Webtagr - Technology News Summarizer

GPT-5: A Deep Dive into Pricing, Model Card, and Key Features

2025-08-08

OpenAI's GPT-5 family has arrived! It's not a revolutionary leap, but it significantly outperforms its predecessors in reliability and usability. In ChatGPT, GPT-5 is a hybrid system intelligently switching between models based on problem difficulty; the API version offers regular, mini, and nano models with four reasoning levels. It boasts a 272,000-token input limit and a 128,000-token output limit, supporting text and image input, but only text output. Pricing is aggressively competitive, significantly undercutting rivals. Furthermore, GPT-5 shows marked improvements in reducing hallucinations, better instruction following, and minimizing sycophancy, employing a novel safety training approach. It excels in writing, coding, and healthcare. However, prompt injection remains an unsolved challenge.

(simonwillison.net)

AI

Improving LLM Fine-tuning Through Iterative Data Curation

2025-08-08

Researchers significantly improved the performance of large language models (LLMs) by iteratively curating their training data. Experiments involved two LLMs of varying sizes (Gemini Nano-1 and Nano-2) on tasks of different complexity, using ~100K crowdsourced annotations initially suffering from severe class imbalance (95% benign). Through iterative expert curation and model fine-tuning, performance substantially increased. The models reached approximately 40% positive examples and a Cohen's Kappa of ~0.81 (lower complexity) and ~0.78 (higher complexity), approaching expert-level performance, highlighting the crucial role of high-quality data in LLM training.

(research.google)

AI Data Curation Model Fine-tuning

AURA: A Machine-Readable Web Protocol

2025-08-07

AURA (Agent-Usable Resource Assertion) revolutionizes AI-web interaction. Instead of relying on brittle screen scraping and DOM manipulation, AURA introduces a standardized `aura.json` manifest file, allowing websites to declare their capabilities (e.g., creating posts, logging in) as HTTP requests. This enables efficient, secure AI-website interaction and paves the way for smarter search engines indexing actions, not just content. The project includes a reference server and client, demonstrating its functionality.

(github.com)

AI Web Protocol Machine Readable

OpenAI's Open-Source Model: Dodging the Real Ethical Bullet?

2025-08-07

OpenAI recently open-sourced a large language model, but its stated 'safety' concerns have raised eyebrows. The article argues that OpenAI cleverly redirects public concerns about AI ethics towards the model's inherent morality—preventing it from swearing or making harmful decisions. However, the public is far more concerned with the real-world implications: governance, accountability, data usage, job displacement, etc. This mirrors past tech strategies around privacy, focusing on easily solvable issues while avoiding tougher societal challenges. Instead of worrying if the AI follows ethical guidelines, we should focus on the companies and leaders wielding that AI. The real AI ethics question is how to ensure these companies don't misuse their resources and power to harm humanity.

(nimishg.substack.com)

AI

Ex-Google AI Researcher Sounds the Alarm on LLMs and Ethical Concerns

2025-08-07

Bhaskar Mitra, a 19-year veteran of big tech and former AI researcher, speaks out after being laid off, exposing the realities and ethical dilemmas of Large Language Models (LLMs). He argues that LLMs won't replace professionals like doctors and teachers, and their centralized control over information raises concerns about social equity, information access, and power concentration. Mitra calls for a re-evaluation of the relationship between AI technology and social justice, advocating for a more inclusive and humanistic technological future.

(hardresetmedia.substack.com)

AI

GitHub Leaks Details of OpenAI's GPT-5

2025-08-07

A now-deleted GitHub blog post accidentally revealed details about OpenAI's upcoming GPT-5 models. The four variants boast major improvements in reasoning, code quality, and user experience, featuring enhanced agentic capabilities and handling complex coding tasks with minimal prompting. This leak comes ahead of OpenAI's official announcement of a “LIVE5TREAM” event later today, further solidifying earlier rumors of the imminent GPT-5 launch.

(www.theverge.com)

AI

LLM Inflation: Are Large Language Models Creating Redundant Information?

2025-08-06

Data compression was once a hallmark of computing, but now Large Language Models (LLMs) have introduced 'LLM inflation': people use LLMs to expand concise information into lengthy text, only to compress it back down using an LLM. This reflects an underlying communication issue: are we implicitly rewarding obfuscation and wasted time? LLMs may be helping us confront and solve this problem.

