Hasty Briefsbeta

All tags

#ai

2203 stories total

Bilingual

IBM Replaced HR Workers with AI
a year ago
- IBM CEO Arvind Krishna confirmed the company replaced hundreds of HR employees with AI.
- IBM's AskHR agent automates 94% of routine HR tasks, including vacation requests and pay statements.
- The company reallocated resources to hire more employees in software engineering, marketing, and sales.
- IBM's AI implementation led to a $3.5 billion productivity improvement over two years.
- IBM's workforce grew to 270,300 employees globally as of its 2024 annual report.
- IBM introduced new generative AI tools at its Think conference, aiming to expand its AI division.
- Other companies like Klarna and Salesforce have also replaced human roles with AI.
The AI jobs crisis is here, now
a year ago
- Duolingo CEO announces shift to an 'AI-first' approach, reducing reliance on human contractors.
- Reports contradict the CEO's statement, revealing that Duolingo has already replaced 100 workers, including writers and translators, with AI.
- Affected workers express shock and disappointment, noting AI's current limitations in quality and creativity.
- The AI jobs crisis is already impacting various sectors, including creative industries, journalism, and voice acting.
- Recent college graduates face unusually high unemployment rates, potentially linked to AI replacing entry-level white-collar jobs.
- Economic trends suggest AI investment may be crowding out spending on new hires, exacerbating job market challenges.
- The AI jobs crisis is characterized by gradual attrition in creative fields and reduced hiring, rather than mass layoffs.
- Critics argue that AI is being used to automate creative jobs instead of dull tasks, contrary to initial expectations.
- Despite hype, enterprise AI adoption shows mixed results, with many companies reporting disappointing returns on investment.
- The broader implications of the AI jobs crisis call for societal reflection on the value of work and the need for protective measures.
Klarna changes its AI tune and again recruits humans for customer service
a year ago
- Klarna is shifting back to human customer service after initially relying heavily on AI.
- The company now emphasizes the importance of human empathy alongside AI efficiency.
- Klarna is piloting a new customer service model with flexible, remote roles for highly educated individuals.
- AI still handles two-thirds of inquiries, improving response times and reducing repeat issues.
- Customers prefer human interaction for complex or sensitive issues, highlighting AI's limitations.
- Klarna acknowledges past over-reliance on AI for cost-cutting, now focusing on quality human support.
- Surveys show widespread customer frustration with chatbots, especially when they fail to resolve issues.
- Human empathy is deemed more important than quick responses in customer service.
Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL
a year ago
- INTELLECT-2 is the first 32B parameter model trained via globally distributed reinforcement learning.
- The model uses PRIME-RL, a training framework for distributed asynchronous reinforcement learning, with components like TOPLOC and SHARDCAST.
- Training data includes 285k verifiable tasks from NuminaMath-1.5, Deepscaler, and SYNTHETIC-1.
- The model improves upon QwQ-32B with modifications to the GRPO training recipe and advanced data filtering techniques.
- Future work includes increasing inference-to-training compute ratio, tool calls, crowdsourcing RL tasks, and model merging.
- The article also includes a detailed math problem solution involving quadratic polynomials P(x) and Q(x).
US Copyright Office Has Thoughts on AI. Big Tech May Not Like It
a year ago
- Big Tech companies rely on others' content to train AI models, raising copyright concerns.
- The US Copyright Office's new report questions if AI training falls under fair use, depending on usage specifics.
- AI companies argue training is transformative, but the office distinguishes between research and commercial uses.
- The report rejects arguments that AI training is inherently transformative like human learning or non-expressive.
- Shira Perlmutter, the Copyright Office director, was fired by Trump shortly after the report's release, sparking controversy.
- Tech leaders, including Elon Musk, have supported Trump, aligning with his federal spending reduction efforts.
Trump fires head of Copyright Office after rpt AI training may not be fair use
a year ago
- Shira Perlmutter, head of the US Copyright Office, was fired by the Trump administration.
- The firing followed the release of a report questioning the fair use of copyrighted data for AI training.
- The report suggested commercial AI tools using copyrighted data might violate fair use laws.
- Representative Joe Morelle criticized the firing as an 'unprecedented power grab' linked to the report.
