Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
GPT-5 rarely screws up and generally feels competent. The model used in ChatGPT is a hybrid that switches between different models. GPT-5 is available as three models in the API: regular, mini, and nano. Each can be run at one of four reasoning levels. This post looks at the key characteristics of the models, how they are priced, and what can be learned from the GPT-5 system card.
Ignore the claims about vibe coding or that you don't need to know how to write code, and instead look for ways to augment what you do. Experiment with AI and discover what value you can extract - there's a lot of grift. Learn for yourself what works and what doesn't. AI in its current form won't ever replace hands-on-keyboards, and it won't replace engineering.
Orchid is a company that lets parents genetically test embryos before pregnancy. Parents can choose their future baby for as little as $2,500. This article features a transcript of an interview with the company's founder, where they debate the scientific, moral, and ethical implications of designing a 'healthy' child and what we lose in separating reproduction from sex. With Orchid's technology, parents in the future will not have to roll the dice with their child's health.
Intel's Chief Executive Lip-Bu Tan and some Intel directors disagree on topics as central as whether the company should stay in the manufacturing business or exit it entirely. Tan recently tried to raise new capital to acquire an artificial intelligence company, but efforts have been stalled by people on the board. Despite Trump unexpectedly calling for Tan's ouster, the company is so far standing by Tan, saying that the board and Tan are all deeply committed to advancing US national and economic security interests. While Intel's chip factories are a money loser, some view them as politically important because they help secure the US semiconductor supply chain.
AI coding assistants can actually be massively money-losing businesses.
Cursor CLI lets developers interact with AI agents directly from their terminal to write, review, and modify code.
Tesla's Dojo team has lost 20 workers to a recently formed startup, - remaining Dojo workers will be reassigned to other data center and compute projects within Tesla.
The National Highway Traffic Safety Administration has cleared Zoox to deploy its fleet of steering wheel-less electric pods for demonstration purposes.
The user experience benefits of local-first are undeniable.
The American economy has split into a rip-roaring AI economy and a lackluster consumer economy.
Sponsored
OpenAI has launched GPT-5, its most advanced AI model yet, now available to all ChatGPT users.
Similar to Claude Code and Gemini CLI, Cursor CLI brings AI coding assistance directly to the terminal and allows developers to switch seamlessly between command-line and editor-based AI workflows.
Elon Musk announced that X will begin inserting ads into Grok's AI responses, aiming to boost the platform's ad revenue using xAI's targeting tech and chatbot suggestions.
GPT-5 excels as a daily driver for most users, with API pricing that aggressively undercuts competitors by up to 12x. However, it is too cautious for writing feedback or for autonomous coding workflows, making it feel like a significant upgrade to an old paradigm rather than a leap forward for developers already using multi-agent tools like Claude Code.
GPT-5 marks the "stone age" for AI. It doesn't just use tools - it thinks with them, like humans first learning to shape stones changed everything. Testing showed it one-shotted dependency conflicts that stumped every other model by using yarn commands like Deep Research uses web search, iterating and reasoning through problems instead of just guessing. While it's worse at writing than GPT-4.5 (producing more "LinkedIn slop"), it's unequivocally the best coding model - creating entire production-ready websites with SQLite databases in one shot, where other models gave scaffolding or plans, automating software engineering from maybe 65% to 72% in one leap.
METR assessed whether OpenAI GPT-5 could pose catastrophic risks before it was externally deployed. This post provides detailed findings from METR's assessment. METR determined that GPT-5 does not have the prerequisite capabilities - by a large margin - to pose catastrophic risk.
Sponsored
YC-backed Momentic uses AI to automate web app testing. Write tests in Plain English ("the login button should be visible") and let AI execute the test. Join hyper-growth companies like Retool, Notion,
Notte is a web agent framework built for speed, cost-efficiency, scale, and reliability. It allows developers to rapidly build reliable web automation agents. Notte provides all the essential tools for building and deploying AI agents that interact seamlessly with the web. The full-stack framework combines AI agents with traditional scripting for maximum efficiency. It enables users to develop, deploy, and scale their own agents and web automations within a single API.