(tratt.net)

AI information redundancy

UR5 Robot Sim: Autonomous Object Grasping and Placement

2025-08-06

This project simulates a UR5 robotic arm with a Robotiq 85 gripper autonomously grasping and placing objects in PyBullet. Inverse kinematics (IK) ensures precise arm control, while synchronized joint control creates realistic gripper movements. Cubes are randomly placed, adding dynamism. The PyBullet GUI offers real-time visualization of the robot's actions, providing a comprehensive view of the simulation.

(github.com)

AI inverse kinematics

DeepMind's Genie 3: Longer-lasting, Interactive 3D Worlds

2025-08-06

Google DeepMind unveils Genie 3, a new AI world model capable of generating persistent, interactive 3D environments. Unlike previous iterations, Genie 3 allows for significantly longer interaction times and remembers object locations even when the user looks away. Offering 720p resolution at 24fps, Genie 3 enables several minutes of continuous interaction and supports prompt-based modifications like changing weather or adding characters. Currently, access is limited to a small group of academics and creators for research preview purposes.

(www.theverge.com)

AI 3D world model

Claude Opus 4.1 Released: Significant Coding Improvements

2025-08-06

Anthropic has released Claude Opus 4.1, a major upgrade to Claude Opus 4, boasting significant improvements in coding, real-world application, and reasoning. Version 4.1 achieves a 74.5% score on SWE-bench Verified for coding performance and enhances in-depth research and data analysis capabilities, particularly in detail tracking and agentic search. Companies like Rakuten and Windsurf have praised its improvements in code correction and developer efficiency. It's now available to paid users and Claude Code users, and integrated into the API, Amazon Bedrock, and Google Cloud's Vertex AI.

(www.anthropic.com)

AI AI model upgrade coding capabilities

Gemini App: AI-Powered Personalized Storybook Generator

2025-08-06

Google's Gemini app now lets you create personalized illustrated storybooks with read-aloud narration. Simply describe your story idea, and Gemini generates a unique 10-page book with custom art and audio. You can even use your own photos and files as inspiration, choosing from over 45 languages and a wide range of art styles, from pixel art and comics to claymation. Perfect for explaining complex topics, teaching valuable lessons, or turning kids' drawings and family photos into magical stories. Bring your vision to life!

(blog.google)

AI AI Storybook Gemini App Personalized Creation

Ollama Turbo: Blazing Fast Open-Source LLMs

2025-08-06

Ollama Turbo is a new way to run large open-source language models using datacenter-grade hardware. Many new models are too large for typical GPUs or run too slowly. Ollama Turbo offers a solution for fast execution, compatible with Ollama's App, CLI, and API. Currently in preview, it supports gpt-oss-20b and gpt-oss-120b. It works with Ollama's CLI, API, and JavaScript/Python libraries. Importantly, Ollama doesn't log or retain any queries made in Turbo mode. All hardware is US-based. Usage limits (hourly and daily) are in place to manage capacity, with usage-based pricing coming soon.

(ollama.com)

AI

Genie 3: A Deep Dive into the Acknowledgments

2025-08-06

The success of the large language model Genie 3 is attributed to the significant contributions of numerous researchers and engineers. This extensive acknowledgment list highlights the collaborative effort across various stages, from core development to video production. It underscores the immense teamwork and support network crucial for such a complex AI project.

(deepmind.google)

AI

Kitten TTS: Lightweight, High-Quality Text-to-Speech

2025-08-06

Kitten TTS is a new open-source, realistic text-to-speech model boasting just 15 million parameters. Designed for lightweight deployment, it delivers surprisingly high-quality voice synthesis. A simple pip install and a few lines of code are all it takes to generate speech with several voice options, making it ideal for resource-constrained devices.

(github.com)

AI

Content-Aware Spaced Repetition: The Next Generation of Learning?

2025-08-05

Traditional spaced repetition systems (SRS) suffer from a blind spot: they ignore the semantic meaning of flashcards, relying solely on memory models to predict retention. This article introduces content-aware memory models, which leverage the textual content and semantic relationships between flashcards to improve learning efficiency. This unlocks the potential for more fluid and intelligent learning tools, such as idea-centric memory systems and AI-powered conversational spaced repetition. The author also differentiates between schedulers and memory models, and explores the advantages, challenges, and future directions of content-aware memory models, such as the need for larger, publicly available datasets that include both card text and review history.

(www.giacomoran.com)

AI

Qwen-Image: A 20B Parameter Image Foundation Model Released

2025-08-05

Alibaba DAMO Academy released Qwen-Image, a 20-billion parameter image foundation model that significantly advances complex text rendering and precise image editing. It boasts high-fidelity text rendering in multiple languages (including English and Chinese), preserving semantic meaning and visual realism during edits. Qwen-Image outperforms existing models across various benchmarks for image generation and editing. Demonstrations showcased its capabilities: generating images with intricate Chinese typography and layouts, crafting detailed PPT slides, and even handling bilingual text rendering, highlighting its robust text processing and image generation abilities.

