Best of GPT — 2025

1. 1
   Article

   Data Engineer Things·1y

   10 minutes are all you need to understand how Transformers work in LLM

   Understanding how transformers work in large language models (LLMs) can be achieved quickly by breaking down the steps involved in the process. Starting from tokenization, where input data is converted into tokens, these tokens are then embedded into numerical representations understood by the model. These embeddings are processed through multiple transformer layers that use attention mechanisms to determine the importance of each token in relation to others. Finally, the processed data is projected back onto the vocabulary to predict the next token in a sequence. This foundational knowledge helps in exploring further intricacies of models like GPT-2.

   270

   4
2. 2
   Article

   Community Picks·1y

   GPT-4.1 Prompting Guide

   The post provides a comprehensive guide on GPT-4.1 prompting using the OpenAI API. It includes open-source examples, advanced techniques, and detailed walkthroughs, empowering developers to efficiently utilize GPT-4.1 capabilities. The guide also encourages sharing your own examples and insights for further learning.

   268

   7
3. 3
   Article

   AI·42w

   BREAKING: GitHub accidentally leaked GPT-5 details (proof inside)

   GitHub accidentally published and quickly deleted a changelog entry announcing GPT-5's general availability in GitHub Models. An archived version of the deleted page serves as evidence of the premature announcement, suggesting GPT-5 may be launching imminently.

   242

   26
4. 4
   Article

   DiamantAI·41w

   GPT-5 just proved something important - the scaling era is over

   The performance gap between GPT-4 and GPT-5 is smaller than previous generational leaps, signaling the end of the AI scaling era where bigger models automatically meant better performance. The future of AI development is shifting toward sophisticated engineering and AI agents built with existing models, rather than relying on massive compute budgets and larger model architectures.

   195

   11
5. 5
   Article

   Hacker News·41w

   gpt-5 leaked system prompt

   A leaked system prompt reveals GPT-5's internal instructions and capabilities. The prompt shows personality guidelines emphasizing clarity and enthusiasm, memory management through a 'bio' tool, canvas functionality for document creation, image generation capabilities, Python code execution environment, and web search tools. It includes specific behavioral constraints like avoiding opt-in questions and copyright material reproduction.

   104

   6
6. 6
   Article

   Atomic Spin·51w

   Tips & Tricks for Better AI Prompts

   Effective AI prompt engineering requires structured formatting with clear sections like Instructions, Context, and Examples. Keep prompts concise rather than verbose, as AI models excel at inferring from limited examples. Using markdown-style sectioning and asking AI to help craft system prompts can significantly improve response accuracy and reliability.

   69
7. 7
   Video

   Fireship·41w

   GPT-5 is here... Can it win back programmers?

   OpenAI released GPT-5, claiming it's the first AI to outperform humans on certain benchmarks, but the reality is more nuanced. While GPT-5 unifies multiple models for better task routing and costs significantly less than competitors at $10 per million tokens, it still has limitations in coding tasks. Testing shows it can generate functional Svelte applications but makes errors with framework-specific rules. The model represents more of a consolidation effort than a revolutionary breakthrough, and programmers' jobs remain safe for now.

   67

   4
8. 8
   Article

   UX Planet·28w

   Scenario-based AI Chatbots for Language Learning

   Scenario-based AI chatbots are transforming language learning by providing contextual, real-world practice environments that reduce anxiety and improve fluency. Companies like Duolingo, Mondly, and Babbel leverage GPT-4 and LLMs to create adaptive conversations simulating authentic situations like ordering coffee or navigating airports. Research shows chatbot-assisted learning produces significant positive effects (g = 0.484) compared to traditional methods, with key success factors including adaptive complexity, authentic contexts, immediate feedback, and multimodal engagement. The approach addresses the gap between classroom knowledge and real-world conversation by building neural pathways connecting words to situations, while creating judgment-free practice spaces that reduce speaking anxiety and build learner confidence.

   65

   1
9. 9
   Video

   Fireship·1y

   GPT-4.5 shocks the world with its lack of intelligence...

   GPT-4.5 was released by OpenAI, but it has disappointed many due to its high cost and lack of novel capabilities. The launch focused on its ability to chat in a more natural way and its lower hallucination rate, but many found it underwhelming. The model is extremely expensive, costing $150 per million output tokens and accessible only to $200 per month Pro users. Despite scaling up the number of parameters and compute, the improvements seem marginal. The AI model is particularly criticized for its performance in programming tasks compared to other existing models.

   58

   2
10. 10
    Article

    GoPenAI·1y

    Not Just Text: RAG That Sees Images and Reads Tables 🧠🔍

    Retrieval-Augmented Generation (RAG) enhances the capabilities of Large Language Models (LLMs) by providing additional context for more accurate responses. This guide demonstrates building a multimodal RAG system that processes not only text but also tables and images from documents. Using the Unstructured library and GPT-4.1, it outlines parsing PDFs, summarizing content, creating embeddings, and storing vectorized data in ChromaDB. The approach aims to improve document understanding by integrating various content types, addressing accuracy, stability, and other potential risks.

    55
11. 11
    Article

    Daily Open Source Tools·51w

    Prompt Management Dashboard

    Promptzy is a web application for managing and organizing AI prompts with cross-device synchronization via Supabase. It features a free GPT-4.1 powered assistant for prompt generation, mobile app installation capabilities, and multiple deployment options including direct web access, npm package installation, and source code deployment from GitHub.