Octo is a very friendly open-source coding helper. It works with any OpenAI- or Anthropic-compatible LLM API. Octo allows developers to switch models at will mid-conversation when a particular model gets stuck. Users have the option to use custom-trained models to automatically handle tool call and code edit features from the main models. Octo has zero telemetry.
Identifying policy-violating content requires solutions capable of deep contextual and cultural understanding, an area that large language models (LLMs) excel at over traditional machine learning systems. However, fine-tuning models for such complex tasks requires high-fidelity training data that is difficult and expensive to curate at the necessary quality and scale. This post describes a scalable curation process for active learning that drastically reduces the amount of training data needed for fine-tuning LLMs while significantly improving model alignment with human experts. In experiments, the process reduced the scale of training data needed from 100,000 to under 500 training examples while increasing model alignment with human experts by up to 65%.
The technique uses two reward parameters during training: safety constraints that penalize policy violations by severity, and helpfulness maximization that rewards both direct compliance and informative refusals with safe alternatives. In tests, GPT-5 with safe-completions achieved higher safety scores than o3 on dual-use prompts while providing substantially more helpful responses, and when failures did occur, they were significantly less severe.
OpenAI's o3 beat Grok 4 in the Kaggle Arena AI Chess Exhibition Tournament. Gemini 2.5 Pro defeated o4-mini 3.5-0.5 for the bronze medal. Grok appears to struggle with end games. o3 could understand the endgame much better and came up with better moves, eventually checkmating Grok.
Stability AI introduced a new enterprise offering that delivers tailored generative AI models and workflows for creative production.
Anthropic credits its commitment to safety, the quality of its research, and its team of top researchers as reasons that talent chooses to go there.
AI isn't thinking and then outputting a solution - every token a large language model generates is selected from adjacent possibilities in probability space.
Anthropic named Hidetoshi Tojo as Head of Japan and plans to open a Tokyo office, expanding its presence in Asia.
President Trump will sign an executive order directing the Labor Secretary to review ERISA fiduciary guidance so that private equity, cryptocurrencies, real estate, and other alternative assets can be included in 401(k) and defined-contribution plans. The move marks a major victory for alternative asset managers after conditional 2020 and Biden-era guidance already allowed limited private market exposure. With Americans holding about $8.7 trillion in 401(k) assets, BlackRock and Empower are already preparing funds that allocate 5-20% to private investments. Proponents argue that broader access can improve diversification and long-term returns, despite higher fees, longer lockups, and transparency concerns.
Ripple is set to acquire Toronto-based stablecoin payments platform Rail for $200 million. The deal is expected to close in Q4. It is part of Ripple's strategy to deepen adoption of its newly launched RLUSD stablecoin. Rail will augment Ripple's global payments network by integrating fast, compliant stablecoin rails into its On-Demand Liquidity infrastructure. This move follows Ripple's prior bid for Circle. It is the latest development in the intensifying race among blockchain firms to control stablecoin payment flows. By combining Rail's payment expertise with its settlement network, Ripple aims to accelerate stablecoin innovation in cross-border finance.
Narrative is a perpetual market that allows users to speculate on trends and mindshare rather than more objective information like prices. Its goal is to make it possible to speculate on any piece of information. It is the latest entry to the InfoFi narrative. Narrative's testnet deployment is live on Monad. A private beta is available for interested users.
Boros is a new protocol from the Pendle team that lets users trade and hedge funding-rate yields on BTC and ETH via Yield Units (YUs), each representing the yield from one unit of the underlying asset until maturity. By allowing traders to go long or short YUs and providing isolated liquidity pools (exLSTs), it introduces a standardized, trustless way to fix or speculate on funding exposure. Over time, Boros aims to become a core pillar of onchain yield markets and extend beyond funding rates to bonds, equities, and other real-world assets. Initially capped at $10 million open interest per market and 1.2x leverage, Boros supports Binance BTCUSDT and ETHUSDT funding rates, with vaults offering PENDLE incentives, swap fees, and implied APR gains.
Onchain order-book data from Polymarket between April 2024 and April 2025 reveals persistent pricing inefficiencies. Leveraging heuristics and LLM-assisted semantic matching, this research classifies mispricings into intra-market “Market Rebalancing” and cross-market “Combinatorial” arbitrage, scanning 17,218 conditions. Opportunities surfaced in 7,051 single-condition and 662 multi-condition markets, with 11 logically linked market pairs enabling cross-market trades. Onchain activity shows arbitrageurs have already extracted about $40 million, highlighting both the scale of inefficiency and the corrective role of sophisticated traders in decentralized prediction markets.
Comet, Perplexity's browser agent, was put to the test of creating a crypto portfolio and investing on Coinbase. It collected market information using various news sources and pricing data and organized it into a digestible format. Comet then navigated to Coinbase and started making trades based on its analysis, marking the first agentic trading on Coinbase. The portfolio includes BTC, ETH, SOL, ADA, and XRP.
Stablecoins mark a foundational shift from batch-based legacy finance to real-time, programmable money, creating new markets and value layers. There are four core layers in the stablecoin stack: issuance (where trust lives), infrastructure, orchestration, and distribution. While most builders chase the middle layers, the most defensible value lies at the edges, specifically issuer network effects and end-user distribution scale. The stablecoin market cap is at $232B and projected to hit $400B by year-end.
Boros' YU tokens are fixed-rate instruments that let long and short traders hedge volatile funding rates. In bull markets, longs pay high funding - buying YU at a fixed rate locks in costs. Shorts in bear markets can sell YU to fix incoming funding gains. Basis traders (e.g., long spot + short perp) can use short YU to preserve favorable funding without closing positions, especially relevant for protocols like Ethena. While Boros is early and short-YU demand dominates, it reflects smart yield engineering for the growing perp ecosystem.
In a joint filing, the SEC and Ripple Labs agreed to dismiss their respective appeals, marking the official end of the long-running case.
Base announced it will start collecting tokenized posts on @baseapp to directly support creators.
Nasdaq-listed Fundamental Global has filed to raise up to $5 billion, with the majority of proceeds earmarked for Ethereum purchases as part of its treasury strategy.
Lido DAO has proposed using up to 70% of incoming assets for $LDO buybacks, targeting more efficient capital use of its $145M+ treasury (stables + stETH).
Character.AI's Feed is the world's first AI-native social platform featuring scrollable content with Characters, Scenes, and Streams. The platform transforms passive consumption into interactive creation, allowing users to remix storylines and build on others' content. Feed is rolling out on the mobile app with multimodal features including Chat Snippets, Character Cards, and Avatar FX video creation.
Meta is adding Instagram Map and post reposting to make Instagram more social and focused on real connections, shifting away from its video-heavy, algorithm-driven feed. While Instagram Map shares user location only with consent and limited updates, and post reposting boosts friends' content in the main feed, some users have raised privacy concerns, despite Meta's emphasis on user control.
Tinder faces declining user engagement with a 7% drop in paying users, prompting the company to introduce new features, including "modes" for different dating goals and college-specific options. The dating app is planning a broader redesign targeting Gen Z users that will feature AI-powered matching expansion and allow users to like specific profile parts. Despite flat revenue of $864 million in Q2, Match Group expects improved Q3 performance and announced $50 million in product development investments.
Digital twins are generative AI models that simulate specific individuals to predict their preferences and behaviors in new situations. These AI proxies can fill survey gaps, enable shorter studies, and forecast population trends by aggregating individual predictions. While showing promise for research efficiency, digital twins raise significant ethical concerns about consent, privacy, and misrepresentation.
UX hiring managers in 2025 prioritize strategic thinking and problem-solving over polished interfaces and tool proficiency. Essential skills include user research, data analysis, AI literacy, accessibility knowledge, and strong collaboration abilities. Portfolios should demonstrate real impact through case studies with clear decision-making processes and measurable outcomes.
macOS 26 has sparked backlash over design changes, particularly the revised Finder and Macintosh HD icons, with users criticizing inconsistencies in perspective, added visual clutter, and a lack of clarity. Apple's new layered glass icon style and enforcement of a uniform squircle shape—often resulting in blurry visuals and “gray box” placeholders for non-conforming third-party apps—has led to usability concerns and a perception that the Mac is becoming harder to navigate.
Rork builds complete, cross-platform mobile apps using AI and React Native.
Polarr offers applications to allow creators to make and distribute photos and video filters.
Namelix uses generative AI to create a short, brandable business name.