Search the archive

👉 Run your agent with the fastest inference layer

GOOGLE MIGHT BE TESTING GEMINI FLASH UPGRADE ON LM ARENA

Google is testing a new Gemini Flash upgrade on LM Arena, reportedly offering incremental improvements over the current Flash version. Despite no official confirmation, past Arena tests have often preceded public launches, drawing attention to potential new labels like "Gemini 3.6 Flash" or "Gemini 4 Flash." The Flash tier attracts interest due to its importance for everyday users and cost-conscious developers, handling most free and pay-as-you-go traffic faster than pricier Pro models.

META IS PLANNING A CLOUD BUSINESS TO SELL AI COMPUTING POWER

Meta Platforms is developing an internal cloud infrastructure initiative to sell access to its surplus AI computing power and hosted models to external developers. This strategic pivot aims to generate a new revenue stream from the company's massive multi-billion-dollar investments in data centers and chips, directly challenging dominant hyperscalers like Amazon Web Services, Microsoft Azure, and Google Cloud.

THE WINNING ESSAYS FOR THE BIG QUESTIONS ABOUT AI

Dwarkesh Patel posted some big questions about AI a couple of months ago and received 600 essay submissions for the contest. The three winners of the competition were Jassi Pannu, an Assistant Professor at Johns Hopkins University; Ege Erdil, the co-founder of Mechanize; and Michael Li, a Master of Public Policy candidate at Harvard Kennedy School. The full essays are available in the post.

AUTORESEARCH: THE FEEDBACK LOOP BEHIND SELF-IMPROVING AGENTS

Autoresearch involves building an 'outer loop' where agents help maintain and improve the primary system using feedback signals, evals, and human input to make progress over time. One of the startups building infrastructure for these self-improving systems is Introspection. This post features an interview with Introspection's co-founder and CEO, Roland Gavrilescu. The discussion covers the shift from agent harnesses to feedback loops, the role of the open-source Pi framework, and why autonomous software factories must first learn from humans.

CONTINUAL HARNESS: AN EFFICIENT SELF-IMPROVING AGENT ON ARC-AGI-3

The ARC-AGI-3 benchmark tests agents' ability to form and update an internal world model through ongoing learning. Continual Harness aims to efficiently enhance agents' self-improvement capabilities on this challenging IQ test.

LEARNING TO REPLICATE EXPERT JUDGMENT IN FINANCIAL TASKS

Frontier models struggle with relatively simple financial tasks. Custom models fine-tuned on high-quality proprietary datasets labeled by expert investors perform better. They are also substantially cheaper. Models of the future will probably feature differentiated intelligence - organizations will have custom models tuned to their specific needs that outperform frontier models.

ZCODE

ZCode is now available on macOS, Windows, and Linux. It combines the best AI agents with existing tools so developers can plan, code, review, and deploy without friction. GLM-5.2 is tuned for ZCode, making agentic coding faster and steadier. GLM Coding Plan subscribers now have 1.5x usage quota in ZCode.

PORTAL: PORTABLE TASK ADAPTERS FOR LLMS

PorTAL is an innovative architecture designed to decouple task fine-tuning from specific base model weights. As the accelerating cadence of foundation model releases creates a compounding financial and engineering burden to re-tune capabilities from scratch, PorTAL offers a way to pay for task adaptation once and amortize it across all future models.

REDEPLOYING FABLE 5

Claude Fable 5 and Mythos 5 have been redeployed. Fable 5 will be included for up to 50% of weekly usage limits until July 7, after which it will be available via usage credits. Mythos 5 has been restored for some organizations in the US. Anthropic is continuing to coordinate with the US government to expand Mythos access to the broader set of domestic and international partners in the Glasswing program.

A NEW LOOK AT AI'S IMPACT ON JOBS: FIRM-LEVEL AI SPENDING AND WORKFORCE ADJUSTMENT

A joint study by Ramp and Revelio Labs analyzed the relationship between firm-level generative AI investments and employment outcomes across over 21,000 companies in the US. The research reveals that companies with high-intensity AI spending grew their overall headcount by 10.2 percent and entry-level positions by 12 percent over the two years following adoption.

HOW OPENAI DELIVERS LOW-LATENCY VOICE AI FOR 900M USERS

This article looks at how OpenAI uses WebRTC to run voice for its 900 million users.

PRODUCT SHAPE IS THE MOAT

Application layer AI companies cannot build sustainable moats against model providers through simple technical adjustments like fine tuning or model routing.

OPENAI PROPOSES 5% STAKE TO TRUMP ADMINISTRATION TO EASE WASHINGTON PRESSURE

The proposal is part of a larger arrangement under which the US government would hold 5% of each of the leading US AI developers through a sovereign wealth fund vehicle.

KOTO REDESIGNS STACK OVERFLOW TO CHAMPION HUMAN KNOWLEDGE IN THE AI ERA

Koto has rebranded Stack Overflow around the idea that trustworthy, human-validated knowledge is the foundation of effective AI, repositioning the platform from a Q&A site to an essential resource for developers and enterprises. The new identity centers on the concept of “Always in build,” introducing an evolved logo, a stack-based visual system, the custom Stack Sans typeface, AI-assisted design tools, and a clearer tone of voice that reflects the collaborative, iterative nature of the developer community. Developed with community feedback and refined after its 2025 debut, the rebrand aims to reinforce Stack Overflow's role as a living source of reliable technical knowledge in the AI era.

WATCHOS 27 HAS BRAND NEW DEFAULT HOME SCREEN FOR APPLE WATCH

watchOS 27 introduces the Dynamic app grid as the new default Apple Watch Home Screen, placing the Siri app at the centre and surrounding it with AI-powered suggestions based on your most-used and recently used apps. The traditional Grid and List views remain available but are now accessed through a shortcut at the bottom of the dynamic grid. Apple aims to make app access faster and more relevant, with early beta testing suggesting the recommendations work well.

GOOGLE RELEASES NANO BANANA 2 LITE, ITS FASTEST AND CHEAPEST AI IMAGE GENERATOR YET

Google's Nano Banana 2 Lite is an image model that can generate visuals in four seconds for under four cents per thousand. It is now available in AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform. Built for speed over fidelity, it replaces the original "legacy" model while retaining prompt adherence, character consistency, and text rendering. Nano Banana 2 Lite is rolling out across consumer products like Search, Gemini, and Photos. Alongside it, Gemini Omni Flash video generation is now open to developers at ten cents per second. The model is a companion pipeline that turns generated images into short clips.

YOU'RE MISSING THE POINT; THIS IS YOUR REAL VALUE IN TECH COMPANIES

AI is most valuable when it complements human expertise rather than replaces it, as businesses that rely solely on automation risk creating generic products while losing the strategic, creative, and problem-solving strengths people provide. A comparison between AI-only, human-only, and hybrid approaches shows that the strongest outcomes come from combining AI's speed and scalability with human judgment, originality, and vision. As AI reshapes industries, the most valuable skills are no longer just knowing how to use the latest tools, but applying critical thinking, strategy, and creativity to solve meaningful problems.

THE STATE OF THE CREATIVE INDUSTRY 2026: WHAT OUR SURVEY TELLS US ABOUT PAY, BURNOUT AND AI

Creative Boom's 2026 survey of 882 creative professionals reveals an industry under significant strain, with 69% reporting burnout, widespread financial insecurity, and 86% using AI despite only 10% believing it has a positive impact. Many creatives feel pressured to adopt AI while worrying about lower fees, increased competition, and the future of the profession, with nearly half of freelancers earning under £30,000 a year and 38% considering changing jobs. Rather than seeking more technology, respondents identified stronger professional communities, networking, and mentorship as the most valuable ways to improve their careers and wellbeing.

FINDING PHOTOS IS SO MUCH EASIER WITH SIRI AI IN IOS 27 THAT I NO LONGER SCROLL

iOS 27's Siri AI improves photo search by letting users find images with natural-language voice commands, making it much easier to locate specific photos in large libraries without endless scrolling. It generally performs well when searching for objects, people, or clothing, and allows users to quickly share results or open them in the Photos app, though it can still make mistakes with more complex or time-based queries. Despite some inaccuracies in the beta, the feature is a significant improvement over previous photo search capabilities and saves considerable time.

CREATE VIRAL ADS BY CHATTING

Turn an idea into ready-to-run automated video ads for TikTok, Meta, and YouTube with the ultimate AI ad generator.

V0 TEMPLATE GALLERY

Discover the best apps, components, and starters from the v0 community.

THE FUTURE OF UX IN AN AI WORLD: WHY DESIGNERS ARE BECOMING STRATEGIC LEADERS

AI is reshaping digital products, pushing UX designers beyond interface work into strategic decisions about trust, autonomy, and how humans and AI should interact. As uncertainty grows around AI-powered features, user research becomes essential for uncovering how people actually behave, rather than relying on assumptions. This shift is opening new career paths, with designers gaining greater influence over product direction, roadmap planning, and cross-functional decision-making.

DESIGNED FOR A DEAD LANGUAGE

Modern language-learning apps like Duolingo and Anki inherited their grammar-drill structure from the Prussian Grammar-Translation method, originally designed in 1788 to teach Latin at scale. That method conflated literacy with fluency, prioritizing gradable metrics over actual communication, a flaw later articulated by linguist Stephen Krashen's distinction between acquisition and learning. Newer tools like Praktika, Langua, and Rosetta Stone instead remove access barriers to immersive, conversation-based learning, which research suggests produces stronger speaking results.

39 PRINCIPLES FOR DESIGNING HUMAN-AI INTERACTION

Designing AI products requires rethinking traditional UI principles because AI is probabilistic rather than deterministic, making user trust, transparency, control, and recoverability as important as model capability. The framework presents 39 design principles covering areas such as setting clear expectations, showing evidence instead of confidence scores, keeping users in control, making AI outputs editable and verifiable, designing for graceful failure, and limiting autonomy based on risk. The overarching goal is to help users rely on AI appropriately by treating it as a collaborative tool whose decisions are understandable, inspectable, and always subject to human oversight.

CREATIVITY LESSONS FROM THE EIFFEL TOWER

The Eiffel Tower offers four creativity lessons drawn from a book on creative thinking, starting with how its design built on earlier works and multiple collaborators rather than solitary genius.

WHY ARE WE ALL OBSESSED WITH ANTI-DESIGN?

Anti-design has resurged over the past decade as a reaction to minimalism's polished, white-space-heavy aesthetic that grew tiresome to audiences.

WHAT TO LEARN TO BE A REAL TIME GRAPHICS PROGRAMMER

To become a hireable graphics programmer, one must understand both the CPU and GPU sides of modern rendering, understanding APIs like DirectX12 and Vulkan, as well as mathematical concepts related to lighting and rendering techniques. Learning about Physically Based Rendering (PBR) is necessary, as it standardizes lighting representation across varying conditions, thus streamlining asset creation. Familiarity with C++ is needed for CPU programming, while knowledge of shader languages like HLSL and GLSL can be beneficial.

META'S AI STORAGE BLUEPRINT AT SCALE

Meta's storage architecture has evolved to meet the demands of modern AI workloads, addressing challenges such as maximizing GPU utilization and speeding up research velocity. By rebuilding the metadata subsystem and adopting a tiered caching approach with prefetching and on-demand hydration, its new BLOB-storage system improves performance, reducing data ingestion times and facilitating more efficient access for researchers.

CLIENT-SIDE LOAD BALANCING AT A MILLION REQUESTS PER SECOND

A client-side load balancer was developed to handle over a million requests per second of internal traffic, migrating routing decisions from a shared edge load balancer, Skipper, to improve latency and system resilience while reducing costs. It included features like occupancy-based bounded load and AZ-aware routing strategies to improve performance and efficiency.

MOST REWRITES SERVE THE ENGINEER, NOT THE BUSINESS

Most code rewrites cater to the preferences of engineers rather than the actual needs of the business, leading to a loss of valuable historical context and solutions that reside in the existing code. A rewrite should only be pursued when there are measurable issues, such as outdated dependencies or technical debt, rather than personal style choices.

DARK MODE WITH WEB STANDARDS

To improve user experience with dark mode on websites, developers should respect the user's system preferences while also allowing custom toggles for per-site settings.

REDEPLOYING CLAUDE FABLE 5

Access to the AI models Claude Fable 5 and Mythos 5 was briefly suspended due to newly imposed US government export controls, but it has now been restored following the lifting of these restrictions. Anthropic is collaborating with government partners to improve cybersecurity measures and create a consensus framework for assessing and addressing potential vulnerabilities and “jailbreaks” in AI models.

ZCODE 3.0

ZCode 3.0 introduces an optimized platform for easy collaboration using AI agents, enabling developers to plan, code, review, and deploy.

OPENWIKI

OpenWiki is a CLI designed to write and maintain documentation for codebases specifically for agents. Users can initialize the tool to generate documentation, update existing documentation, and customize various model settings.

SANDBOXING AN AI AGENT

The increasing autonomy of AI agents raises security concerns as they gain the ability to execute commands without explicit human approval, creating potential risks associated with private data access and untrusted content. Implementing sandboxes provides isolated environments for these agents, improving containment, resource management, and recovery processes while allowing for experimentation and parallel execution without risking the host system.

HOW WE KEEP GPUS RELIABLE ACROSS DATABRICKS AI

GPU failures during distributed training can primarily be categorized into crashed jobs, silent slowdowns, and numerical corruption, each posing distinct challenges for maintaining reliable performance. To address these issues, a comprehensive health check system at Databricks AI uses both proactive stress testing of workloads and continuous monitoring to detect and mitigate failures across the GPU infrastructure.

A TALE OF TWO MODEMS

Apple plans to equip the US version of the iPhone 18 Pro with Qualcomm modem hardware to support mmWave technology, while the international versions will rely on Apple's proprietary C2 modem, which doesn't have mmWave capability. This choice may prioritize carrier marketing over user experience, as Qualcomm modems offer faster speeds at the expense of battery life.

OF COURSE META PLATFORMS IS GOING TO BE A CLOUD

ARXIV'S NEXT CHAPTER: UPDATES ON OUR SPIN OUT FROM CORNELL UNIVERSITY

arXiv will become an independent nonprofit organization on July 1 after 25 years at Cornell University.

META CAPS INTERNAL AI TOKEN SPENDING AFTER COSTS APPROACH BILLIONS IN 2026

Meta is implementing spending controls on employee AI usage after internal costs surged towards billions in 2026, driven by excessive token consumption tracked on a leaderboard.

BUILDING GIN: SIMPLE OVER EASY

Gin is a Go web framework that balances simplicity with functionality, avoiding the complexities of reflection and excessive magic found in other frameworks while ensuring backward compatibility and high performance through efficient routing and explicit control.

ONE IN FOUR AMERICANS DISCOVER PRODUCTS THROUGH INFLUENCERS IN 2026

TV and radio advertising reach 41% of Baby Boomers+ as a product discovery channel but only 19% of Gen Z. Social media influencers and bloggers reach 41% of Gen Z but only 13% of Boomers+. The inversion is nearly perfect, and it shows up in brand performance as well. Gen Z consumers who discover products through influencers show higher consideration across categories, with Dunkin' at 44% versus 38% among all Gen Z and e.l.f. Cosmetics at 28% versus 21%. Across all US adults, 27% now cite influencers as a discovery channel, ranking 9th overall behind friends and family at 51%, retail browsing at 44%, and search at 42%.

STUDY: THE GREAT BLOGGING COLLAPSE

Most once-successful blogs have seen steep declines following Google's Helpful Content Updates and the rise of AI-generated search results. In a study of 100 blogs, median traffic fell 85%, with only 21 still growing. Blogs that rely solely on search traffic and affiliate monetization are struggling, while those offering original, experience-based content perform better.

WHAT DOES YOUR LOGO SOUND LIKE?

Sonic branding is becoming more important as attention shifts off-screen, but most brands still rely on visual identity. Sound can drive stronger recall by tapping into emotional processing and working when users aren't looking. Building a consistent audio system across touchpoints helps reinforce recognition over time. As more media consumption happens in audio-first environments, brands without a defined sonic identity may struggle to stay memorable.

HOW A BRANDED SUBREDDIT HELPS CURSOR OWN BOFU DISCUSSIONS

Cursor turned its branded Reddit community into a high-intent marketing channel by answering buyer questions where purchase decisions already happen. Its subreddit has 143,000 members, ranks for 482 keywords, and brings in about 9,600 monthly organic visits by appearing for pricing and comparison searches that often rank above its own website. Senior employees answer pricing, billing, and product questions in public, which builds search visibility, gives AI tools accurate content to cite, and creates a steady source of customer feedback. The strategy only works when the community trusts the moderation, so it keeps constructive criticism visible instead of removing negative posts.

BUILD YOUR AI BRAIN

Building a personal “intelligence layer” helps marketers capture insights and experiences instead of letting them stay scattered across tools, roles, or past work. This knowledge often gets lost or reset when using AI, creating constant rework. By continuously feeding it into a centralized system, it becomes reusable, improves outputs, and compounds over time. The post includes a screenshot showing how this can look in practice.

HOW SONY SOLD 400 MILLION WALKMANS

Sony built the Walkman by removing features instead of adding them. It cut recording even though research and retailers said customers wanted it. That decision lowered the price to $150, made the device smaller, and gave it a clear purpose. The cassette Walkman sold 200 million units and the Walkman brand reached about 400 million sales across all formats. Use customer data to understand existing behavior, but test new ideas by the behavior you want people to adopt instead of the feature list they say they want.