15:[[47990675,{"by":"shad42","descendants":121,"id":47990675,"kids":[47991503,47991481,47991871,47991683,47991108,47991000,47991419,47996350,47991486,47991796,47991669,47991972,47993677,47997993,47992009,48001088,47993487,47996437,47993579,47994866,47998078,47991208,47991347,48001671,47991477,47992502,47992777,47995564,47993353,47991214,48002289,47990998,47997618,47995867,47993958,47997498,47991142,47991128,47998321,48159555,47991058,47995623,47994979,47994031,47992814,47994647,47999058,47992077,47993129,47993142,47991147,47997631],"score":182,"time":1777756887,"title":"The agent harness belongs outside the sandbox","type":"story","url":"https://www.mendral.com/blog/agent-harness-belongs-outside-sandbox"}],[45008915,{"by":"yuer2025","descendants":8,"id":45008915,"kids":[45012858,45010742,45012938,45009080,45009222,45012620,45035409],"score":10,"text":"$16","time":1756079852,"title":"Prompt engineering is collapsing – GPT-5 just proved it","type":"story"}],[44854141,{"by":"hitradostava","descendants":16,"id":44854141,"kids":[44857586,44867585,44854608,44854547,44854549,44854161,44854414],"score":10,"text":"It seems that OpenAI have got the PR machine working amazingly. The Cursor CEO said it's the best, as did Simon Willison (https://simonwillison.net/2025/Aug/7/gpt-5/).

But I've found it terrible. For coding (in Cursor), it's slow, fails with tool calls often (no MCP just stock Cursor tools) and stored some new application state in globalThis - something that no model has ever attempted to do in over a year of very heavy Cursor / Claude Code use).

For a summarization/insights API that I work on, it was way worse than gpt-4.1-mini. I tried both mini and full gpt5, with different reasoning settings. It didn't follow instructions, and output was worse across all my evals, even after heavy prompt adjustment. I did a lot of sampling and the results were objectively bad.

Am I the only one? Has anyone seen actual real-world benefits of GPT-5 vs other models?","time":1754821009,"title":"GPT5 is worse than 4.1-mini for text and worse than Sonnet 4 for coding","type":"story"}],[44742251,{"by":"thunderbong","descendants":173,"id":44742251,"kids":[44742988,44743067,44742992,44743820,44743551,44742714,44742627,44743066,44743072,44742540,44744623,44743288,44744432,44743683,44743150,44742712,44742883,44746610,44744147,44745141,44745994,44745535,44742952,44742822,44744704,44743364,44743858,44743685,44743767,44805970,44744046,44745148,44743112,44743475,44744081,44744054,44772428,44745135,44742537,44742535],"score":353,"time":1753935400,"title":"I know when you're vibe coding","type":"story","url":"https://alexkondov.com/i-know-when-youre-vibe-coding/"}],[44739632,{"by":"BUFU","descendants":284,"id":44739632,"kids":[44741016,44741689,44740213,44746471,44741610,44740957,44741699,44742374,44740421,44740138,44741736,44743503,44745508,44740951,44743926,44741971,44753542,44745543,44745394,44744219,44743166,44740851,44743454,44741524,44741768,44742851,44743714,44740965,44746288,44741538,44743696,44740210,44741651,44741407,44742610,44742367,44765154,44741040,44742604,44745942,44741059,44740838,44740289,44741501,44742219,44740531,44743954,44742073,44742362,44750405,44742966,44740972,44740656,44743244,44743187,44742411],"score":560,"time":1753910343,"title":"Ollama's new app","type":"story","url":"https://ollama.com/blog/new-app"}],[44739556,{"by":"simonw","descendants":454,"id":44739556,"kids":[44740344,44740134,44740693,44740786,44740192,44742107,44739834,44740001,44740290,44740083,44740139,44740291,44740947,44739963,44743097,44739887,44740259,44740043,44740311,44740048,44743960,44740560,44742925,44740073,44746042,44740579,44740744,44741272,44742109,44740256,44740241,44740003,44740898,44747498,44740202,44740219,44740997,44740297,44740433,44753920,44741471,44748363,44742719,44740242,44740013,44740212,44740746,44739976,44745986,44744012,44740488,44740913,44741182,44743734,44742390,44747665,44740829,44742294,44747979,44740720,44743914,44740037,44744553,44742314,44740184,44746285,44752863,44757766,44745670,44740101,44745863,44744967,44740361,44741504,44741954,44740090,44744404,44740474,44741373,44759241,44740364],"score":758,"time":1753910000,"title":"Vibe code is legacy code","type":"story","url":"https://blog.val.town/vibe-code"}],[44624604,{"by":"mettamage","descendants":2,"id":44624604,"kids":[44624917,44624729,44629995],"score":2,"text":"I'm just using Claude, copy/pasting code, reading it, making sense of it, thinking about if this is what I want and then proceed. It's a bit slower than the pure vibe coding approach, but I like to stay in the loop and adjust course as quickly as possible.

