The deal would give Meta a 49% stake in the data-labeling startup and bring over co-founder Alexandr Wang to lead a new "superintelligence" lab aimed at outperforming OpenAI, Anthropic, and Google. The massive investment follows Llama 4's underwhelming launch, but it's unclear if the investment also includes greater access to Scale's training data created for other AI labs in addition to highly sought-after AI research talent.
Mistral AI's Magistral is a multilingual reasoning model designed for domain-specific tasks, offering enhanced interpretability and traceability. Available in two versions, Magistral Small and Magistral Medium, the model excels in multilingual reasoning and speeds up responses with its Flash Answers feature. It is ideal for diverse applications, from legal research to creative writing, providing transparent reasoning crucial for compliance in regulated industries.
Tech is spending $80 billion on AI infrastructure while skeptics say we're building another dot-com bubble. Who's winning? OpenAI doubled to $10 billion ARR in a year. Anthropic tripled to $3 billion. Random startups are hitting $100 million ARR in months. The median AI startup hits $2 million revenue in year one - that used to take years. Yes, Apple says LLMs still suck at reasoning. Yes, we might overbuild data centers. However, when the richest companies on Earth are footing the bill, their losses mean cheaper compute for everyone else.
A car wash tested two punch cards: one needed 10 stamps for a free wash, the other needed 12 but started with 2 stamps already filled. Same math, but the pre-stamped card got 2X more completions. That's the Endowed Progress Effect - we chase goals harder when we feel we've already started. Notion uses this by giving new users pre-filled templates instead of blank workspaces. Give users artificial progress and they'll finish what they think they've started.
Scaling AI-native companies requires rethinking job descriptions for AI agents. An Agent Job Description (AJD) should define the desired outcomes, necessary context, security, and budget for an AI agent. The development of AI agents ranges from simple chat platforms to complex custom codes, requiring careful trade-offs between capability, complexity, and context based on the task's specificity and scale.
Recent IPOs from CoreWeave and Circle signal an open IPO market, with both companies displaying significant investor interest and high valuations. CoreWeave achieved a $20 billion valuation, while Circle's shares jumped 168% on its first trading day. This trend indicates strong investor appetite for high-growth, category-defining companies despite economic uncertainties.
Sponsored
AI agents are gaining autonomy—and access. As they interact with internal tools, APIs, and systems, over-permissioned agents can leak data, perform unintended actions, or expose your platform to risk. Without scoped access, auditability, and strong defaults, they're a liability waiting to happen.
Conversational analytics for developers.
Bilanc is an AI-powered platform that tracks and measures the impact of AI tools on developer productivity using insights from merged pull requests.
Agent Simulate is a testing platform that allows developers to simulate AI agents with thousands of digital humans, enabling faster debugging and safer deployment.
VCs are starting to buy up traditional businesses and juice them with AI. It's the old private equity roll-up strategy - buy lots of small companies, merge them, cut costs - but with a Silicon Valley twist. Instead of using debt like PE firms do (which is expensive at 5% rates), VCs are using their massive cash piles to acquire service businesses and transform them.
There's "revenue" and then there's revenue - and they can be wildly different numbers. ARR is whatever companies want it to be, while GAAP revenue follows actual accounting rules. Example: a company showing 50% ARR growth might report 67% GAAP revenue growth - a 17-point gap. Why? GAAP lags because it recognizes money as delivered, not when signed. With AI scrambling pricing models, these gaps are getting dangerous. Less than 1/3 of public companies even disclose ARR anymore.
Bloomberg Beta has announced its fifth $75M fund to invest in early-stage startups. The firm maintains its consistent strategy of supporting founders with innovative ideas without seeking to reinvent VC. Under partners Roy Bahat, James Cham, and Karin Klein, Bloomberg Beta has helped create 93 millionaire founders, focusing on meaningful startup collaborations.
The most successful raises from pre-seed founders build leverage with "pre-traction" - waitlists, design partners, or domain expertise proving demand exists before revenue does.
