News for hackers. Stuff that matters.

walpurginacht

i'm enjoying langgraph the most, it's got nice integration with langfuse for observability and langchain (despite its shortcomings) with its plethora of integrations makes things easier to do, and the way to define the graph and transitions is pretty intuitive

currently trying out pydantic's own graph based approach to use together with dspy

Re:Ask HN: Relatively SoTA LLM Agents from Scratch? on Friday December 12, 2025 @05:21AM (#46242690)
I'd suggest you take a read on HuggingFace's writeup when they trained smolLM3

https://huggingface.co/spaces/HuggingFaceTB/smol-training-pl...

rare detailed insight on the entire process

Re:Ask HN: Battling Depression on Thursday December 04, 2025 @02:35PM (#46151812)
I think it's outright disrespectful to just assume people in such circumstances are just looking for external factors to step out of responsibility. Have you ever thought that they do want to make a move, but just cant because their minds are broken?

you cant apply the "you actually just have to step up and get out of your comfort zone" advice to people whose minds are severely altered by their condition. You fix the condition first, or give a remedy, then when they start to get a grip on things you move to making changes to get them out of the situation

thank god you only speak for yourself, please keep it that way

do you have an evaluation in place that necessitates complex stuffs? If not I'd start simple with proven stuffs and collect usage data to determine what's next
Re:Otary now includes 17 image binarization methods on Saturday October 11, 2025 @05:23PM (#45552824)
Love this, gonna take a look when I get home for some weekend hacking

in your experience, which binarization (adaptive or otherwise) do you often practically use?

Re:Ask HN: Do you roll your own agent or use a framework? on Friday October 10, 2025 @12:39AM (#45535420)
used langchain and churned out of it due to it's abstraction level. Nowaday I either just use pydantic-ai or dspy.
Re:Dead Internet Theory on Tuesday September 23, 2025 @10:09PM (#45355486)
opened twitter (now known as X), it's the same automated slop posted over and over again with a singular goal of farming engagements

opened facebook, my favourite groups are now filled with automated reposters

opened youtube, the glaring content farm videos following an obvious template with the same art style narrated by an obvious AI tts greet my feed

it is as if incentifying any sort of engagements through rewards, be it adsense money or just clout, directly correlates with the amount of automated braindead slops

i wish all other sites have the feature to just hide AI-generated content. Future content moderation is going to be insane

Yeah, what was once a minor incovenience has now become amplified, really sad
It sucks so bad. Before I've caught covid I barely get sick, or if so just a night of rest will bring me back to health. Now just a common cold lasts minimum a week

Not to mention me being unable to reach the levels of aerobic fitness and mental clarity I've had before I got covid

Re:Ask HN: Does anyone else pronounce CLI as "clee"? on Wednesday September 10, 2025 @06:03AM (#45195495)
I pronounce it as cee ell eye but from now on it's clee I guess
Re:AI hype is crashing into reality. Stay calm on Monday September 08, 2025 @09:16AM (#45167880)
to be honest with the level of hype it's been sustaining, it's about damn time people get disappointed with unrealistic expectations they set (or let others set) for themselves. All the while people who merely view it as a fancy tool with all of its shortcomings, and treat it as such in building something with it, will keep finding benefits from it

With this, hopefully we can stop having peope using AI as a term exclusively for llms

We're about to, or have already hit the "data" wall recently as it gets increasingly hard to get more data to train on, exacerbated by more and more content being generated that risks autophagy

it's less and less about gathering more data now and either we're gonna try to engineer more data into existence, or it's back to the deep learning boom where it's more common to resort to optimizing architecture and training methodologies to squeeze out more performance from what data we have

Re:Show HN: Dumb Site to Rate Horses on Sunday September 07, 2025 @03:32AM (#45156159)
Reminds me of facemash
Re:Ask HN: When was the last time you visited Stack Overflow? on Saturday September 06, 2025 @08:08AM (#45148566)
They really have the reputation of being way too hostile and gatekeepish, to the point they even made a post about it

https://stackoverflow.blog/2018/04/26/stack-overflow-isnt-ve...

Re:Ask HN: When was the last time you visited Stack Overflow? on Saturday September 06, 2025 @08:04AM (#45148544)
couple of hours ago. I still use it from time to time, after I used ai assistants to help me narrow down what is it that im specifically tryin to solve. Sure claude/deepseek/chatgpt/grok/whatever assistant you use can do that for you, but I enjoy reading the discussion part of stackoverflow when it's not just a bunch of stupid debate of semantics / people being too pissy about their "rules" and whatnot