I haven't used Claude Code yet. So, genuine question, am I missing out? What makes it special versus just using Claude in this copy/paste/reflect way?","time":1753015080,"title":"Ask HN: Claude Code vs. Just Claude","type":"story"}],[44620633,{"by":"hubmusic","descendants":40,"id":44620633,"kids":[44620656,44620844,44627860,44621396,44624277,44624014,44622391,44622316,44624552,44621225,44621948,44622263,44621603,44620890,44621745,44621321,44620759,44620740,44627725,44620781,44620737,44620764,44627728,44620861,44620977,44624764,44620728,44621630,44621216,44620765,44621725],"score":22,"text":"Curious to see what others think. There is west village in nyc or rich neighborhoods in SF, LA where people live and spend as if they are from another planet. Houses, vacation homes, yearly trips, spa, jetski, ski trips and more.

Then, there are some people who barely get by or get by with barely anything left after rent, healthcare, and household expenses.

America has entered this era of distribution of wealth that is really sad. I don’t subscribe to socialism or communism but it’s hard to see that some people are just never going to make it.

As someone who graduated in 2007 and financial crisis struck, it’s been disaster after disaster. I know some lucky people made it in tech and now have $2m or more at their name. But, there are people like me who don’t have much to show for. Just a sad reality.","time":1752969584,"title":"Ask HN: How do you digest fact that you are not successful by 40","type":"story"}],[44223758,{"by":"fallat","descendants":8,"id":44223758,"kids":[44229684,44234507,44230736,44230700,44230976],"score":25,"time":1749471888,"title":"Makefile.md – Possibly Use(Ful|Less) Polyglot Synthesis of Makefile and Markdown","type":"story","url":"https://zoo.dev/blog/polyglot-makefile-markdown"}],[44085343,{"by":"dougsan","descendants":77,"id":44085343,"kids":[44085607,44085761,44085999,44085582,44085822,44085689,44085791,44085687,44085748,44085505,44092500,44085623,44085572,44086057,44085344,44091817,44085868,44085871,44085785,44091501,44093369,44094298],"score":124,"time":1748144458,"title":"Claude Opus 4 turns to blackmail when engineers try to take it offline","type":"story","url":"https://techcrunch.com/2025/05/22/anthropics-new-ai-model-turns-to-blackmail-when-engineers-try-to-take-it-offline/"}],[43976895,{"by":"badmonster","descendants":37,"id":43976895,"kids":[43978880,43978937,43982083,43979788,43982526,43979108,43977761,43979264,43979102,43984790],"score":181,"time":1747165684,"title":"Build real-time knowledge graph for documents with LLM","type":"story","url":"https://cocoindex.io/blogs/knowledge-graph-for-docs/"}],[43740065,{"by":"schappim","descendants":45,"id":43740065,"kids":[43740270,43740305,43740301,43740589,43741050,43746790,43793272,43741008,43740242,43740647,43741681,43741920,43745017,43741628,43742831,43801945,43740586,43744044,43747219,43741804,43741677,43744452,43744815],"score":22,"time":1745102656,"title":"Ask HN: What did you learn too late in your tech career?","type":"story"}],[43620279,{"by":"jcartw","descendants":769,"id":43620279,"kids":[43620583,43621834,43620573,43620729,43620640,43620560,43620800,43622235,43620648,43620475,43621046,43620957,43621090,43626285,43628707,43626956,43620490,43627900,43631100,43622901,43620631,43621372,43621656,43621063,43625658,43621385,43621978,43621003,43620744,43620645,43628358,43622991,43620819,43620852,43627357,43624601,43628168,43623098,43627166,43631554,43623399,43620731,43627063,43620835,43628238,43629377,43628227,43628594,43620574,43629888,43624214,43620806,43626992,43629562,43620912,43621892,43629382,43635374,43621380,43620892,43625145,43622801,43620615,43627615,43621198,43629145,43627675,43628177,43631780,43626948,43621009,43622352,43622385,43630756],"score":667,"time":1744109966,"title":"Brazil's government-run payments system has become dominant","type":"story","url":"https://www.economist.com/the-americas/2025/04/03/brazils-government-run-payments-system-has-become-dominant"}],[43590998,{"by":"ses425500000","descendants":38,"id":43590998,"kids":[43594848,43591383,43591697,43592384,43598623,43593797,43592408,43596947,43598060,43593560,43592934],"score":170,"text":"Hi HN,

I’ve been working on an OCR pipeline specifically optimized for machine learning dataset preparation. It’s designed to process complex academic materials — including math formulas, tables, figures, and multilingual text — and output clean, structured formats like JSON and Markdown.

