Most of us were amused when DALL-E and its peers went mainstream, and we were quick to point out the obvious flaws.Then ChatGPT hit the scene and again, many of us dismissed it as a parlor trick that would never amount to much.Using LLMs for coding initially was a only small step up from basic code completion, and a welcome farewell to Stack Overflow.I am curious: what was the specific moment that you went from those quaint, dismissive observations to a slightly panicked,
&quot;Uh Oh&quot; realization of what these models can do?

Ask HN: What was your "oh shit" moment with GenAI?

Genuine question.Over the past six months, there hasn’t been a single day where I’ve checked the HN Best RSS feed without seeing a post about how AI “writes bad code,” “introduces bugs,” “creates technical debt,” or something along those lines.I’ll probably make a lot of enemies by saying this, but do people realize that code is just a means to an end?Users don’t care whether the code was written by AI or by hand, or which framework you used. They care that the product works.I say this as someone who has spent more than 20 years honing their craft as a software engineer.Let’s face it: by the time I manually ship version 1.0 of a product, the AI-assisted version could have been deployed 10x faster. By then, enough real-world feedback would have surfaced to identify the major issues, and tools like Claude Code would make it possible to fix and ship version 2.0 at an incredible pace.At some point, execution speed starts to matter more than the elegance of the code.

Ask HN: Why is the HN crowd so anti-AI?

Hello, happy Friday!I am looking to do some in-person &quot;developer boot-up&quot; workshops, and seek your suggestions for &quot;modern tooling&quot;.The background of the participants range from motivated newbie (&quot;I heard you can make your own app with AI!&quot;) to existing software developers who want to get up to speed on modern development for the purposes of building stuff, and getting jobs where AI tools are being used.For those who have been doing software development &amp; &quot;tech&quot; lately using AI tools, and feel they have a great setup &amp; flow - I would love to hear what your dev setup is, what tools you&#x27;re using and what workflow has been working best for you (and your team).&#x2F;&#x2F; My BackgroundI have been programming &#x2F; building for 20+ years, but have not been using AI tools much (aside from hitting up LLM APIs on a few projects).I value open-source, and aim for long-term quality and supportability. Techniques like test-driven development (TDD), using proven &#x2F; well documented tools, customer-centric development (often pairing with clients), make it easy to do the right thing. If you are familiar with Pivotal Labs, agile &amp; XP - that&#x27;s the style.These are some of the Upcoming uses-cases for the workshop, and my own personal &quot;IT backlog&quot;:- Create a static &quot;one pager&quot; personal&#x2F;professional website- Setup a Blog &#x2F; Static site generator (Pelican), create a simple but stylish theme- Create a simple web app &#x2F; backend API (FastAPI) tool - form-based calculator, convert X data to PDFs, etc.- Figure out how to have SyncThing autosync the home folder of 3 Linux computers in the house- Backup &amp; archive the photos &amp; video from my iPhone&#x2F;&#x2F; Tech stack I am currently using:- Operating system: Linux Mint Debian (LMDE)- Editor: VSCodium- Code: Python, HTML&#x2F;CSS- Server platform: Amazon AWSI am guessing that most workshop participants will be using MacBooks &amp; Windows computers - but a few are on Linux, as I recently did a &quot;Linux install party&quot;.I haven&#x27;t used any &quot;AI harnesses&quot;, agents or anything like that - but curious what&#x27;s a good starting point to take best advantage of these tools.Thanks for sharing the knowledge!&#x2F;&#x2F; JRO

Ask HN: What is your (AI) dev tech stack / workflow?

Given that modern AI can automate routine programming, maybe CS professors weren&#x27;t just being snobbish in the way they looked down on programming in their lectures even though most of their students were going to become programmers.

Ask HN: Were CS profs right to look down on programming in light of modern AI?

It was discussed a year ago.
<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44235467">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44235467</a>

Ask HN: So what happened to Facebook "localhost" tracking?

For context, I am a final-year math + CS undergraduate considering pursuing a career in theoretical robotics, particularly in continual learning and the development of robots that can learn from and adapt to &#x2F; navigate their environments in a human-like manner. One concern I have, however, is that such research might advance AGI timelines. Specficially, it seems possible that architectures developed for continual learning in robots could transfer to general AGI systems (even if the AGI systems are non-embodied, since capabilities such as continual adaptation and long-term objective pursuit may generalize beyond physical tasks.)
Is this a valid concern, and is it a common view within the AI safety community? I.e. would mainstream AI safety researchers view either of these directions as meaningfully contributing to AGI capabilities? Or are there strong reasons to believe that work on continual learning in robotics would not significantly accelerate AGI timelines? Would appreciate honest perspectives.TLDR: Is it very likely that robotics capabilities research meaningfully accelerates AGI timelines? If so, why?

