Hacker Newsnew | past | comments | ask | show | jobs | submit | clusterhacks's commentslogin

Even pre-AI, my experience as an adjunct led me to believe there was a lot of cheating on homework and assignments.

Individual programming projects for students wrecked me as a lecturer. The time investment for me for meetings, reviews, and understanding 30+ projects well enough to grade against a higher-level, abstract rubric just took too much time.

Are you happy with the MS of Ed program you are in? I sometimes think about making the same transition to high school math teacher. I find that I do very well with engaged high school students (I volunteer in various ways at my local high school) but haven't quite found my footing with students who (maybe rationally) don't engage with the material.


Yes, I do well with engaged students too. I haven't found out how to manage students who don't stay engaged, they all have some sort of hidden trauma, maybe someone told them they were bad at math (more mild form of trauma), to more extreme forms of trauma related to family relationships and the such. It feels like being a teacher is learning now to support the students that have been through the trauma, and how to create a safe space in the classroom (so they feel okay saying they don't know something and are okay with mistaking mistakes), which isn't easy and I'm still learning how to do this.

...especially with megakernels...

I don't understand why megakernels haven't become the default choice. The speed advantage is crazy. I made a few changes to the openjet megakernel to run in a dual gpu (older gen, not blackwell) box and doubled my token generation speed on qwen3.8-27B (4-bit though).

I keep debating adding a third matching gpu to see if I can push the speed a bit more and move back to the 8-bit quant, but I'm stuck on the idea of just building a Blackwell gpu system despite the cost factor. Plus a third gpu is going to going to be weird for the tensor split and P2P data syncs and my workstation is old with 3rd gen PCIe etc etc.


Architectures (like Gemma4) have things like sliding window attention that make it pretty impractical to do on a GPU, you would have to do additional things to make it work, and the talent to do that, even with LLM assistance is very small.


Ah, that makes sense. I haven't bothered to look at current low-level NN architecture details. I'm a little surprised that there isn't a bumper crop of relatively well-informed talent to do gpgpu work. Although, when if I think about it, it has always been a little bit of a niche area.


I have experience as a coach in middle and high school in the US. I can say emphatically that school sports are relatively low impact in terms of time opportunity cost and individual family spending. The "club" or "travel" team private organizations are where the money and time really go.

I am also a parent of a kid who participated with a soccer club team for many years. She is a multi-sport athlete who loved several sports and we found the club experience in soccer actually quite good for her emotional and physical health. I will say that family experience with club sports varies based on the quality of the organization. She also played with her club in falls and her other sport in the spring at high school, so she got to do both. This was fun for her and us.

Assume a kid just wants to get better at some sport. I'm a player development nerd and believe most kids would be better served with 1-1 training with a much lower time investment that just joining a club or travel team. I do 1-1 sessions with players for 30 minutes 2x/week for 6-8 weeks in the fall. Players, of all natural athletic levels, will make about two years of progress compared to just participating on the spring team. That's with one whole hour per week of time invested. I love working with players this way because I am measurable-metric driven, so a kid can see their progress. I'm also a little bit qualitative about it too, because sometimes, the goal is just to build confidence and comfort.

The approach I'm describing looks like lots of hobbies. Imagine a kid wants to learn piano or guitar or drums. It's basically the same abstract approach. Work with an experienced teacher, commit to a regular+reasonable practice schedule, repeat for 6-8 weeks, and see if it fits and feels good.

The social aspect of being a good club or travel team is nuanced, but I will just say that my kid was incredibly well-served by having a good group of friends not connected to her school. And she had a great group of friends in school.


Running locally for me is mainly about learning, maintaining control+privacy, and helping shift my coding+design process to leverage LLMs. I guess if you made me boil it down to a single word to justify the cost I would just say: tuition.

Sounds like we have similar boxes - mine has a 10 core CPU, 64 GB of ram, and a 2070 Super. My motherboard had two unused PCIe3x8 slots and doesn't support Blackwell GPUs. I bought a couple of brand new Ada generation RTX 2000s with 16GB of memory for under $1400 to get to 40 GB of VRAM. That will easily run Qwen3.6-27b at a 6-bit quantization and 80,000 token context size. It isn't fast (19-21 t/s), but using pi-coding-agent is fine.

Now, my instinct is that I am giving up SOTA performance on agentic coding with this setup and LLM. But the gap between my setup and SOTA commercial models is small enough that it doesn't matter to me.