Some features:\n • Multi-stage OCR combining DocLayout-YOLO, Google Vision, MathPix, and Gemini Pro Vision\n • Extracts and understands diagrams, tables, LaTeX-style math, and multilingual text (Japanese/Korean/English)\n • Highly tuned for ML training pipelines, including dataset generation and preprocessing for RAG or fine-tuning tasks

Sample outputs and real exam-based examples are included (EJU Biology, UTokyo Math, etc.)\nWould love to hear any feedback or ideas for improvement.

GitHub: https://github.com/ses4255/Versatile-OCR-Program","time":1743830553,"title":"Show HN: OCR pipeline for ML training (tables, diagrams, math, multilingual)","type":"story","url":"https://github.com/ses4255/Versatile-OCR-Program"}],[43201709,{"by":"tgma","descendants":29,"id":43201709,"kids":[43204607,43202303,43209303,43206665,43204720,43204664,43204315,43215971,43212489,43202035,43213291],"score":59,"time":1740717093,"title":"Google's Sergey Brin Asks Workers to Spend More Time in the Office","type":"story","url":"https://www.nytimes.com/2025/02/27/technology/google-sergey-brin-return-to-office.html"}],[43059579,{"by":"b4rtazz","descendants":156,"id":43059579,"kids":[43059858,43059832,43060683,43059934,43059793,43060700,43060342,43063898,43059942,43061219,43060110,43074451,43059885,43060693,43061871,43060775,43059899,43060285,43061835],"score":306,"time":1739635889,"title":"Deepseek R1 Distill 8B Q40 on 4 x Raspberry Pi 5","type":"story","url":"https://github.com/b4rtaz/distributed-llama/discussions/162"}],[42950976,{"by":"finding_theta","descendants":194,"id":42950976,"kids":[42985194,42985885,42985479,42985732,42984700,42984867,42985131,42992120,42986123,42984724,42986336,42984916,42988025,42984703,42985717,42985097,42985490,42988613,42986529,42985525,42984737,42987253,42986166,42985874,42989623,42986756,42985895,42985992,42985645,42986151,42986215,42986424,42986958,42984944,42985456,42986856,42990069,42985372,42985424,42985545,42985135,42985384,42985459,42984680,42984911,42984623,42985128,42985210],"score":384,"time":1738773235,"title":"Are LLMs able to notice the “gorilla in the data”?","type":"story","url":"https://chiraaggohel.com/posts/llms-eda/"}],[42984739,{"by":"hekike","descendants":3,"id":42984739,"kids":[42985070,42984834],"score":7,"text":"$17","time":1739037675,"title":"The Two Types of Future Developer","type":"story"}],[42917965,{"by":"sb13","descendants":2,"id":42917965,"kids":[42918000,42918205],"score":1,"text":"With the advent of DeepSeek, I was thinking if I should consider self-hosting open source LLMs vs using OpenAI APIs. Who here has done this already? And what are your main reasons, privacy or cost?","time":1738589077,"title":"Ask HN: Self-hosting LLMs, is it worth the setup?","type":"story"}],[42829034,{"by":"me2too","descendants":91,"id":42829034,"kids":[42845865,42850490,42847715,42850731,42852169,42845700,42848510,42847406,42845956,42845827,42845836,42854111,42847558,42855114,42851411],"score":198,"time":1737884533,"title":"Using AI for Coding: My Journey with Cline and LLMs","type":"story","url":"https://pgaleone.eu/ai/coding/2025/01/26/using-ai-for-coding-my-experience/"}],[42845070,{"by":"hexxeh","descendants":200,"id":42845070,"kids":[42845185,42845100,42845102,42845879,42845134,42846791,42850006,42845081,42845435,42845512,42845218,42845085,42845266,42849154,42845210,42846949,42845258,42849986,42845067,42845215,42845108,42845162,42846672,42850948,42846020,42850907,42845385,42847470,42846385,42845497,42848250,42845400,42847907,42850702,42847165,42852747,42850012,42847473],"score":1263,"text":"https://github.com/google/pebble","time":1738008549,"title":"Google open-sources the Pebble OS","type":"story","url":"https://opensource.googleblog.com/2025/01/see-code-that-powered-pebble-smartwatches.html"}]]

user:	8thcross
created:	September 26, 2015
karma:	7