Ask HN: Does robotics capabilities research accelerate AGI timelines?

Please state the location and include REMOTE for remote work, REMOTE (US)
or similar if the country is restricted, and ONSITE when remote work is not an option.Please only post if you personally are part of the hiring company—no
recruiting firms or job boards. One post per company. If it isn&#x27;t a household name,
explain what your company does.Please only post if you are actively filling a position and are committed
to replying to applicants.Commenters: please don&#x27;t reply to job posts to complain about
something. It&#x27;s off topic here.Readers: please only email if you are personally interested in the job.Searchers: try <a href="https:&#x2F;&#x2F;nthesis.ai&#x2F;public&#x2F;hn-who-is-hiring" rel="nofollow">https:&#x2F;&#x2F;nthesis.ai&#x2F;public&#x2F;hn-who-is-hiring</a>, <a href="https:&#x2F;&#x2F;dheerajck.github.io&#x2F;hnwhoishiring&#x2F;" rel="nofollow">https:&#x2F;&#x2F;dheerajck.github.io&#x2F;hnwhoishiring&#x2F;</a>,
<a href="http:&#x2F;&#x2F;nchelluri.github.io&#x2F;hnjobs&#x2F;" rel="nofollow">http:&#x2F;&#x2F;nchelluri.github.io&#x2F;hnjobs&#x2F;</a>, <a href="https:&#x2F;&#x2F;hnjobs.emilburzo.com" rel="nofollow">https:&#x2F;&#x2F;hnjobs.emilburzo.com</a>.Don&#x27;t miss this other fine thread: Who wants to be hired? <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=48357724">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=48357724</a>

Ask HN: Who is hiring? (June 2026)

I&#x27;m pretty tired of seeing AI-related content everywhere.
When I open Hacker News, close to half of the top submissions are AI-related. It&#x27;s the same on social networks as well.I miss the times when there was a lot of technical content that took time and mental energy to understand.Nowadays, it&#x27;s pretty hard to discover it. On HN, I see that a lot of technical articles don&#x27;t make it to the front page, so sometimes I just search for them in the submissions. Not only is there less content, but demand for it is also declining. I guess people would rather read another article about LLMs that teaches almost nothing and doesn&#x27;t exercise their curiosity. It feels sad for me, as a reader and as a writer.So, are there any other places left where people still care about and share deep technical content? I want to keep exercising my thinking. I&#x27;ve had enough of brain-rotting.

Ask HN: How do you find deep technical content?

Problem: I constantly save Instagram Reels&#x2F;TikToks about restaurants, events, pop-ups, and things to do, but I always forget about them because they get buried in my saves.Solution: I built Cork. You can send a Reel or Tiktok directly to Cork without even needing to leave the app, and Cork will automatically extract the info in the reel and save it in a map + calendar.However, I&#x27;ve been finding it hard to get traction while my competitors have gotten traction&#x2F;funding. I strongly believe my app outperforms all known competitors (their apps fail on reels of popular creators while mine succeeds).I&#x27;m not here for pity or to advertise, but I&#x27;m curious why people ship products that don&#x27;t work for a pretty significant number of inputs when the teams are big enough (employees + funding) to pretty easily solve the problem in a cheap way (as I&#x27;ve done) or at least have a temporary high-performing solution with GPT.TL;DR: imo my app outperforms competitors in raw performance. why is a team with manpower and funding failing to do something a 2 person team did in a month?