On the one hand, sure, why not have a default install throw a bunch of bells+whistles via skills and extensions.

But I like pi precisely because it is so minimal. I want understand and work around the simplest possible agentic coding setup, find the sharp edges, maybe even improve my prompting ability. And doing all three with a locally hosted LLM.

At some point, if I don't understand the foundations, am I just punting on actually thinking about what I'm doing?

Of course, making individual choices about how to do agentic coding are precisely just making individual choices. People should do what makes them happy and productive.


This may be too naive, but I created a user on my linux box who doesn't have very many permissions. Then I sudo to that user, use firejail to start pi in a dev project directory, and let it have at it.

My projects are usually very limited with respect to external dependencies and that is part of prompts or markdown files describing various project goals, plans, and current state.

My operating theory is that this probably won't get my systems borked. I wasn't patient enough to dig deeper.


FTA's conclusion:

"Is this decline a distinct change from the recent behavior of the labor share in the U.S.? Along the two key dimensions we investigate, our answer is no. <later> ... and they provide little evidence that it will evolve differently from past episodes."

This conclusion seems to be against "this time is different" arguments. Should we be generally encouraged by similarity to past declines pre-2000 or bearish and think that there is more drop to come like the 2000-2007 and 2007-2019 periods they graph out?

I guess there is no way to predict other than check back in after time passes.


> I guess there is no way to predict other than check back in after time passes.

Welcome to the dismal science of economics, where the rear-view mirror is crystal clear but the windshield is totally fogged up.


Bearish, they are already working as hard as they can to replace everyone with AI or at least Actual Indians playing AI


Huh. Same problem, and I run with llama.cpp. In my case, Gemma4-31B (4-bit quant though) will just stop sometimes.


I used to agree with you but now do not. I now think the floor for this market is probably no worse than the annual revenue of cell phone plans in the US market. So say, $250 billion.

Now, that probably doesn't justify the valuations and hype being thrown around, but I think it gets at a real revenue number.

I also don't know how that number fits into the funding rounds already raised and VC dreams of IPOs for these two.

This isn't coming from deep analysis on a verifiable source, but I started asking people in my social circle (includes white-collar and blue-collar folks) about their LLM use. The biggest surprise in 2026 for me was that almost all of these people told me about regular (and sometimes sophisticated) use.

A more intriguing observation - I work on the side with high school students and have two college kids of my own. Their LLM usage (and their peers) is much, much lower than expected . . . that's a little counterintuitive given "popular" perceptions I read.


> I used to agree with you but now do not. I now think the floor for this market is probably no worse than the annual revenue of cell phone plans in the US market. So say, $250 billion.

I don't think we're talking about the same thing. I'm talking about what their IPO is going to do to their share price.

In any case, $250b revenue translates to, best case scenario, $50b profit. On an investment of $1t. It does not look good for those companies making up the $1t investment.


Gotcha. I'm past the point of having any confident thoughts about what happens to their share price at IPO.

What about the idea that there is a high likelihood that the potential share price for OpenAI and Anthropic are both going to be pretty divorced from a rational market price for either?


Interesting idea and reasonable number, but cell phones need a lot of infrastructure and they need interconnection. The risk here is that in the future a combination of near-sota open weights models optimised to use as little resources as possible and a reasonable drop in compute price, will make possible for small and tiny providers to compete with Anthropic/ OpenAI or even for people to run their own private models for most applications. Then large, expensive sota models would only be used for research and to answer the small subset of general user queries that need that kind of intelligence.


--what this means for the valuation of the AI companies

Probably nothing. Most users have no idea what an LLM is or how it runs. Anecdotally speaking, I see many LLM users default to whatever their day job provides to them. And even slightly more sophisticated users seem ok with paying for their openai or anthropic subscriptions.

Maybe we will see a small but dedicated group of open weight model users who prefer local llm, but everybody else will just consume from the big providers? The scenario might look something like OS choices today - a small, committed group of Linux users vs the vast majority of other users running Windows, MacOS, or Chrome?


Prices from OpenAI and Anthropic have really jumped in the past month. I work for a big giant company and our Github co-pilot costs increased as of today, June 1st. Our internal estimates are that our bill will double or triple. How much are we willing to pay? I don't know, but nobody wants to be "left behind".

I think there's actually a big market opportunity here. Somebody, like Dell or HP, should start selling turnkey on-prem LLM servers.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: