AlexC04

The analogy meanders a bit in getting to the point, but I generally enjoyed the journey of the read and also, I think this is starting to look like an increasingly important perspective.

Increasingly I've been asking the LLM to do things that are only seconds of work to do manually on my own. Earlier in the process of learning how to use these things I think that was a somewhat reasonable decision in the name of learning how these things work.

Now... It's maybe the case that getting a bit more practice in with keeping on top of how we do things the manual way ... might still have value? both for the token cost and perhaps for the value that gets unlocked through understanding your systems a little better?

I'm also a little torn since the leaders in the field of "tokens for everything" like Cherny and Karpathy are pretty vocal about not writing any code on their own anymore. While they also don't really pay for tokens the way the rest of us do (I'm out of pocket for mine) is there something to be said for joining that frontier wave of developers?

One other thing you might want to check out for running locally. (I have not independently verified yet, it's on the TODO list though)

https://docs.vllm.ai/en/latest/api/vllm/model_executor/layer...

vLLM apparently already has an implementation of turboquant available - which is said to losslessly reduce the memory footprint required by 6x and improve inference speed by 8x.

From what I understand, the steps are:

1. launch vLLM 2. execute a vLLM configure command like "use kv-turboquant for model xyz" 3. that's it

I've got two kids under 8 years old, a full time job, and a developer-tools project that takes like 105% of my mental interests... so there's been a bit of a challenge finding the time to swap from ollama to vLLM in order to find out if that is true.

SO buyer beware :D - and also - if anyone tries it, please let me know if it is worth the time to try it!

oooh! I really like your graph! https://github.com/adrq/agentbeacon/blob/main/docs/screensho... I'll have to pok in and see what you're using for that!

this is pretty interesting - just from the screenshots it looks like an electron app?

have you or had you considered pursuing a vscode extension? (and by extension cursor?)

In my free time I'm also building out a set of agentic tools and I've found that there's so much you get "for free" by working in the space where developers already are.

Git integrations, text editing, highlighting, etc...

Now - I'm still in pre-alpha local development only so who knows what my adoption story will look like after I actually get it onto the vscode marketplace, but I dunno - just making converstaion I guess :D

Anyways, this really seems like an interesting project and I'll take a deeper look outside of office hours! Thanks for sharing.

oooh ... I wonder if it would be sufficient to just write a markdown gramar for TS?! I should ask my AI what it thinks.. I'm sure it'll tell me I'm absoloutely right and a very good and smart boy.

I first leanred about tree sitter a couple months back when I started looking at what was inside the NPM fodler for claude. It's a really cool library.

One of the things it made me think about is whether it made sense for using when editing large markdown files would it be more efficient to convert a document form markdown to DOM then back again for the purposes of editing a large markdown file via code agents? (or a json)

The theory being that agents are always asking me for pemission to use sed in bash to edit markdown files -- could tree-sitter do the same thing using its code-editing capabilities? And would that difference be materially impactful? Could I lower the token cost of writing an extensive plan by choosing a format that allows me to use tree sitter?

I really haven't explored that much yet since I've been working on other things but it was more just one of those things that make you go hmmmmm... maybe someone else knows :)

yeah - I thought the first half of the article (up until "introducing ace") was actually really great thoguh.

this is really good. it aligns closely with what I've been writing on the same subject (but is way better :) )

nice share. thanks.

OH! I commented before I realized it was a product pitch ... literal seconds before she introduced ACE.

If I'm honest, the idea of a collaborative agentic coding interface is pretty novel and interesting. Still impressed with the article just the nature of my appreciation has changed.

but how does it perform on pelican riding a bicycle bench? why are they hiding the truth?!

(edit: I hope this is an obvious joke. less facetiously these are pretty jaw dropping numbers)

to directly answer this bit:

> Feels like a fundamental bottleneck for production agent systems, so would love to compare how you're thinking about the latency vs accuracy tradeoff.

I'm really not focusing on latency right now. My short term goal is to prove the thesis that `ail` can improve same-model performance on SWEBench Pro vs. their own published results.

Can I run swebp with GLM-4.6 and get a score better than their published `68.20` https://www.swebench.com/?

The argument is that the latency right now just isn't the part we should worry about. If we're reducing the time to code something from ~6 weeks to 1 hour... then does it really matter tha we add an other 30 minutes of tool calls if we get it 100% right vs. 80% right?

Make it work -> Make it right -> make it fast.

I'm still on the first one tbh :rofl-emoji:

so - my approach is still being built and I'm still very hand wavy around how it is going to come together, but effecively I'm building pipelines of prompts. Rather than running our LLM sequences as long running sessions where the entire context gets loaded on every turn (a recipe for rot), we unlock the ability to introduce a thinking layer at each step in between the process.