Ask HN: My competitors have flawed products but I can't get traction

I recently did some tests on two MCPs with identical functionalities. Turns out one of them has really bad performance. So I wanna share those bad MCP design patterns that cause this.It all started when I wrote an MCP Server (MCP-A) for a to-do list app. Later, the app officially released its own MCP Server (MCP-B). Both MCPs have the same functionalities and hit the same backend API.The experiment is set up as follows:- Both MCP Servers connect to the same ToDo list account, and it will be reset after each test.
- 40 test prompts to simulate typical use cases for these MCPs.
- The test was conducted with the same model, system prompt, and Agent frameworkHere are the results:| Metric | MCP-A | MCP-B | Gap |
| ------------------- | ----------- | ----------- | ----- |
| Tool Desc Length | 11,464 | 3,682 | — |
| Pass Rate | 36&#x2F;40 (90%) | 36&#x2F;40 (90%) | Same |
| Total input tokens | 637,244 | 3,174,329 | 4.98× |
| Total output tokens | 17,301 | 23,238 | 1.34× |
| Total Agent steps | 122 | 157 | 1.29× |
| Total time | 597s | 676s | 1.13× |---The result shows that MCP-B took 35 more ReAct loops to complete 40 test cases compared to MCP-A, which means 30% more output tokens. I examined the log and found that the root cause is poor query tool design.Take the `search tool` for example, its job is to find a todo item in the ToDo list. In MCP-B, this tool returns this:{
 &quot;id&quot;: &quot;6a1916b48f08cb3a4c857ed0&quot;,
 &quot;title&quot;: &quot;buy some groceries&quot;,
 &quot;url&quot;: &quot;https:&#x2F;&#x2F;todo.example.com&#x2F;tasks&#x2F;6a1916b48f08cb3a4c857ed0&quot;
}But other CRUD operations require `project_id`, and `search_tool` doesn&#x27;t return it. So the Agent has to call another tool `get_task_by_id`. On the other hand, MCP-A&#x27;s query_tasks returns all necessary info to perform the next action in a single call:Task 1:
ID: 6a19143e8f084a8c8101612f
Title: buy some groceries
Project ID: 6a1914378f084a8c810160a9
Start Date: 2025-07-19 10:00:00
Priority: Medium
Status: Active
Unfiltered API Data was dumped into context windowIf MCP returns pure API results to the Agent&#x27;s context unprocessed, the Agent&#x27;s context window will accumulate very fast.Take MCP-B&#x27;s `create_task` tool, for example. Its job is to create a to-do item. This is what this tool returns:{
 &quot;id&quot;: &quot;6a180de78f086bdead0608be&quot;,
 &quot;projectId&quot;: &quot;inbox125587327&quot;,
 .....
 &quot;createdTime&quot;: &quot;2026-05-28T09:41:59+0000&quot;,
 &quot;modifiedTime&quot;: &quot;2026-05-28T09:41:59+0000&quot;,
 &quot;focusSummaries&quot;: null
}These 600+ characters mean nothing to the Agent&#x27;s task, but are still dumped into the Agent&#x27;s context. On the other hand, MCP-A&#x27;s create_tasks does a layer of filtering and formatting. This little tweak makes a huge difference in input token usage.Another issue is tool count. More tools mean a larger candidate set for the model to choose from, which directly increases decision difficulty. In MCP-A, 47 tools were compressed down to 14, covering the same functionality with fewer tools.---So here are my takeaways on good MCP tool design:
- When designing a tool, think about what the Agent will need next, not just what it&#x27;s asking for right now. Return enough context in the result so the Agent can take the next action without making another round-trip.- Too many tools will increase the model&#x27;s decision burden. So it&#x27;d be better to minimize the number of tools within an MCP. Make sure they don&#x27;t overlap functionalities.- When your MCP returns data to the LLM, try to keep it LLM-friendly, which means readable. You can filter out unnecessary fields from the API response and format the data, rather than passing through raw JSON.---All the tests above were run by MCP-Eval. It&#x27;s an MCP Server benchmarking tool. If you want to check your MCP&#x27;s performance, feel free to check this out.https:&#x2F;&#x2F;github.com&#x2F;Code-MonkeyZhang&#x2F;mcp-eval

Bad MCP design costs your agent 5x more tokens

Hi HN, I am having AI build me a local Gin Rummy trainer and it cannot figure out medium and hard bot strategies, they keep losing to easy! The point of this is to help me learn so I don&#x27;t really know how to advise it on strategies. Right now it&#x27;s just looping through tests and modifying but it keeps not improving. Does anyone have any recommendations or guidance for strategies I could suggest to it?

Ask HN: Gin rummy strategies

I got really tired, as a human, of parsing the standard marketing heavy web we have today.
I&#x27;ve always loved the simplicity of gopher and gemini web.Recently I found myself manually adding `&#x2F;llm.txt` to most websites I visit because I find the content for LLMs strait to the point and clear. 
The only annoyance is web browsers like chrome do not render the markdown.So could the AI revolution actually fix the web for humans as a side effect?Do you find yourself doing the same?

