Lovable found that forcing a subscription-only model hurt retention because user demand was irregular and many users rejected ongoing commitments. It introduced credit top-ups instead and A/B tested pricing, and a 20% premium over subscription credit prices performed best while a 40% premium reduced adoption. The 20% markup increased engagement and bookings, and paid plan retention rose by 7%, even though upgrades to higher plans dropped by nearly half. Within one week, 20%+ of bookings came from top-ups, and subscription revenue still grew. Flexible pay-as-you-go options matched real usage patterns better and drove stronger long-term value than subscription-only billing.
Learn how to use AI to define what actually matters, so you can trust your metrics to guide real product decisions. The hosts will cover how to distinguish vanity metrics from decision metrics and validate metric quality. The free event will take place on February 18 at 10:30 AM PT.
Lockdown Mode is an optional security setting in ChatGPT for users at higher risk of targeted attacks. It mitigates prompt injection and data exfiltration by restricting external interactions, limiting live web access, and disabling tools that cannot meet strict data safety guarantees. This allows admins to control approved apps and actions. It is available on enterprise-focused plans, with broader rollout planned. Elevated Risk labels mark features that introduce added network or data exposure risk, giving users clearer visibility and choice when enabling them.
Bing has launched a new AI Performance report in its Webmaster Tools. It shows how often a site's pages are cited in AI-generated answers across Microsoft Copilot, AI summaries, and partner tools. It reports total citations, unique cited pages per day, average cited pages, grounding query phrases, and page-level citation counts. It updates daily using aggregated sample data. Citation counts show how often content appears as a visible source, and they do not measure rank, traffic, clicks, or authority.
Pinterest captured massive traffic from Google Images by turning UGC into an SEO engine. Search crawlers read text rather than images. Pinterest structured board titles, pin descriptions, and comments into indexable pages, then built templated pin, board, and programmatic explore pages designed to match high volume queries. Each user action created a new searchable entry point, forming a self-reinforcing growth loop that drove hundreds of millions of organic visits. A login wall gated full content behind registration, converting search traffic at scale until Google's 2018 algorithm update penalized the tactic and erased millions of rankings.
AI citation behavior favors front-loaded and entity-rich writing over long narrative structure. 44.2% of citations come from the first 30% of a page, while only 24.7% come from the last third. Key facts buried deep are 2.5X less likely to be cited. Citation-heavy passages use definitive statements, include questions, and show high-entity density (using specific names like Salesforce or Pipedrive). Paragraph analysis shows 53% of cited sentences sit in the middle of a paragraph, so strong facts can appear after the first sentence if the paragraph appears early on the page. Citation winners also use a balanced tone and a readability level near grade 16, similar to outlets like Harvard Business Review.
Recent baby formula recalls involving Nestlé and Danone renewed scrutiny of crisis communication in consumer safety categories. Once risk is credible, containment should begin immediately through rapid recalls and transparent updates rather than fixed timelines. Nestlé leaned on clear language and visible executive leadership to signal accountability. Danone took a more formal tone and moved more slowly, which limited reassurance for consumers. In high-risk sectors, brands need prebuilt escalation plans and real-time sentiment tracking.
Set a hierarchy in ChatGPT that lets current conversation instructions override project, account, and memory settings, addressing only direct contradictions.
AI makes prototypes easy to create, but not all are built to last. Choose tools based on whether the work should validate ideas quickly or evolve into real products.
Teams shift between intentional strategy, undisciplined drift, and reactive survival. Real progress requires noticing the drift and reclaiming deliberate focus.
Startups win when they find buyers who urgently want change because they are coping or blocked. Product-market fit exists when real demand comes first.
AI-first companies now expect PMs to run evals, prototype with code, understand model tradeoffs, and automate with agents, meaning the PMs who thrive will be technical builders, not just coordinators.
Plaid achieved their best-performing direct buy to date, adding $382k in pipeline, earning 70 leads, and achieving 20.1x ROI. This case study breaks down their strategy, results, and example ads. Read the
OpenClaw creator Peter Steinberger is joining OpenAI to work on bringing agents to everyone. OpenClaw will move to a foundation and stay open and independent. While Steinberger sees the potential in OpenClaw becoming a huge company, he doesn't find the prospect appealing and would rather change the world with OpenAI.
CapCut users will soon have access to Seedance 2.0, ByteDance's latest AI video generation model. The model is already available to users on the Chinese version of CapCut, Jianying. It can generate coherent story lines, create realistic voice-overs and background sounds, and generate complex character actions. Seedance 2.0 can currently only generate videos of up to 15 seconds. It generated some controversy on launch, with critics raising privacy and copyright concerns.
One of the key steps in the origin of life is the emergence of an RNA molecule that could replicate itself. RNA can carry genetic information and also fold into three-dimensional structures that act as catalysts. This suggests that early life was protein-free, with RNA handling heredity and catalyzing a simple metabolism. A team of scientists has discovered a 45-base-long piece of RNA that can make copies of itself, adding to our knowledge of the conditions in which life emerged.