So before each turn is sent into the LLM we (potentially) run a local process to assemble a bespoke context of only what is required for that specific turn.

If a tool call is not going to be needed on the prompt, we don't include it in the system prompt on that round.

I'm still formalizing the spec at the moment and think I'm about six months to a year out before I have a full human ready UI running.

This is the foundational paper I'm basing the tool on: https://github.com/AlexChesser/ail/blob/main/docs/blog/the-y... while the spec starts here: https://github.com/AlexChesser/ail/blob/main/spec/core/s01-p...

Essentially I'm trying to build an artificial neocortex and frontal lobe to provide a complete layer of Executive Function that operates on top of our agents - like Claude Code (or whatever else).

I'm basing the roadmap on the about 100 years of cognitive science. We've legitimately had names for all these failure modes (in humans) since the 1960's. We have observations of what we're witnessing in agents from 1848.

We have the roadmap from Psychology.

this is a pretty important piece and the research backs you up. Moving that context out of your system prompt dynamically is going to help reduce your lost in the middle effect. Context rots almost immediately. I've got a project that is being built to address this directly as well, but I'm still very early days.

Keep it up! you're on the right track.

Hong, K., & Chroma Research Team. (2025). Context rot: How increasing input tokens impacts LLM performance. Chroma Research. https://research.trychroma.com/context-rot

Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P. (2024). Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12, 157–173. https://doi.org/10.1162/tacl_a_00638

I looked at the JS over here http://codebase.es/riffwave/riffwave.js and find this line

     this.dataURI = 'data:audio/wav;base64,'+FastBase64.Encode(this.wav);

(edit... more accurately the whole RIFFWAVE function blows my mind)

Really quite revolutionary (in terms of my understanding)... I already knew about making JPEGS etc, so this makes perfect sense but just seeing him build all the headers for a wav file in memory then "spit it out" to the page is stunning.

Kudos, Bravo, Standing slow clap, etc...

Thanks for sharing.

I remember the time you helped me with some canvas stuff. Thanks again. 12K is well deserved :)

Could this ban not be bypassed if people knew the IP address megaupload was hosted on?

How was it taken down? Just DNS or soemthing else?

yeah - that was way too small for me. I had to zoom a lot. For a small project like this it's sort of a "who cares" decision, but I'd hate for that size to be a design "norm".

I agree with you 100%, and from an educational and learning theory perspective that's the way to go. Something that is engaging, interesting, exciting and cool is highly motivating and therefore has a much greater chance of being able to "sneak in" the learning while the kids think they're just having a bit of fun.

Unfortunately, that's now how it works in England. It's completely bass-ackwards and all performance and learning has to be measured so that it can be inspected, compared to others' and reported on.

Unless there is a political "sea change" in England with respect to this issue, your "preference" is worthless. Edexcel, is a private publisher who writes the exams, and sells the textbooks and makes millions of pounds a year on the system as is. Inspection and Regulation can be billed at a tremendous markup and it pays for a lot of lobbyists.

Unions in England were broken by Thatcher, so the 3 (4?) teacher's unions (yes, plural) are powerless to lobby against the system ... in a brilliant case of divide and conquer, they spend as much time arguing amongst themselves as lobbying for teacher's right to be trusted.

The office of inspection of schools has a motto of something like "Better education through inspection" ... how disjointed is that? Not through "instruction" or "enjoyment" or "making cool stuff" but "inspection"

Sorry ... your preference is irrelevant.

I feel very passionately about that, every day they're failing their children and they can do so much better.

It's a large part of the reason I've quit teaching but, Que Sera right?

There are two sides to that story. There's quite a bit going on that has been left out. England's school system really doesn't work in a way that teachers can just hand out whatever grade a student "deserves".

The kid was handed a syllabus, a marking scheme and told "if you do this you'll get an A". He threw out the mark scheme and did his own thing.

He then tried to get an A for doing whatever the hell he wanted.

I applaud him for pursuing his own interests and taking charge of his own learning - I really do - but the system doesn't allow for that.

You don't hand in creative writing and expect to get a math credit. You don't hand in physics to your chemistry teacher.

I wish the system in England were different. Maybe I'd still be teaching there if it were, but you're casting a lot of aspersion and judgement on a subject that you don't seem to know much about.

Worse than that, difficult courses have much lower pass rates and scores. Teaching them is risking your standing on the league tables. It also precludes your ability to teach the 4 credit GCSE in ICT which (at my last school) got a 100% pass rate.