Patrick Collison's darkest moment came six months in when a data center switch failed at 4 am and took down Stripe's entire payments API for 8 hours.
Sponsored
Attackers used stolen domain admin credentials to infiltrate a hospital network - aiming to deploy remote access tools, steal sensitive data, and unleash ransomware.
The Texas Department of Transportation is notifying the impacted individuals of a data breach of its Crash Reports Information System (CRIS). The breached data includes full names, addresses, driver's license numbers, license plate numbers, and car insurance policy numbers. An attacker obtained the data via a compromised account.
Sensata Technologies confirmed hackers stole sensitive employee data during a March-April ransomware attack. The stolen information includes Social Security numbers, financial details, and health insurance data.
Researchers discovered two Secure Boot bypass exploits affecting 50+ device makers. Microsoft patched CVE-2025-3052 (DT Research firmware tool vulnerability) but left CVE-2025-47827 (IGEL Linux OS flaw) unpatched. Both allow attackers with physical access to install pre-boot malware, defeating Secure Boot's core security purpose against "evil maid" attacks.
At Gartner's Security Summit, analysts advised harnessing AI and cybersecurity hype constructively rather than letting it drive hasty decisions. They recommended using protection level agreements and outcome-driven metrics for fact-based security conversations with executives. Security teams must develop AI literacy while protecting organizational AI investments from emerging threats.
Most process injection techniques typically begin by allocating memory, writing to that memory, and then executing it. However, the researchers in this post explored the possibility of process injection while only using the execution step. The researchers first achieved their goal by dropping a malicious DLL and then executing it by abusing that LoadLibraryA automatically appends “.dll” to strings that are passed to it. The researchers then took this one step further and used CreateRemoteThread with SetThreadContext and NTCreateThread to execute code in a remote process without needing to write a DLL to disk.
This post presents a list of security decisions that can be made early in a startup's lifecycle to enhance long-term security while providing the most value early on. The list includes organizational security and IAM-related decisions such as choosing an Identity Provider, adopting security keys, and using a Mobile Device Management (MDM) solution. The list also includes architectural decisions like adopting Infrastructure as Code and using managed infrastructure where possible.
The Containerization package allows applications to use Linux containers. Containerization is written in Swift and uses Virtualization.framework on Apple silicon.
Swimlane's platform leverages agentic AI and automation to resolve security, IT/OT operations, and compliance issues.
Claude-C2 provides an MCP server to communicate with a C2 server.
Marks & Spencer has partially restored online orders following an April cyberattack, with select fashion ranges now available in England, Scotland, and Wales. Services remain limited - no Northern Ireland delivery, Click & Collect, or next-day options.
Researchers from iVerify detected behavior indicative of tampering and spyware in iPhones belonging to high-value targets in an EU member state's government, US political campaigns, media organizations, and an AI company. The hackers targeted the individuals via the “Nickname” feature in iOS, which sends a notification whenever someone's iCloud photo or name changes. Apple has fixed the vulnerability being used and claimed that it has uncovered no evidence of exploitation.
ConnectWise is updating digital certificates for key software to address security concerns - users must install these updates by June 13 to avoid potential issues.
Starting next month, Microsoft will expand the types of blocked attachments in Outlook Web and the new Outlook for Windows to improve security by preventing potentially harmful files from being sent or received.
Sponsored
Delve ran 4 ads across 2 TLDR newsletters and brought in 66 leads, with a meaningful chunk from enterprise. The result: $1M in attributed pipeline and a 52x ROI.
Media consumption time has grown slowly over the past decade while the volume of available content has exploded, fragmenting consumer attention. As users toggle between platforms, devices, and formats, the traditional focus on total time spent or audience size fails to capture what truly drives monetization. Commercial factors like platform scale, pricing, and distribution explain some of the variance in revenue, but a significant portion depends on how engaged the audience is and why they are consuming the content. Formats that encourage focused and purposeful engagement, such as books, live events, and gaming, consistently outperform low-focus formats like social video or background audio.