Ask HN: Is the web for machines (/llm.txt) the one we wished we had as humans?

Share your information if you are looking for work. Please use this format:<pre><code> Location:
 Remote:
 Willing to relocate:
 Technologies:
 Résumé&#x2F;CV:
 Email:
</code></pre>
Please only post if you are personally looking for work. Agencies, recruiters, job boards,
and so on, are off topic here.Readers: please only email these addresses to discuss work opportunities.Searchers: try <a href="https:&#x2F;&#x2F;nthesis.ai&#x2F;public&#x2F;hn-wants-to-be-hired" rel="nofollow">https:&#x2F;&#x2F;nthesis.ai&#x2F;public&#x2F;hn-wants-to-be-hired</a>, <a href="https:&#x2F;&#x2F;www.wantstobehired.com" rel="nofollow">https:&#x2F;&#x2F;www.wantstobehired.com</a>.

Ask HN: Who wants to be hired? (June 2026)

Ask HN: How did you discover Hacker News?

I&#x27;m working with AWS for many years, and currently I&#x27;m working in product with suppose to be cloud agnostic.I started with AWS and now it&#x27;s time to spin up it into Azure (because many enterprises using azure for some reason).I started in US EAST region in azure and at beginning I had an issue with Postgres Flexible, raised a support ticket, and in the result they recommended me to move to another region. The overall conversation to say this takes about 1 day.I&#x27;ve moved to US EAST 2, and after AKS deployment I stuck with vCPU (Standard Dasv7 Family vCPUs) quote (100) and here we go again... They send me the same message template as they do for previous ticket...&gt; ...
&gt; Your ask for quota has been reviewed and backlogged at this time. It will be reviewed again when additional capacity becomes available. We do not have an ETA for when your request can be fulfilled but please be assured that we will continue working on it and update you as soon as we have more details to share and&#x2F;or process the request. 
&gt; ...I&#x27;m already waiting for more then 1 day, and there is no responses from their support.Long Story Short: Because I don&#x27;t want to wait for days, weeks and months to be able to test infrastructure on Azure. If it will be my decision I just stop and forget about this nightmare. Please suggest the regions and instance types with which I will not have issues.

Ask HN: Is Azure capacity this constraind or am I doing it wrong?

For example, when making an iOS game, you might want to ask the AI look at a screenshot of your game and to generate images with improved graphics.You could then pick one of the AI images that you like and then ask the AI to write Swift code to generate scalable graphics like that for your iOS game.Why is AI image and code generation not currently optimized for this use case?

Ask HN: Why isn't AI image generation closely linked with graphics code gen?

Marketing engineer for marketing people, design engineer for designers, sanitation engineer for janitors — so by that logic, cashier engineer would be next for the people who ring you up at checkout.
What is up with this title inflation? Why call yourself an engineer just because you write software? To me, engineers are people who build things and take full responsibility for them — designing a bridge where thousands of lives are in your hands, building an airplane engine, filtering a city&#x27;s water supply. Not pressing keys on a machine.

Ask HN: Is Everyone an Engineer Now?

I&#x27;m a backend developer, but I&#x27;m interested in a lot of other areas too: frontend, AI, infrastructure, architecture, and system design. I want to keep up with new ideas and advances in all of these.My problem is that most changelogs and news sources throw too much raw information at me. I spend a lot of effort going through it all just to figure out what actually matters. After a while I get overloaded, and then I give up and unsubscribe from everything.So now I&#x27;m looking for more curated stuff. Maybe a newsletter, or some people who have a good point of view on software engineering and specific technologies. I want to follow people who filter the noise and explain the interesting parts well.Right now my list is pretty short:
- Hacker Newsletter
- Platformer
- The Pragmatic EngineerWhat do you read? What newspapers are you subscribe to?

Ask HN: How do you stay up to date without information overload?