If you funnel every kid in your school through a course on which 100% of them get 4 GCSE credits and target measure on the league tables is percentage of students who get 5 GCSEs, then you're going to do it. Remember, funding, promotions and everything else is based on your standing on the league tables.

No. Actually Teaching programming in your computer classes is a recipe for failure.

I was an ICT teacher in the UK for 5 years. Our requirement is that we teach units to an accepted common standard.

http://www.teach-ict.com/contributors/Ritchie_King/gcseproje...

Is an example of one marking scheme that I found (on a quick search).

See page 7 and notice that the requirement is that the student "identify" on a scale of 5 marks

- A clear statement of the problem, giving some background detail and identifying user(s)

- Consideration of possible alternative solutions with adequate justification given for the chosen method

- Quantitative objectives or user requirement

Now if a student just goes out and makes a mobile app that doesn't cover any of the requirements so for the above section I'd have been forced to give a 0.

The UK school system has a number of very deep, systematic problems that are preventing it from delivering a good education in IT.

It's a shame, it's frustrating and it's not going to change any time soon.

I actually laughed at this. It's a total Canandian "in joke"

Thanks so much for this! The graphs are exactly the kind of "conceptual flow" I like to have to give me a 1000 meter view of what's going on.

I'd like a simpler explanation of how it all works. With colorful flow charts and graphs.

I'm really interested in how it compares to what express does. How is it the same, how different? However since I'm reading this from the office between doing actual work I've not got time to read this version.

I may or may not remember to come back and check it out after work.

Anyways, all of that is intended as a round of constructive criticism. At the end of the day, it is what it is.

Although it isn't my intention, I do understand that this might come across as rude, but how is this different from the last few times it has been posted here on hackernews?

http://www.hnsearch.com/search#request/all&q=nodebeginne...

6 months ago, 4 months ago... It may be a good book, but it isn't new, or as far as I know, changed significantly enough to warrant being listed on the front page.

Klick Inc, Toronto Canada

Digital agency. We're just shy of 200 employees and growing. One of Canada's 50 best managed companies. We're looking for Web and Mobile Application developers, Front end web, Project Managers, Senior project managers, Account managers, Reporting analysts, Creative designers.

If you're a programmer, we're language agnostic with respect to your past experience, mostly .NET in practice (and rapidly expanding in the mobile space).

You can browse the positions at http://www.klick.com/career_opportunities/ and if you've got any questions, you can email me directly: alex.chesser at gmail

I've got a good breadth of experience and honestly believe that employees are treated really well here. There's a good focus on work-life-balance, weekly paid lunch, lots of free snacks, a bottomless ice cream fridge (careful ... it's delicious and really tempting!)

Also - steps from Yonge & Bloor station so pretty good for a commuter (I guess union might be a touch better but oh well).

Any questions, email me and I'll tell you all about it. I can even pass your resume on to HR for you.

The E-2 can be had for as little as $100,000

http://www.h1base.com/visa/work/Compare%20work%20Visas/ref/1...

I thought they were going to be posting videos of the lectures... but when I logged in today and couldn't see the it anywhere. Did I misunderstand something? Is it posted? If yes, where?

My take on that article is that the fact that node is written in "blub" could equally be seen as an advantage.

Popularity will also mean a broad based skill set ready to work for you when it comes hiring time.

Of ever developer who's worked in Ruby, PHP, C#, Blub... they've all had to have their hands on Javascript. So this one-ring-to-rule-them-all isn't necessarily about being the best & most powerful language ... but it's the one that might be easier to hire for.

What of the social cachet attached to working in the "cool" language?

I wrote project in node and I loved it. I learned a mountain about Javascript. And in that respect, it's been a resounding success.

It has even been mildly successful running at ~1000 uniques a day, ~3000pv. In retrospect, it was absolutely the wrong tool for the job. Totally stupid. Should have used PHP & MySQL. (Or even node+mysql for that matter).

Regardless - I'm now a lot clearer on what a good use of node is vs. what a bad use is. (Hooray for deliberate professional practice).

I've read about some other languages that "already do what node says it does" - I've heard "Ruby's twisted something-or-other does that", either LISP or SCALA or something.

But I couldn't work in those languages and simultaneously increase my knowledge and understanding within my current professional practice (in a way that is directly relevant). With node/javascript I am able to get both.

I'm not actually sure if my point is very well made. It's certainly not a rebuttal to what you've said (Or even a very effective redirect for that matter).

I'd love to know more about the other stuff & the competition (as you say), but I was attracted to node. I've only got so many hours a day to program. I've got to start somewhere and picking up node (to me) seemed a really fantastic place to start.