Google's new AI tools are slashing traffic to news sites by replacing traditional search with chatbot answers. HuffPost and Business Insider have seen search traffic drop by over 50% in the past 3 years. The New York Times fell from 44% to 36.5%. In response, outlets such as The Atlantic and Washington Post are focusing on subscriptions, live events, and stronger reader relationships. Many were already battling low public trust and rising competition. Now they also face the challenge of protecting their content from being used to train AI without permission.
Marketing AI hackathons help teams quickly adopt AI through hands-on experimentation. The format involves setting a date, defining constraints, choosing a few tools, brainstorming ideas, building simple projects in a day, and sharing demos the next day. Tools like ChatGPT, Clay, Figma Make, Lovable, and Replit make it easy to build workflows, design assets, or internal utilities without heavy coding. During its internal hackathon, MKT1 built several practical tools in under an hour each, including a public hackathon idea generator using Lovable, a UTM link builder in Replit, and a screenshot formatter in Figma Make.
This post shows how to use Perplexity Labs to create an interactive quiz from any PDF, turn stats into a creator media kit, and build boardroom-ready slide decks. Prompts and examples are included.
Metrics evolve as platforms mature, from hits and registered users to more meaningful measures like daily active users and time spent. With AI, there's still little agreement on what counts as meaningful use, whether it's casual prompts, enterprise adoption, or token counts that vary by context. Companies like Google track detailed behavioral signals, but AI products don't have clear equivalents yet. Comparisons to smartphones or the early internet lack context since AI is more accessible and built on existing infrastructure. For now, measuring AI's true impact remains unclear.
Bain & Company's report spotlights 120 insurgent brands in the US consumer goods sector that drove 39% of the category's incremental growth. These high-growth brands include newcomers like Goodles and veterans like Chobani and Celsius. Defined by rapid growth and independence, insurgents are reshaping categories like food, beverages, and personal care by meeting unmet needs with founder-led authenticity and a disciplined playbook. With more incumbents seeking volume-led growth and improved capital conditions, M&A interest in insurgents is rising.
Jewellery is catching on with British men, especially younger ones. More men are buying jewellery for themselves and doing it on their own terms. In the past year, 17% bought it for themselves. Among Gen Z, that rose to 24%. Necklaces or chains were most popular at 33%, followed by watches at 32%. Nearly half of buyers, 47%, shopped mostly or entirely online. Price and discounts influenced 39%, far more than online ads or celebrity endorsements. Independent jewellers led at 39%, ahead of marketplaces at 30% and high street stores at 27%.
Zillow's early growth relied not on traditional marketing but on building features that stirred curiosity and emotional reactions, such as its controversial home value estimator.
88% of subscription brands expect direct acquisition costs to rise in 2025, prompting a major shift in budget toward bundling, partnerships, and aggregator platforms that deliver higher quality subscribers with lower upfront spend.
GitHub engineers tackle platform problems by focusing on understanding their domain, bridging concepts with platform-specific skills, and focusing on knowledge sharing. Due to the wide impact radius of platform changes, they make sure to test deeply with Infrastructure as Code, end-to-end testing, and self-healing capabilities.
Conviva's streaming analytics platform experienced a P99 latency spike for a single customer due to a bottleneck in a shared type registry used by its DAG engine. The issue was due to a high contention on an atomic counter within a concurrent hash map used for read operations. The solution was to switch from flashmap to ArcSwap, which uses a read-copy-update approach, eliminating the counter contention issues by atomically swapping the entire data structure.
Traditional observability tools are becoming obsolete due to the rise of LLMs, which can analyze telemetry data as effectively as human experts. This is shown by an AI agent successfully diagnosing latency spikes in a demo environment. This means that in the future, AI will assist in various development and operational tasks, from suggesting code improvements to autonomously addressing system behavior issues.
Engineering Management has become increasingly difficult due to layoffs, shrinking teams, and budget freezes. Companies now expect EMs to handle people management, technical vision, and some coding, often for multiple teams. At the same time, EMs don't have any control over resources like budget and headcount.