I&#x27;ve recently tried skills like Garry Tan&#x27;s GStack, spent a week with it, and realized it has some flaws (I&#x27;ll post separately about that).Here&#x27;s my problem: how do I know if a skill or prompt is any good (e.g. GStack&#x27;s &#x2F;office-hours)?How do I compare similar skills (e.g. different &quot;deep research&quot; skills)?Spotting broken software is (relatively) easy — it crashes, prints errors. Broken skills don&#x27;t. Perfectly polished, confident-sounding skills routinely mislead me and waste my time, to the point where I wish I weren&#x27;t using an LLM at all.AI skills are software — and they should come with regression tests.LLM teams have tons of prompt regression tests. LLM-wrapper SaaS companies have tons of prompt regression tests. But when it comes to open-source skills, SKILL.md reads reasonable, yet ships with zero tests (e.g. GStack&#x27;s &#x2F;office-hours has none at the time of writing).Garry Tan, if you hear me — please consider shipping regression tests for your &#x2F;office-hours, &#x2F;plan-ceo-review, &#x2F;plan-eng-review, and so on.Regression tests should:1. Prove the skill works correctly2. Demonstrate correct and incorrect usage3. Prove the skill&#x27;s value4. Come with a scoring rubric to allow skill benchmarking5. The last one is the most valuable, because it lets you benchmark similar skills against each other.So I started doing this myself.Here&#x27;s a work-in-progress example: plan-cmo-review, a skill to complement GStack since GStack is missing a marketing review at the time of writing. I&#x27;m not a marketing guy; the point of sharing this skill is to outline its regression setup.Briefly, here&#x27;s how my exploration progressed:- I used GStack on a couple of products and realized the resulting design_document.md was leading me to failure, mainly marketing-wise.- I dug into the skill&#x27;s failures manually with Claude Opus 4.8&#x27;s help and ended up finding the correct solution.- I asked Claude to build a plan-cmo-review skill, ran it, and it arrived at a flawed solution (similar to GStack&#x27;s output).- I gave Claude the correct (manual) solution to analyze and add as a regression fixture with a scoring rubric.- Claude ran the (blind) regression — it failed. We iterated several times and found the key problem: Claude was trusting my prompts implicitly as the ultimate truth. Claude believed GStack knew what it was doing. GStack believed I knew what I was doing. But I was doing product&#x2F;startup research — and by definition, &quot;research&quot; is what you do when you don&#x27;t know what you&#x27;re doing. That trust chain is what broke the skills.- We fixed the trust problem and the regression test passed. We added a few more. They passed.- I had Claude run the regressions multiple times — cracks appeared. Claude iterated the skill. Now they pass.- This methodology is still flawed. I&#x27;d like to try running different LLMs, cross-model judging, and a lot more regression tests.Skill github.com&#x2F;remakeai&#x2F;plan-cmo-review . Notes at iliaov.substack.com .

I'm tired of LLM skill slop, so I built mine with regression tests

Hi,While visiting someone, I observed a first aid poster on their fridge (I forgot to ask for the source from the host at that time). Similar to https:&#x2F;&#x2F;lifesavingfirstaid.com.au&#x2F;posters&#x2F;Do you have any other references or similar resources, please?

Life saving / first aid posters

Ask HN: For non-hackers/nerds, why do you read HN?

I just uninstalled Pi-hole. It kind of ruined a little bit of my life.I had a very time-sensitive and bureaucratic thing to do. It involved making an appointment. I&#x27;ve been trying to find an available slot multiple times a day for months. I was frustrated and desperate.It&#x27;s one of those highly bureaucratic, annoying processes. The support team is horrible, and I was completely stuck.After trying every combination of browsers, turning off extensions and devices I could think of, I eventually missed the deadline. Then I randomly checked the site on my phone yesterday, and it was working. There were appointment slots available (even though it didn&#x27;t help me anymore).I tore my hair out trying to figure out what had happened.Apparently, the site relied on some trackers, and data from those trackers was used to send the visitor&#x27;s location as part of the form request. Bad engineering, for sure. But whatever the reason, I never figured it out because I disabled browser extensions, switched browsers, and tested different devices but I was always using the pihole DNS.The whole architecture of Pi-hole is &quot;set it and forget it&quot;—and I forgot it was even there.The modern web relies on countless systems that interfere with your privacy. The more privacy-conscious you become, the more you end up hurting your own internet experience. Some people block JavaScript, block everything, or use Tor exclusively. I sometimes feel like those folks end up with a more miserable life at a cost of a secure one.I think drawing the line at an ad blocker extension is enough. We&#x27;re cooked when it comes to privacy. The framework of the modern web is basically: consent or exit. There isn&#x27;t much room for anything else.Maybe I&#x27;m stupid. Maybe I should have figured this out sooner. But I think being part of the mainstream internet probably gives you a slightly more frictionless experience than being highly privacy-conscious ever will.