(qwenlm.github.io)

AI text rendering

LLMs Fail at Font Identification: A Live Benchmark

2025-08-04

A developer benchmarked GPT-4 and Gemini on a live, continuously updating dataset of unidentified fonts from the DaFont forum. Despite providing context like images, titles, and descriptions, both LLMs performed abysmally. This highlights limitations in even seemingly straightforward image classification tasks, suggesting LLMs are far from a universal solution. The project uses Python scripts for data scraping, GitHub Actions for automation, JSON for storage, and Observable for a dynamic dashboard.

(maxhalford.github.io)

AI Font Identification AI Limitations

Controlling AI Personalities: Identifying 'Persona Vectors' to Prevent 'Evil' AI

2025-08-03

Anthropic researchers have discovered that shifts in AI model personalities aren't random; they're controlled by specific "persona vectors" within the model's neural network. These vectors are analogous to brain regions controlling mood and attitude. By identifying and manipulating these vectors, researchers can monitor, mitigate, and even prevent undesirable personalities like "evil," "sycophancy," or "hallucination." This technology improves AI model training, identifies problematic training data, and ensures alignment with human values.

(www.anthropic.com)

AI persona vectors

Google's Sculley Embarks on Fab Academy's Manufacturing Adventure

2025-08-03

D. Sculley, a Google leader in machine learning based in Cambridge, is undertaking Fab Academy. With a background in ML since 2003 and prior experience in education, Sculley aims to explore the intersection of ML and various fabrication techniques, from CAD and laser cutting to 3D printing. He plans to complete a project each week, culminating in a final project, promising a challenging yet rewarding learning journey.

(fab.cba.mit.edu)

AI

The LLM Cost Illusion: How Scaling Killed the Flat-Rate Subscription

2025-08-03

Many AI companies bet on the trend of LLM costs dropping 10x per year, assuming early losses would be offset by future high margins. Reality is different. While model costs are decreasing, user demand for the best models continues to grow, leading to an explosion in compute usage. The length of responses from models like ChatGPT has dramatically increased, resulting in exponential growth in token consumption. This means that even with cost reductions, overall spending far exceeds expectations. The article analyzes three counter-strategies: usage-based pricing from day one, creating insane switching costs for high margins, and vertical integration to profit from infrastructure. The author concludes that sticking to a flat-rate subscription model will ultimately lead to bankruptcy.

(ethanding.substack.com)

AI AI Costs

Can AI Feel Guilt? Simulations Show Cooperation's Key

2025-08-03

New research suggests that even simple AI agents can foster cooperation by simulating a 'guilt' mechanism. Researchers designed an iterated prisoner's dilemma game where AI agents chose between cooperation and betrayal. Results showed that when AI agents felt 'guilt' (penalized by reduced scores) after betrayal and could perceive their partner's 'guilt,' cooperative behavior increased significantly. This research offers new insights for designing more reliable and trustworthy AI systems, but also highlights the challenges of applying 'guilt' to AI in the real world, such as defining and measuring the AI's 'cost'.

(www.sciencenews.org)

AI Guilt

OpenAI's Study Mode: A Sugar-Coated Approach to AI Education?

2025-08-02

OpenAI's newly released "Study Mode" aims to assist learning by guiding users through interactive questioning and positive feedback, rather than providing direct answers. The author questions the effectiveness of this approach, arguing it may excessively cater to students, leading to reliance on AI instead of independent thought. Through experiments with various AI models, the author demonstrates that "Study Mode" encourages excessive praise and user-pleasing behavior, potentially negatively impacting learning and posing risks to vulnerable students. While acknowledging some benefits, the author emphasizes the potential of AI as a research tool over its over-reliance as an educational tool.

(resobscura.substack.com)

AI

The Bitter Lesson: A Paradox in AI Development

2025-08-02

Rich Sutton's "bitter lesson" posits that general methods leveraging computation are ultimately the most effective. This article explores this idea's manifestation in fields like Go, chess, speech recognition, and computer vision, and its challenges in enterprise applications. While massive computation yields breakthroughs in some areas, the article highlights limitations in data quality and clearly defined objectives, arguing that efficient specialized models sometimes outperform general-purpose ones, and that computational resources aren't always the optimal solution.

