"Only 0.5% of doorways are used to leave with bank money"
Look, I only need one doorway, I'm not coming back to prove all your doorways are usable. If you cover the obvious ones I'm going to use the less obvious ones.
That the "batteries included" bit, but it's not "burning tokens" in the sense that using Pi with the same batteries would (of course) use just as many. I can strongly recommend omp.sh if you just want to be up and running without hassle and a far better (for me) developer experience.
Yeah, saying it kills the minimalist bit misses the point of omp being a separate project. It took a minimalist-base of pi and added tools that make it usable out of the box. Hence batteries.
One example here is that it added MCP extensions to make the harness usable with MCP servers long before Pi figured out that that was the right move. A lot can be said about development of MCP over that timeframe, but I appreciated the bias towards adoption, since progress is driven by empirical experimentation rather than theory in this domain right now.
I've long championed assembling developer tools yourself from scratch and I've been an advocate of vanilla Emacs and customizing it yourself as an ideal developer experience (I realize this opinion is controversial). But OMP is essentially the VSCode to pi's Emacs.
I ended up choosing omp as my daily driver, mostly because I see the velocity of change in AI development being far too fast for me to make a worthwhile investment into customizing the environment myself, since that investment will have a relatively short half-life. With Linux, which I learned more than 30 years ago and still believe to be a fantastic investment, I can largely use the same knowledge I gained then to administer systems today. I fear that much investment in the specific tools that I add to my pi harness will expire before 2027. So I'm sort of drafting OMP until things settle down.
Yeah I just gave it a run and it starts off burning 6% of my tokens on start.
Pi sits at 0%.
I'm so use to the command keys in Pi that having to type something out in OMP slows me down. I'll stick with Pi, I've built it to my needs and having WSL finally working. I still have to run 2 CLIs, one for LLama.ccp and the other for Pi, not sure if this is normal.
OMP provides a significant amount of structure to the work. That costs some tokens in terms of the system prompt. When I use these systems, I balance the return of the extra guidance in the system prompt with the cost of the context. You've talked about how much context it takes, but you haven't talked about any difference in behavior.
omp offers some nice tools, like /shake, to manage context efficiently and offset some of that consumption.
Not sure about your point with command keys. OMP has all kinds of shortcuts and is highly configurable. It seems very unlikely that you cannot achieve what you're looking for in OMP in terms of shortcuts. Specifics would be immensely helpful here.
And yeah, you'll need to run your inference engine separately from your harness, for the same reason the web server and the web browser are different applications.
I know that's the way omp is often introduced but I don't think it's fair for either omp or pi - after lately switching my usage more towards omp I'd say by this point they're almost more different than Pi and Codex. And I like them both.
I am wondering what use people get out of this. What I like about pi is that, using proprietary harnesses to begin with, I already somewhat know what kind of features I use and features I wish I could use. Asking pi to self modify itself (with a smart enough model) is a one time cost for a fully vetted feature that I personally approve. Do users of OMP find all of its additions useful?
Yeah, Advisor and "time-travelling rules" (TTSR) work great (they're injecting corrective prompts automatically either via another critic LLM or a bunch of regex rules, respectively)
It automates a lot of the babysitting of agents, and lets me use a couple of smaller models with safeguards instead of burning tokens on a galaxy brain just because it listens to instructions in AGENTS.md.
Blocking slop code, like `unwrap_or_default()` in Rust that agents use to weasel out of error handling.
/omfg automatically adds new rules, so I have plenty of project-specific enforcement (e.g. disallowing making certain fields public, which agents love to do to take a shortcut that ruins the architecture).
It’s the same kind of people that would use omz, they probably have no idea what 90% of the batteries are or do. But it sure makes them feel smart and cool.
Like other comments, I can't help feel uncomfortable with the way everything was written defensively, where every example is used to justify the author's actions.
The author concludes near the end:
> That said, I have no judgment to pass on you here.
And follows with this as the very next sentence:
> We are all human beings, we all make poor choices and mistakes in our lives.
Which seems to be clearly judging the students for their "poor choices and mistakes".
I feel the author is out of touch with the students and the times; If so many of them are leveraging AI perhaps the better move is to find ways to align that usage with learning. He quotes:
> "it is better 100 guilty Persons should escape than that one innocent Person should suffer"
But if there is such a scenario in a classroom rather than a courtroom, then perhaps the methodologies should be revisited?
It feels like the author is the ACKCHYUALLY meme, striving to prove themselves technically right, rather than empathizing with the students, which I feel shows a lack of maturity for an educator more than 2 decades older than his students.
I understand that education and evaluation is difficult in this AI era, but I can't come to agree with the author's actions or conclusions in this case.
With regard to judgment, there is a difference between judging a person's character and judging their actions. That was the point there. The actions clearly violated course policy.
I am curious what leads you to believe that there's a desire for someone to agree with the actions when the retrospective opens by saying the situation should have been addressed better.
Assuming that the students were using generative AI when told not to...
The way you leverage AI for learning is to do the work yourself and ask it to review and provide feedback, or to ask it to explain some concept, ask it questions to check your understanding, ask it to generate questions for you to answer, etc. If the entire point of the exercise is the exercise, then generating the answer... misses the point.