    39
12. 12
    Article

    Medium·38w

    GPT-5 System Prompt Leaked : 7 Prompt Engineering Tricks to learn

    Analysis of a leaked GPT-5 system prompt reveals seven key prompt engineering techniques including identity locking to prevent prompt injection, knowledge anchoring for temporal context, multimodal toggles for routing, personality injection for behavioral control, content safety as first-class instructions, self-denial of hidden mechanisms to prevent conspiracy theories, and dynamic retrieval gates for up-to-date information. The techniques demonstrate advanced strategies for building robust AI systems through careful prompt design rather than fine-tuning.

    29

    1
13. 13
    Article

    Prompt Engineering·44w

    Prompt Management Dashboard

    Promptzy is a web application for managing and organizing AI prompts with cross-device synchronization via Supabase. It features a free GPT-4.1 powered assistant for prompt generation, supports multiple installation methods including web, mobile app, npm package, and source code deployment. Users can categorize prompts by system prompt, task, image, or video, and sync data across devices using Supabase credentials and a username.

    29

    2
14. 14
    Article

    DigitalOcean Community·38w

    Build An AI Customer Support Agent With GPT-OSS

    Learn how to build an AI customer support agent using GPT-OSS-120B on DigitalOcean's Gradient AI Platform. The tutorial covers creating knowledge bases from various data sources like ticket logs, FAQs, and product documentation, then deploying the agent either through API or the no-code control panel interface. Includes practical examples using a fictional biotech company and guidance on data preparation, agent evaluation, and integration.

    24
15. 15
    Video

    Theo - t3․gg·28w

    GPT-5.1 is built for normies

    GPT-5.1 represents a shift toward consumer-focused AI with improved conversational tone, customizable personalities, and enhanced safety guardrails. The release prioritizes warmth and accessibility over developer features, with API access delayed. Testing reveals better mental health safeguards and reduced sycophancy compared to GPT-4, though the model's personality options and emoji-heavy responses may not appeal to technical users. The instant variant shows adaptive reasoning that adjusts token usage based on query complexity, while safety evaluations demonstrate meaningful improvements in handling sensitive content.

    18

    1
16. 16
    Article

    Windsurf·23w

    GPT 5.2 is now available in Windsurf!

    GPT-5.2 is now available in Windsurf for 0x credits (limited time for paid and trial users). This release represents the biggest leap in intelligence for GPT models in agentic coding since GPT-5, positioning it as the state-of-the-art coding model at its price point and making it the default model in Windsurf.

    17

    1
17. 17
    Article

    Visual Studio Blog·48w

    Better Models, Smarter Defaults: Claude Sonnet 4, GPT-4.1, and More Control in Visual Studio

    Visual Studio Copilot now defaults to GPT-4.1 for improved performance and offers expanded model choices including Claude Sonnet 4, Claude Opus 4, OpenAI o3 mini, and Gemini variants. The update introduces sticky model selection, simplified model switching, and a new Copilot Consumptions panel for tracking usage under GitHub's consumptive billing model. Users can monitor premium requests and automatically fall back to standard models when quotas are exceeded.

    17

    1
18. 18
    Article

    Collections·40w

    Comparing GPT-5 and Claude Opus for Agentic Coding

    GPT-5 and Claude Opus are compared for agentic coding tasks through implementing a landing page with newsletter signup. GPT-5 excels in initial accuracy and accessibility for beginners but has slower iteration cycles. Claude Opus offers faster iterations and better professional integration through MCP tools, though it requires more debugging initially. The choice depends on whether developers prioritize initial accuracy or iterative speed, with Claude Opus suited for professionals and GPT-5 better for consumer developers.

    16
19. 19
    Video

    ThePrimeTime·30w

    I predicted the future

    A commentary on AI model releases and their incremental improvements, particularly focusing on GPT-5's performance on coding benchmarks like Swebench. The piece discusses developer fatigue with frequent AI announcements that promise revolutionary changes but deliver marginal improvements in practical coding scenarios. It also touches on companies' hasty decisions to replace developers with AI, only to reverse course shortly after.

    15

    1
20. 20
    Article

    Community Picks·1y

    Generate SQL Queries with AI in Seconds

    AI Query leverages advanced AI models like OpenAI GPT and Google PaLM 2 to enable users to generate complex SQL queries using simple English prompts. It supports multiple database engines and provides tools for defining database schemas and translating SQL to English. The service offers various features, including unlimited query generation and explanations, database schema management, and query history. Two pricing plans are available with options for faster AI response and priority support.

    15
21. 21
    Article

    Machine Learning Mastery·46w

    Your First OpenAI API Project in Python Step-By-Step

    A comprehensive beginner's guide to building a Python application that integrates with OpenAI's GPT-4 API using FastAPI. Covers obtaining API keys, setting up virtual environments, creating a REST API endpoint, and testing the integration through a web interface. Includes code examples and step-by-step instructions for creating a functional chatbot API.

    14
22. 22
    Video

    Hacker News·1y

    Deep Dive into LLMs like ChatGPT

    Large language models (LLMs) such as ChatGPT are built through a complex pre-training process involving the downloading and processing of large quantities of diverse, high-quality internet texts. Common Crawl data, along with filtering steps like URL filtering, text extraction, and language filtering, are critical components. Tokenization converts these texts into a sequence of symbols for neural networks to process. These networks are trained to model the statistical relationships between tokens to predict the next token in a sequence. Inference is generating new data from the trained model by predicting subsequent tokens based on a given input.

    12