Fusion energy startup Helion has hit a key milestone: plasma inside its Polaris prototype reactor has reached 150 million degrees Celsius. The company's ultimate goal is to produce plasmas that hit 200 degrees Celsius. Its reactor design is different from its competitors, requiring plasmas about twice as hot to function as intended. Helion plans to extract energy using the fusion reaction's magnetic field rather than from heat. The company is contracted to sell electricity to Microsoft starting in 2028, but this will be through a larger commercial reactor called Orion, which is currently being built.
Cognitive debt is the idea that AI-generated codebases are at risk of falling into a state where nobody knows how they work, leaving them inextensible, unobservable, and hard to debug. This is already happening in AI-driven codebases. However, when controlling for the size and scope of a project, cognitive debt tends to be at least as bad, if not worse, in pre-AI code bases than in AI code bases. This fact is obscured because a lot of traditional engineering work is either managing crippling cognitive debt or avoiding it at enormous cost. The best AI users are already pretty good at avoiding cognitive debt, and they're only going to get better at it.
Bias toward action is defaulting to the smallest responsible step that produces real feedback while pre-committing to guardrails. It helps companies learn faster while making sure that being wrong doesn't kill them. It's about optimizing for learning speed while investing in foundations that make you faster. Fast teams use a lot of information, but they don't wait for perfect information. Their tight feedback loops allow them to learn quickly.
Meta plans to add facial recognition to its Ray-Ban smart glasses later this year. Its 'Name Tag' feature will let wearers identify people and get information about them via the company's AI assistant. The company considered adding facial recognition to the first version of the glasses in 2021, but pulled back over technical challenges and ethical concerns. Meta plans to first release Name Tag to attendees at a conference for the blind before making it available to the general public.
Companies that don't rebuild how they work from first principles will get outshipped by teams half their size six months from now. Engineering teams need to throw out the playbook and write a new one for the age of AI. This post discusses one startup's new playbook, which puts agents first as the primary users who do the work. The engineer's role in the framework is to make sure the agents can do the work well.
The Legacy Survey of Space and Time (LSST), a project by the Vera C. Rubin Observatory in Chile, will help fill out the gaps in our knowledge of the contents of the solar system and greatly advance our census and our knowledge of the contents of the solar system.
Some companies have a lot of power over the open source projects they contribute to, and this can become a problem if they make decisions based on their own agendas.
OpenClaw (formerly ClawdBot and MoltBot) is an open-source agent assistant that operates with the same permissions as the user who installs it - and can act on their behalf.
>> WANT TO GET CERTIFIED IN AI SECURITY? With the _Foundations of AI Security_ [5] series, you can learn how to deploy secure AI across the enterprise directly from experts at Zenity and AWS. Attend all three virtual sessions to earn earn the Foundations of AI Security Professional Certificate!
OpenClaw's creator will join OpenAI to develop accessible agents and ensure OpenClaw remains open and independent within a new foundation. OpenAI's support will provide access to cutting-edge models and resources, aligning with the creator's vision of making these tools widely usable. OpenClaw's community will benefit from a supportive structure that encourages innovation and data ownership.
OpenAI introduced an optional Lockdown Mode for higher-risk workflows and added “Elevated Risk” labels for capabilities in ChatGPT, ChatGPT Atlas, and Codex that can increase exposure to prompt-injection attacks.
xAI's Grok Build plans to introduce Parallel Agents and Arena Mode, enhancing its capabilities from a coding assistant to a full IDE. Parallel Agents allows up to eight coding agents to work simultaneously, while Arena Mode facilitates competition among agents to determine the best output. The update also includes a new UI resembling a browser-based IDE, dictation support, and nonfunctional GitHub integration, with Grok 4.20 expected soon.
Both CEOs of OpenAI and Anthropic claim that human-level AI is just around the corner. However, Transformers have their fundamental limitations. While they are very powerful and have taught us a lot about what general intelligence is, overcoming these problems will require research that could take decades. The advances that made modern GPTs possible were discovered gradually over several decades, and each discovery came with new and more nuanced issues.
OpenClaw's recent security issues highlight a larger problem with agent ecosystems as new attack surfaces, where malicious skills can exploit agent + tools + marketplace combinations. Current safeguards like prompts are inadequate. Agents need sandboxing, scoped credentials, restricted tools, and proper logging to manage risks.
Anthropic and OpenAI have unveiled new "fast modes" for LLM inference, offering differing performance enhancements. OpenAI's model speeds exceed 1,000 tokens per second using Cerebras chips, but rely on a less capable model, while Anthropic's fast mode supports real models with up to 2.5x speed through low-batch-size inference. Despite OpenAI's technical achievement, the utility of faster but less capable inference remains debatable, focusing more on OpenAI's exploration of Cerebras' potential.
Text has to pass through a tokenizer before GPT-5.x is able to understand any input. A tokenizer is a compression layer that converts raw text into a sequence of integer IDs. Every design decision baked into the tokenizer has downstream consequences for cost, accuracy, multilingual performance, and hallucination rates. This article looks at OpenAI's open-source tokenizer library, tiktoken, and explains how it works.
Human judgment is the backbone of AI quality. Yet most orgs scaling AI lack clear decision frameworks, consistent interpretation policies, and reliable oversight mechanisms. Welo Data [14] builds AI data quality systems that govern human and machine judgment. Turns out that