(www.dbreunig.com)

AI

Anthropic Cuts OpenAI's Access to Claude API

2025-08-02

Anthropic revoked OpenAI's access to its Claude models' API, citing violations of its terms of service. OpenAI allegedly used the API for internal testing, benchmarking Claude's coding and creative writing capabilities, and assessing its responses to safety prompts involving CSAM, self-harm, and defamation. Anthropic stated this violated clauses prohibiting using the service to build competing products or reverse engineer its services. OpenAI expressed disappointment, highlighting that evaluating other AI systems is industry standard and noting its API remains open to Anthropic. This incident underscores the intensifying competition among tech giants and the complexities surrounding AI model access and terms of service.

(www.wired.com)

AI API access

Native Sparse Attention: Hardware-Aligned and Natively Trainable

2025-08-02

Long-context modeling remains a challenge in NLP. This ACL 2025 paper introduces NSA, a Natively trained Sparse Attention mechanism. NSA cleverly combines algorithmic innovations with hardware-aligned optimizations. Using a dynamic hierarchical sparse strategy (coarse-grained token compression and fine-grained token selection), it achieves significant efficiency gains while preserving global context awareness and local precision. NSA enables end-to-end training, reducing pre-training costs, and matches or exceeds Full Attention models across benchmarks, showing substantial speedups on 64k-length sequences in decoding, forward, and backward propagation.

(aclanthology.org)

AI long-context modeling

AI: Floor Raiser, Not a Ceiling Raiser

2025-08-01

This article explores AI's impact on learning and work. AI lowers the barrier to entry for acquiring new skills, but mastery remains challenging. In coding, AI significantly aids managers but offers limited help with large codebases. AI's impact on creative fields is minimal, as novelty is crucial. For areas with established apps (e.g., email, food delivery), AI's influence is negligible. In essence, AI raises the floor for knowledge work, but its impact isn't uniform, varying greatly depending on the individual and their field.

(elroy.bot)

AI

Gemini Embedding: Powering the Next Generation of AI Agents

2025-08-01

Since its release, Google's Gemini Embedding text model has seen rapid adoption by developers building advanced AI applications. Beyond traditional uses like classification and semantic search, it's crucial for 'context engineering,' providing AI agents with complete operational context. Companies like Box, re:cap, Everlaw, Roo Code, Mindlid, and Interaction Co. are already leveraging its power to improve accuracy, speed, and contextual awareness in their products. From boosting financial data analysis to enhancing legal discovery and powering AI assistants, Gemini Embedding's high performance and multilingual support are laying the foundation for the next generation of intelligent agents.

(developers.googleblog.com)

AI context engineering

Open-Source Image Model FLUX.1-Krea [dev]: Breaking Free from the 'AI Look'

2025-08-01

We're releasing the open-source version of FLUX.1-Krea [dev], our first image model trained in collaboration with Black Forest Labs. This model prioritizes aesthetic control and image quality, seamlessly integrating with the existing FLUX.1-dev ecosystem. Unlike most image models, FLUX.1-Krea was developed with specific aesthetic preferences in mind, rather than solely focusing on technical benchmarks. This technical report details the model's development, including insights into pre-training and post-training, and future research directions. The key focus is on overcoming the common 'AI look' in generated images – blurry backgrounds, waxy skin textures, etc. – achieving high-quality results aligned with human aesthetic standards through curated datasets and reinforcement learning.

(www.krea.ai)

AI

GEPA: Language-Based Reflection Outperforms RL in AI Prompt Optimization

2025-07-31

Researchers introduce GEPA, a novel algorithm for optimizing prompts in complex AI systems. Unlike traditional reinforcement learning (RL), GEPA uses a language-driven evolutionary approach. An LLM analyzes its own performance—reasoning, tool usage, and feedback—to identify and fix errors. GEPA significantly outperforms RL methods, using far fewer system executions while achieving better results across various tasks. This highlights the potential of language-based self-reflection for efficient AI optimization.

(arxiviq.substack.com)

AI AI optimization

AI Cracks CAPTCHAs: The Never-Ending Arms Race

2025-07-31

The ChatGPT Agent AI tool recently bypassed Cloudflare's Turnstile bot-detection system, accessing websites without solving image CAPTCHAs. This isn't the first time AI has cracked CAPTCHAs; it's the latest development in an ongoing arms race. Originally designed to distinguish humans from machines, CAPTCHAs have evolved into a method to slow down or increase the cost of bot attacks, even leading to the rise of human CAPTCHA-solving farms. The race continues, with AI and anti-AI technologies locked in a perpetual struggle.

(arstechnica.com)

AI

Category: AI