- Legal experts noted the report could be a setback for AI companies relying on copyrighted data.
- President Trump also fired Librarian of Congress Carla Hayden, whose department oversees the Copyright Office.
- The White House provided vague reasons for Hayden's dismissal, citing concerns over DEI initiatives and children's books.
Show HN: Vom Decision Platform (Cursor for Decision Analyst)
a year ago
- No-code editor with Excel-formula powered decision flow for branching, variable calculation, and decision tables.
- REST API integration compatible with OpenAPI for easy system integration including legacy, ERP, CRM, and event buses.
- AI Copilot for task automation and policy improvement suggestions.
- Connectors with popular data providers and services, including caching and cost-saving data fetch skipping.
- Policy A/B testing for risk-controlled iterations and continuous improvement.
- Versioning system with logging of modifiers and approvers for consistency and auditability.
- Security features include API keys authentication, encrypted databases, and a secret manager.
- Scalable cloud application capable of online decisions in under 30ms and millions of batch executions in minutes.
- Decision History logs every decision for auditing, debugging, and analysis.
- Vom platform supports various applications like credit engine, anti-fraud, approval automation, and dynamic pricing.
- Vom's vision is to help companies thrive in a data-driven, automated future with stricter regulations and AI/ML support.
- Vom focuses on user-friendliness, performance, and security using the latest technology.
- Backed by industry leaders, Vom offers solutions for companies of all sizes.
Show HN: Vibemail – Chrome extension for vibe emailing
a year ago
- VibeMail enhances Gmail with AI-powered writing assistance.
- Features include real-time AI suggestions and tab completions.
- Private context comments can be added using //.
- Shortcuts like Shift + Tab or Shift + X enhance usability.
- Seamless integration within Gmail without switching tools.
- Developer information and privacy policy details provided.
Build Your Own Siri. Locally. On-Device. No Cloud
a year ago
- The article discusses the potential of running AI models locally on devices for privacy and efficiency.
- It introduces a mini-course on building a local voice assistant that understands natural language and executes app functions offline.
- The course covers fine-tuning LLaMA 3.1 8B with LoRA, creating a function-calling dataset, and running inference locally.
- It emphasizes the importance of MLOps principles even for local-only AI systems to ensure reliability and avoid silent failures.
- The system overview includes dataset generation, instruction tuning for function calling, and testing the model in the MAC ecosystem.
- The article is aimed at developers building privacy-first mobile apps and teams deploying apps in sensitive environments.
- The mini-course is free and includes hands-on lessons with a GitHub repository for practical implementation.
Welcome to the age of paranoia as deepfakes and scams abound
a year ago
- Nicole Yelland conducts thorough background checks on unknown contacts due to past scam experiences.
- Digital imposter scams are increasing, with AI tools making fake personas easier to create.
- Reports of job-related scams nearly tripled from 2020 to 2024, with losses rising from $90M to $500M.
- Scammers impersonated a real company and requested personal details during a fake job interview.
- AI startups like GetReal Labs and Reality Defender are emerging to detect deepfakes.
- Old-fashioned verification methods, such as code words and time-stamped selfies, are being used to confirm identities.
- Recruitment verification services use rapid-fire questions and the 'phone camera trick' to spot fraudsters.
- Researchers face challenges with fraudulent survey participants, leading to stricter screening methods.
- Common sense and scrutiny of too-good-to-be-true job offers can help identify scams.
Show HN: Helixdb – Open-source vector-graph database for AI applications (Rust)
a year ago
- HelixDB is an open-source, high-performance graph-vector database written in Rust, designed for RAG and AI applications.
- Powered by LMDB via Heed3, it offers reliability and modern features tailored for AI and vector-based applications.
- Performance highlights include being 1000x faster than Neo4j, 100x faster than TigerGraph, and on par with Qdrant for vectors.
- Features native support for graph and vector data types, making it ideal for RAG applications.
- Provides the easiest way to store relationships between nodes, vectors, or both.
- Ensures data integrity with ACID compliance and robust storage via LMDB.
- Includes a CLI tool for installation, setup, query compilation, and deployment.
- Supports TypeScript and Python SDKs for querying the database.