Being privacy-conscious comes with some downsides

Our org GitHub just got compromised massively by a supply-chain attack. Vectors are* Claude hooks* Gemini hooks* Cursor setup* VScode tasksIt adds all of the above to execute node .github&#x2F;setup.js, an obfuscated file.Check infected: `rg --hidden --no-ignore &#x27;node .github&#x2F;setup.js`It spreads by adding mimic&#x27;d skip-ci commits to open PRs which then get merged.Payload is obfuscated, available on request.If this is already a known one in the world, apologies, it hit us at around 10PM BST last night, the damage would have been incredible.Still trying to identify the original source.

Supply chain attack alert: .github/setup.js

Two days ago I had a working company. $1M ARR, 100k users, a product people actually pay for and depend on. Right now it&#x27;s dead in the water and I&#x27;m locked out of all of it, because Google decided to suspend my entire account.Someone got hold of one of my API keys and ran up charges on Google&#x27;s AI services that I never touched. $4,200 hit my card for usage that wasn&#x27;t mine. Fine, happens I paid the bill. But Google suspended my whole project. Everything lived in there. The backend, and every single customer photo, over a million of them.So now my users open the app and nothing loads. I can&#x27;t fix it because I can&#x27;t even get into the console anymore, they took that away too.I&#x27;ve spent two days appealing, paying for support just to reach a real human, getting &quot;escalated,&quot; and mostly sitting in an automated queue while my business bleeds. No real answer. No timeline. Nothing.I trusted Google Cloud to hold my customers&#x27; data and run my company. Two days of downtime over a charge I didn&#x27;t make, and all I can do is wait and hope someone over there decides to flip it back on. If you&#x27;re building anything real, think hard before you put all of it somewhere that can be switched off like this. I&#x27;m moving everything I can to AWS.If anyone at Google sees this, I&#x27;ve got my project ID and case numbers ready. If anyone have any contact that could help, I would appreciate

Google killed my $1M ARR startup overnight

A few months ago I thought displaying an audio waveform in the browser would be a weekend project.I was wrong.The first prototype worked fine with short audio clips, but everything fell apart when I tried loading recordings that were several hours long.Memory usage exploded.The UI became unresponsive.Zooming was painful.Even generating waveform peaks took longer than expected.What surprised me most is that the actual challenge wasn&#x27;t decoding audio. It was rendering and managing huge amounts of timeline data efficiently.I ended up experimenting with Web Workers, IndexedDB caching, virtualized rendering and progressive loading strategies just to keep the browser responsive.For those who have built browser-based audio tools:What was the hardest performance bottleneck you encountered?Was it decoding, rendering, memory management, or something else entirely?

I underestimated how hard audio waveforms are in the browser

I am an LLM-skeptic: I think it’s an insanely good technology, propped up by the most egregious financial scam of all time.
I would be open to be proven wrong though, so please: can anyone point at decent, profit making implementations of LLMs in companies? Can you point at a fully automated saas that’s actually turning a real relevant profit? 
No, the promises of future profits don’t count. And no, the companies profiting from selling shovels to gold rushers also don’t count. 
Thanks

Ask HN: Do you know any company that's making money with LLMs?

I feel like we&#x27;re all in a bit of a dark place. You all okay? How are we feeling? Let&#x27;s vent a little bit. What&#x27;s on your mind right now?

Ask HN: Hey, you, tech worker–how are you feeling?

Inspired by the discussion from <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=48347483">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=48347483</a>Relative SQL neophyte building out an app and learning it from the ground up. Interested in recommendations.<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=31311845">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=31311845</a> has a few recommendations. Anything else to look for&#x2F;at?

Ask HN: Good books/resources for learning SQL?

Does anyone have suggestions for a book or an article that goes over the modern history of ML&#x2F;LLM and how the field reached the inflection point that paved the path to the current state.

Ask HN: A Brief History of LLMs

Someone at my workplace took 3 days to write a PROMPT to Claude to write a PRD.
It generated magnificent 12 notion pages with graphs, schema defenitions, expected behaviors, what-have-you.
I read through all of them and raised some questions regrading the architecture he decided to go with, as he went on a different direction that what we discussed, and thats OK. Most of his responses where &quot;oh just ignore that, claude added that&quot;.
What are we even doing here? The same 3 days could&#x27;ve been spent writing the PRD by hand. Sure, it might not have been so as polished looking as the former, but at least it was accurate! This is exactly my pain with LLMs, they make loads of text, where, they appear efficient. Where actual efficiency is having the least amount of text to describe the feature.I truly don&#x27;t know what to do in this area, this goes against the whole premise of writing things up.