the existence of AI summaries have made me realize I don't really want to know all this stuff anyway, at least before there was some pride and sense of achievement in the process of understanding something.

this is neat. i think digg should come back.

Funny enough, CECOT only exists because of this. MS-13 started in the United States, and only spread to El Salvador because of deportations, making El Salvador completely unlivable.

That is a very simple explanation to an obviously more complex issue.

You cannot discard the role of US Immigration & Deportation policy in the rise of MS13 gang.

Please read some books on the matter if you disagree. My recommendation is "Unforgetting: A Memoir of Family, Migration, Gangs, and Revolution in the Americas"

And is entirely correct.

You consented to making your content public explicitly.

Trying to understand your comment ...

Are you saying that when you make something, literally anything, available for the public to read, then you are ipso facto giving permission for people to copy it and do with it what they will?

the crawler isnt copying it, its using it to inform a model. yes, if you publish things publicly, people will consume it. you dont get to dictate how they do.

also using "enthusiastic consent" to make this argument is super cringe. drawing a connection to someone taking advantage of you because you are asleep is super bizarre.

Actually, people learn for hours this way. It just needs to be tempered with support. This is literally what a teacher does when they give you an assignment, give you feedback on it, then give you another activity. You're thinking of hard testing - where you get a lot of things wrong. I did research on how to keep learners engaged, and as long as they were getting over 50-60 precent of the problems/activities "right" they kept going - so we actually built an adaptive system that optimized for this type of growth, which means throwing them an easier question every now and then so that they have some confidence to go with the knowledge.

The reason some kids are not motivated to do math is because they believe they are bad at it, and nobody likes to do something that makes them feel dumb or not good. The kids that love math (like myself - I remember I was SO excited about math as a kid) are good at it, and teachers are constantly complimenting them.

You can hack the brain to feeling good while learning. I just need like 200M but I might be able to do it without.

So that adaptive system sounds really cool and useful for those bursts of formative assessment. And if the baseline is 100% boring textbook reading, ofcourse that is better. But if they work 9-5 every day, why don't schools do that at the moment do you think?

I don't mean to be negative, but can you name a research paper where a normal students 7 hour school day was changed to formative assessment all week long for a semester or more, and the grades improved? Or identify why students on Duolingo don't tend to learn as well as those taking in person classes? Why does the effect of the Testing Effect/formative assessment tail off when it is used more and over greater time periods?

The point is, the research on learning methods usually researches them in isolation and in 5-15-60 minute study sessions. Maybe doing 15 minutes of formative assessment improves your grades, but puts you off doing other work or makes you less effective in later lessons, when done regularly. There is a lack of useful longitudinal research in Education for the depressing reason that the two main effects are your genetic background, and private teacher attention.

I don't think kids are not motivated about math because they are bad at it. It doesn't seem like a subject that most kids see themselves using every day, because the adults they see around them don't seem to (and in fact largely don't) use Maths. While I value and enjoy Maths, and know how knowing maths makes a massive difference to your experience or life, kids can't, and really won't have good reasons to understand or be motivated through that. e.g. it's not a fear of negativity (though in class under pressure it is). It's an absence of motivation in the first place.

I absolutely love magic links. But I also love email, as a technology. I find it so reliable and robust, I genuinely think more things should be built on email.

As with all things, you need to know your audience. If you are making a product for people over 60, probably a simple username/password would work best.

can confirm. no matter what tool the org uses overall, there is an excel sheet somewhere that is the place stakeholders look.

It's a Google sheet now, but yes.

I'm not going to ask them why there's a spreadsheet based on a Kanban based on a bunch of tickets created in some scrum meeting which are finally added as tasks in an epic just to be rearranged later.

That's above my pay grade, but I question the efficiency of moving task titles between three systems, and never commenting on the tasks themselves. Feels like a lot of extra steps.

To answer the headline - No.

I find that working and adjusting my prompts and context is way higher value than A/B testing LLMs.

After all, I will never expect 100% accuracy.

I feel like they are reaching commodity status and the result quality is so similar, that it just doesn't really matter what you use.

+1

There are already many prompt/LLM routers available.

We've never found value in them, for similar reasons as mentioned above.

Thats great feedback - thank you! I would then assume most people would end up using open source models as they tend to be cheaper with faster inference

I can save you a bit of market research and tell you that’s unfortunately not the case yet in the market today. There are a few reasons for it - the main one in my opinion being that it’s hard to measure the value vs cost of switching to an open LLM, so it’s generally perceived as same/lower value, higher cost (not in terms of inference, but in terms of overhead). What is considered however are cost saving options around the foundational models: going for the mini models, prompt caching, batch inference etc. Some tooling in that area might be interesting.

We've actually found the opposite -- Every client project has been based on GPT4 or Gemini, with one exception for a highly sensitive use case based on Llama3.1.

The main reason is that the APIs represent an excellent cost / performance / complexity tradeoff.

Every project has relied primarily on the big models because the small models just aren't as capable in a business context.

We have found that Gpt4o is very fast, when that's necessary (often it's not), and it's also very cheap (gpt4o batch is ~96% cheaper than the original GPT4). And where cost is a concern and reasoning doesn't need to be as good as possible, gpt4o mini has been excellent too.

Thank you again - Great context for me

Thanks, do you think there is value in evals? / do you use evals while testing prompts?

Jumping in based on our experience in case it's helpful:

- Evals are very useful and a core part of the best practice for creating LLM apps

- There are already excellent solutions for model and prompt eval (etc.) including Parea from YC and many others

the incel to HN pipeline is now complete

that's not appropriate here

but sexism is? awesome.

My parents grew up in the Former Soviet Area so I will share this anecdote: Both of my parents studied Chemistry. Math is really important in our family. You are shamed if you don't understand something logical. My dad still had to do military service even if he studied. There was no strong incentive to study other than the fact that it was free and something to do for people who had no jobs, only 1 hour of TV per day, etc.

My hot take is that lack of entertainment and the fact that education was one of the only free things available to them was a large contributing factor.