- Current focus areas include expanding vector data type capabilities, enhancing query language robustness, and implementing a test suite.
- Long-term projects include developing an in-house graph-vector storage engine and network protocol libraries.
- Licensed under AGPL, with managed services and enterprise support available for selected users.
Databricks and Neon
a year ago
- Databricks announces acquisition of Neon, a serverless Postgres company.
- Neon's architecture separates storage and compute, designed for modern developers and AI systems.
- Neon's team includes Postgres experts and database veterans.
- Neon was founded to disrupt the database industry with a focus on developer experience.
- Neon's architecture enables features like database branching similar to Git.
- AI agents create over 80% of databases on Neon, up from 30% last year.
- Neon's founders have strong backgrounds in databases and open source.
- Databricks and Neon share a commitment to open source and technical innovation.
- OLTP databases are a $100B market ripe for disruption by developers and AI.
- Existing Neon customers will continue to receive support and innovation under Databricks.
- Databricks plans to share more at the Data + AI Summit in June.
Show HN: acmsg (automated commit message generator)
a year ago
- A CLI tool in Python for generating git commit messages using AI models via OpenRouter API.
- Analyzes staged changes in git repositories to generate contextual commit messages.
- Supports multiple AI models through OpenRouter and allows editing of generated messages.
- Automatically commits changes with the AI-generated message upon confirmation.
- Requires an OpenRouter API key for operation.
- Installable via pipx, NixOS flakes, or standalone Nix profiles.
- Configuration file located at ~/.config/acmsg/config.yaml.
- First run prompts for OpenRouter API token configuration.
- Licensed under the MIT License.
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
a year ago
- AlphaEvolve is a Gemini-powered coding agent designed for discovering and optimizing algorithms.
- It combines large language models (LLMs) like Gemini Flash and Gemini Pro with automated evaluators to verify and improve algorithmic solutions.
- AlphaEvolve has enhanced efficiency in Google's data centers, chip design, and AI training processes.
- It has contributed to faster matrix multiplication algorithms and solutions to open mathematical problems.
- AlphaEvolve uses an evolutionary framework to improve upon promising ideas, verified by automated evaluation metrics.
- The agent has been deployed in Google's computing ecosystem, improving data center scheduling, hardware design, and AI training.
- AlphaEvolve discovered a heuristic for data center scheduling that recovers 0.7% of Google's worldwide compute resources.
- It optimized a Verilog rewrite for matrix multiplication circuits in upcoming Tensor Processing Units (TPUs).
- AlphaEvolve sped up matrix multiplication in Gemini's architecture by 23%, reducing training time by 1%.
- It achieved a 32.5% speedup for the FlashAttention kernel in Transformer-based AI models.
- AlphaEvolve advanced mathematical frontiers, improving solutions to problems like the kissing number problem in 11 dimensions.
- An Early Access Program is planned for academic users, with potential broader availability in the future.
- AlphaEvolve's general nature allows application to any problem solvable via algorithms, with potential in material science, drug discovery, and sustainability.
DeepMind unveils 'spectacular' general-purpose science AI
a year ago
- Google DeepMind developed AlphaEvolve, a system combining large language models (LLMs) with algorithms to solve major problems in mathematics and computer science.
- AlphaEvolve has practical applications, improving Google's tensor processing unit designs and optimizing global computing resource usage, saving 0.7% of total resources.
- Unlike specialized AI tools, AlphaEvolve is general-purpose, leveraging LLMs to generate solutions across diverse scientific domains.
- The system evolves solutions by iteratively refining modifications proposed by the LLM, guided by evaluator algorithms.
- AlphaEvolve outperformed specialized tools like AlphaTensor, discovering a faster method for matrix multiplication in some cases.
- The system builds on DeepMind's earlier FunSearch, expanding capabilities to handle more complex algorithms and larger codebases.
I built a type-safe .NET casting library powered by AI
a year ago
- ArtificialCast is a lightweight, type-safe casting and transformation utility powered by large language models (LLMs).
- It enables seamless conversion between strongly typed objects using type metadata, JSON schema inference, and prompt-driven reasoning.
- Key features include zero configuration, bidirectional casting, schema-aware inference, and LLM-powered transformation.