Like I can easily find online how to encode if statements in the lambda calculus, or how to make a half-adder out of logic gates, or how to prove that bijections are invertible. It's always been cheating to not do it yourself. School intentionally gives you easy problems with known solutions, not open research problems where you should use any and all tools and information you can find.
> If so many of them are leveraging AI perhaps the better move is to find ways to align that usage with learning
Should the course be updated to better leverage AI? Sure.
But as a student, if you're explicitly instructed not to use AI for a particular course/assignment and you do it anyway, that's academic dishonesty. No matter how much you disagree with the principle of the thing, you still lied and cheated.
Absolutely crazy and useless article IMO in that they never mention AI.
They cite readers being interested in learning - well what's the best way to do so nowadays? Ask ChatGPT.
Even worse, you can ask ChatGPT to describe the learnings of any reasonably popular nonfiction book directly. No it's not a perfect substitute, nor will the learnings be as memorable, but it's reasonably good enough. That's why I haven't bought a nonfiction book in a long time.
Imo Eventually we will circle back to nonfiction quality books, for one just to decrease screen time :) for second we actually need them even for chatgpt
I think it's just that in the past, you would need X free time, Y motivation level, and Z skill level (partly tradeable for X and Y) to create something like this, and now all of those values are smaller. So it's 'economics' in the straightforward sense, because not everyone could afford to spend X hours on something they weren't going to monetize, and also in the sense of the incentives shaping people's productive/creative activity.
(As for whether it's a good thing, I'm conflicted. It's cool to see more stuff like this, but if it crowds out the more intense passion projects, we lose something too.)
I believe that part serves to emphasize the main point.
To elucidate for this particular example, before AI agents existed, projects with this level of data and depth might've needed to be monetized in order to justify the resources to bring it to life.
Now that the barrier is much lower, the creator didn't mind spinning up and providing it with no monetization (e.g. no ads).
> before AI agents existed, projects with this level of data and depth might've needed to be monetized in order to justify the resources to bring it to life.
The site was registered a month ago. If it ever actually gets and stays popular, see if it has no ads then. The economics of personal interest sites didn't change in the sense that they're any harder to make now compared to the mid-90s. The economics that changed is the Internet now has billions of users and attention being monetized can make you a lot richer now than then. It's very hard for people to see an opportunity to become a multi-millionaire by clicking a button but then not do it because they'd rather maintain a slightly better experience for their audience.
It's, of course, not even loading right now, which gets at the other problem with this "economics" framing. The biggest expense in running an actually popular website isn't creating and publishing the first version. It's paying for the infrastructure to serve a meaningful proportion of the entire world. This is clearly not a static site. It's presumably doing some computation on the server side to draw from all these hundreds of different probability distributions that works well enough for a few dozen simultaneous requests but not thousands. There might even be some kind of perverse anti-economics going on here where thought devoted upfront to efficiency and architecture might have made this far more scalable, but the "publish the first thing that passes a basic sanity check" vibe-slop ethos doesn't usually get you that.
Vibe slop? The structure of the site is quite solid, the UX decisions are spot on, the data looks good to me that knows little about this stuff, and it is entertaining and informative. This was not a one shot, a human out a lot of effort into this.
Yeah it is imperfect and was maybe not ready for the HN front page hug. But I say, bravo to the person or team!
> the data looks good to me that knows little about this stuff
I majored in history. I'd say the data is misleading. I drew a boy "In the Ganges Plain, 1411 BCE". According to the summed up data he had a 54% risk of dying of neonatal tetanus, sepsis or dysentery. Which isn't wrong as such, but it would probably have been better to sum it up with all the other diseases, parasites and infections that would be needed to reach the 54%. Like resporatory infections... "Other causes" is listed at 14%, I assume that malnutrition is part of this, but malnutrition would be the leading cause in the 4-8 year age group, and less so in others.
The life stories themselves are over dramatic. Which isn't weird, considering you don't have a lot of source material dealing with the "boring" lives. Like if you think of Scandinavia in the VIKING ages you'll find a lot of research and data on the people who went viking. In reality 95% of the populations were farmers who didn't even see the coast in their lifetimes. Unless the AI is specifically tuned to value the few sources on everyday farmin in Scandinavia higher than the ones published on viking culture in general... Well.. I think you get it.
That being said. Many people probably don't know that you could have 18 children at only see one of them grow to the age of 20 a few hundred years ago. So the data is accurate enough to be educational in my opinion.
I do think the "typical marriage age" might be suffering from AI safe-guards. For most times I've tried it seems to land oddly close to 18. Which is unfortunately much higher than reality.
I live in Denmark. So yes. I have seen a map of Scandinavia. People didn't travel. Even here in Denmark which is tiny compared to Norway and Sweden, if you only ever go in a 10-15km radius of your home then there is a good chance you won't see the ocean.
This is true. I concede. However I am not sure that the majority of people were as imobile as you say even in mideveal times there were religious pilgrimages, fairs. Maybe the lowest stratum of non free society lived in a land locked mannor and never saw the sea in Denmark but I wonder if it was the majority of people in a lifetime?
It's possible, but not easy. You need to export the internal state of your TLS stack to Wireshark in order for it to decrypt the traffic. This can be done via LD_PRELOAD type shenanigans but it's difficult and fragile.
Look, I only need one doorway, I'm not coming back to prove all your doorways are usable. If you cover the obvious ones I'm going to use the less obvious ones.
What a silly article.
reply