sound negative, but if Facebook had become nothing, odds are Boz would be nothing more than another engineer. It's all too common a tale to hear people spout their nonsense and think they got to where they are because of some inherent talent or skill, but at the end of the day, getting to somewhere like Boz these days is just luck. Does anyone really think you could start as an L4 at something like Amazon these days and end up as a CTO? Nope.

Same here, I hate it. Instead of just reading I find myself also dedicating brain power trying to decide if it's worth reading or not based on the first few sentences... if someone can't put in the effort to actually write something themselves I absolutely do not want to put in the effort to read it.

LinkedIn is basically unusable at this point. I actually did used to use it a fair amount before but I've since deleted it and just use email notifications to see any notifications from recruiters.

What I don't get is how these people don't feel shame in their super obvious blatant use of LLMs for everything, even responding to posts. Maybe it's just me but when I'm attaching things to my name like that, I would absolutely not want everything to be obviously slop shit. Do they think people can't tell or something? I know at least every technical person I know can immediately tell (most of the time) when writing is LLM generated.

I mean, this is perhaps the most vibecoded website possible so... deff AI.

Can you elaborate on that more? Why just a filler character?

Really interesting paper, thanks for the share.

The point their making in that paper reminds me of this paper some people shared around work earlier this year, https://arxiv.org/pdf/2512.14982 (Prompt Repetition Improves Non-Reasoning LLMs)... I wonder how OPs question would fare (or the questions presented in the paper you posted) given double repetition.

Haha that's a great intro to it. I remember very clearly I was obsessed with minecraft griefing/team Avolition at the time and found it to ask if someone could make me a hacked minecraft client for my first thread (yikes). Wasn't a great intro to the community for me but ended up staying after that. Also was very significant to my computing career, glad I was on it. I think that era of the internet was very special and I miss it dearly. Also seems like back in 2015 or something someone somehow got into my account and Omni deleted it. Sad.

Anyone here used to be on HackForums? I lived on there from 2011-2013/4... was pretty young back then but I learned so much and attribute it to getting to the spot i'm at now. I remember being so obsessed about "rep" on there, at one point I hit over 1k and it was like a huge achievement for me lol. Seems like that was the precursor to "followers" on X/Insta/whatever. Looking back, it wasn't the best place for a kid to be either, but seems I turned out fine. I really do miss forums and I have found myself looking for a new community like that to join lately.

Typically when I ship anything that is going to be customer facing I am a nervous wreck. Although I'm only 4 years into my career. Does it get better with time?

When I was studying my CS degree (2018-2022) it was actually encouraged to work together outside of classes with fellow students. We had a CS discord run by the school (or maybe volunteers from the school? I dunno) with diff channels for different classes. I also worked a ton on studying with my friend who was in a lot of the same classes as me. I think that was insanely helpful, especially for a lot of the theory heavy classes it was great to have someone to bounce ideas off.

"On paper, it looks like a smart decision. Reduce paper, reduce staff, reduce operating costs. But what gets overlooked is the hit to the customer experience." As always.... this philosophy is basically killing any customer experiences these days. Hopefully some day profits will take enough of a hit to actually start resulting in more effort in the customer experience.

There are some things that I feel really shouldn't be enshitified by AI and this is one of them. Sad world TBH.

Lol yeah exactly, if kids want to go on social media they will 100% find a way around it.... putting in these identity requirements will literally do nothing except require us as adults to give over more information. I'm sure everything will pass through congress fine though because "someone think of the children!!!!".

I agree, I think kids should have limited access to the internet. I pretty much did and it worked out for me but I have seen so many reports about it causing harm in schools and personal life. (Specifically I think LLMs should not be used in education also, but different point) However, I think the main problem people have with this "think of the children" narrative is that it will force EVERYONE to give up their credentials to access the internet, not just kids. And the general consensus is that we as adults do not want to and should not have to prove our identity to access the internet.

So far this is some of the best uses of ML I've seen to date! This is one of the few things you can point at and say "AI made the world a better place" IMO (this and medical research).

Because the transformer, which all of these models are foundationally built off of and didn't invent themselves (bar google) wasn't invented? The amount of effort it took humanity to generate all the data that was required for the models to get to the point they're at now is absolutely not even comparable to how much effort it took to build the model code. Yeah, it's complicated, but if they didn't rip off all of humanities combined output it wouldn't even matter if the transformer got invented.

Just a heads up (you may already know this) -- In the local LLM community there's a pretty strong disdain for Ollama specifically due to numerous reasons provided in this blog post: https://sleepingrobots.com/dreams/stop-using-ollama/

What are some examples of features that they have added since December that you use? I originally had setup openclaw back in Jan(?) and had it generating some news summaries for me and stuff. But ran out of ideas... would like to try it out again.

Agreed, I went through the blog post and a few other pages trying to figure out what the benefit of Iroh is since I have never heard of it... was struggling lol.

Earlier this year I had used it because I would rather have a IDE-like exp and be able to actually look at the code. However, recently switched to using claude code VS code extension and it's basically the same thing (plus at Amazon we can only use Claude Code)

Here is a one page proof that pi is irrational - https://heuklyd.github.io/papers/pdf/Niven-1947.pdf

Wow that's actually sick as hell, somehow hadn't heard of this. maybe I will go and blow $700 on a new ram kit... thanks for sharing!

Wait so this makes it so I can use my DDR5 as well as my VRAM combined? This is actually sick if so. Maybe I will actually have to go out and buy some more DDR5 (currently only have 32GB...)

This seems really cool actually. Just read thru the README and watched the demo gif. Any examples of what you've built with it so far? Might have to play around with it tonight. I've been getting pretty heavy into local LLMs since getting my 5090. Amazing what Qwen can do given how small it is (running 35B at Q4 quant).

IIRC when they make a big enough architecture change to the model they will need to rerun pre training . So not like they’re feeding it more data (they will be but will be a drop in an s3 bucket compared to their dataset reserves) but rather training models with different architectures.

Lol ++. Although my local host for agent/codex stuff is a raspberry pi I connect to on LAN from my gaming/powerful desktop for sandboxing. However my use case seems to be the exact problem they are trying to solve! Might have to take a look into it at some point.