- ArtificialCast is part of the BIGPISS stack, which replaces conventional logic with inference, structure, and prompt design.
- The tool is fully functional but fundamentally unsafe, as it generates outputs that may appear correct but are not guaranteed to be accurate.
- It is intended for educational and demonstrative purposes only, not for production use.
- Requirements include .NET 9 SDK or later, Ollama installed locally, and a local model like gemma3:4b.
- The project serves as a satire of emerging AI engineering trends, highlighting the dangers of trading correctness for convenience.
MinorMiner: We turn your kid's maths homework into Bitcoin
a year ago
- MinorMiner is a platform that converts children's math homework into Bitcoin mining operations.
- The platform breaks down complex Bitcoin mining algorithms (SHA-256) into simple arithmetic problems for children to solve.
- Children (referred to as 'computation partners') complete math quizzes, which are then used to perform hash calculations for Bitcoin mining.
- The system uses a Python library called CUDAAAAGH to distribute computations across children globally.
- MinorMiner plans to expand into AI training by leveraging children's math skills for matrix algebra computations.
- The company aims to optimize speed through parallelization, curriculum changes (teaching XOR operations), and teacher incentives.
- Future goals include creating a 'human computer' where children act as hardware components like CPUs and memory.
Grok wants people to know claim of white genocide in S.A. is contentious
a year ago
- Grok, an AI model from Elon Musk’s xAI, sent strange responses on X, focusing on debunking white genocide claims in South Africa even in unrelated posts.
- Grok repeatedly emphasized that claims of white genocide in South Africa lack credible evidence, citing official data and court rulings.
- The AI’s responses have since been fixed to address the actual queries, but X has not commented on the issue.
- The incident followed news of Afrikaners receiving refugee status under a Trump executive order, which cited alleged genocide—a claim Grok disputed.
StackAI (YC W23) Is Hiring Pydantic and FastAPI Wizard
a year ago
- Vision: Democratize access to Large Language Models for positive impact.
- Product: No-Code platform integrating AI models, data sources, and SaaS.
- Traction: 85,000+ users, 300+ paying customers, launched 16 months ago.
- Team: Founded by 2 MIT PhDs, backed by Y Combinator and Google.
- Values: Default to action, transparency, boldness, strong opinions loosely-held.
- Responsibilities: Develop backend systems, manage databases, integrate AI models.
- Required Skills: Python, FastAPI, SQL/NoSQL, AWS, Docker, Kubernetes.
- Preferred Skills: AI model integration, RAG pipelines, embedding generation.
- Culture: Fast-paced, user-focused, collaborative, with opportunities for learning.
LLMs are making me dumber
a year ago
- LLMs are being used to shortcut learning processes, such as coding and math homework, leading to less depth in understanding.
- There is a trade-off between output speed and depth of learning, with urgency driven by rapidly improving AI models.
- Historical analogies like calculators and GPS suggest some skills can be offloaded, but intelligence is harder to confine.
- Arguments against LLMs include becoming just a wrapper for models and potential stagnation if models don't improve as expected.
- Arguments for LLMs include significant short-term output gains and the need to act quickly in a fast-evolving field.
- Using LLMs as tutors could balance learning and output, but repetitive tasks are crucial for ingraining skills.
- Finding balance involves automating small tasks while preserving critical thinking and long-term project skills.

first prev12next

About|Login

#ai

IBM Replaced HR Workers with AI

The AI jobs crisis is here, now

Klarna changes its AI tune and again recruits humans for customer service

Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL

US Copyright Office Has Thoughts on AI. Big Tech May Not Like It

Trump fires head of Copyright Office after rpt AI training may not be fair use

Show HN: Vom Decision Platform (Cursor for Decision Analyst)

Show HN: Vibemail – Chrome extension for vibe emailing

Build Your Own Siri. Locally. On-Device. No Cloud

Welcome to the age of paranoia as deepfakes and scams abound

Show HN: Helixdb – Open-source vector-graph database for AI applications (Rust)

Databricks and Neon

Show HN: acmsg (automated commit message generator)

AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms

DeepMind unveils 'spectacular' general-purpose science AI

I built a type-safe .NET casting library powered by AI

MinorMiner: We turn your kid's maths homework into Bitcoin

Grok wants people to know claim of white genocide in S.A. is contentious

StackAI (YC W23) Is Hiring Pydantic and FastAPI Wizard

LLMs are making me dumber