Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Duolingo improved next-day retention from 47% to 55% by using challenges and other gamification features. The app packages accomplishments as a “challenge”, which is either a high-level goal (e.g. "Learn Vietnamese") or a smaller goal (e.g. "study Vietnamese 7 days in a row). Duolingo provides visual motivations to show daily progress and uses external reminders to drive users back to the app.
This is a case study on how Ryan Serhant used YouTube to scale his brokerage firm. It shares tactics on how to win clicks via eye-catching thumbnails and maximize retention within the first 30 seconds of a video and highlights the importance of having an intentional content mix.
TikTok’s new AI-powered “Creative Assistant” tool exists within its ad tools and guides marketers through the campaign creation process. It can show you top-performing ads in your niche and produce ad scripts. These options have been available in TikTok’s Creative Center for some time, but now they’re built into the app to streamline the process.
According to Litmus’ 2023 State of Email Design Report, 43% of marketers said there was an improvement in email performance when GIFs were used in their marketing emails. This guide provides GIF examples from brands, a list of considerations to think about when inserting GIFs into emails, and the steps in creating your own animated GIFs via Adobe Animate.
Merge, the G2 Leader for Unified APIs, started out with two founders encountering a huge problem. They decided to talk to 100 different companies to understand how prevalent the problem was before they reached any sort of solution. In doing this, they understood customer pain points intricately, which ultimately informed their messaging strategy and equipped them to grow through founder-led sales. This article contains more details about their journey to date.
Marketing can only generate an effective pipeline if it targets the same buyers as sales. However, a recent analysis from LinkedIn that measured the targeting overlap between sales and marketing across 7,000 B2B companies found it to be only at 16%. In theory, this number should be 100%. In addition, the average overlap between brand and demand marketing is only 6%.
You don’t need to achieve statistical significance for qualitative research. Instead, you merely need 12-13 respondents to achieve ‘saturation’ in your study.
404 errors are not a direct ranking signal, but they do have a negative impact if valuable backlinks are pointing to them.
This post provides a quick summary of Forbes’ 2023 Top 50 Creators list—46% are women, 28% are people of color, and MrBeast is the reigning top creator.
Sponsored
“Data is the new oil” was a common slogan of the last decade, but this might not hold true anymore with the advent of generative AI. On one hand, quality data still matters as LLMs can be greatly influenced by training data quality. On the other hand, fine tuning models only requires ~100 data points, which can be easily curated. LLMs can also now generate great synthetic data, which lowers the usefulness of “real” data.
Super lightweight JSX syntax highlighter at 1KB after being minified and gzipped.
html2pdf.js converts any webpage or element into a printable PDF entirely client-side using html2canvas and jsPDF.
GDevelop is a full-featured, no-code, open-source game development software. You can build games for mobile, desktop, and the web. GDevelop is fast and easy to use. The game logic is built up using an intuitive and powerful event-based system.
This is a dictionary of algorithms, algorithmic techniques, data structures, archetypal problems, and related definitions.
An online book covering the internals of Deno, a fast run-time for JavaScript.
Sponsored
Sponsored
BinDiff is a comparison tool for binary files. It is able to quickly find differences and similarities in disassembled code. BinDiff helps researchers identify and isolate fixes for vulnerabilities in vendor-supplied patches and can be used to port symbols and comments between disassemblies of multiple versions of the same binary, making tracking changes over time easier and allowing organizations to retain analysis results.
Sponsored
Level up in 5 minutes a day with interactive challenges in math, logic, CS, and more. Sharpen your mind — master new skills — unlock career possibilities. Try Brilliant free for 30 days.
Workers AI is an AI Inference-as-a-Service platform powered by a global network of GPUs that enables developers to run AI models with just a few lines of code.
Sponsored
The semantic layer is key to delivering the ‘last mile’ of analytics — making data available in tools such as Power BI, Tableau, and Excel. But implementing it in the data-intensive insurance industry can get tricky. Watch this tech talk
After a huge seed round and building a team, ML stack, and data pipeline, the Mistral team has released an extremely performant 7B model. It was initially released via a magnet link (torrent) along with a minimal GitHub Repo and permissive license.
Meta is launching its own AI chatbot assistant along with various AI characters across its platforms. The chatbot is designed as a general-purpose assistant, uses Microsoft's Bing for real-time web results, and can generate images similar to OpenAI's DALL-E.
OpenAI has reintroduced a feature in ChatGPT that enables it to access current web information using the "Browse with Bing" functionality. It is initially available for Plus and Enterprise subscribers.
Llava is a method that inserts vision capabilities into language models. It has now been instruction-tuned and run through an RLHF process. Llava improves instruction following and reasoning abilities.
Large models like ChatGPT have been used to improve search results, but many of these tests are hard to repeat because they're hidden or secret. Researchers have introduced RankVicuna, an open-source tool that can reorder search results using language models.
Large language models are powerful, but often too big for some devices due to their computational demands. This new method, QA-LoRA, makes models more efficient by cleverly changing how they're built and fine-tuned.
Sponsored
Want to know what users are doing inside your app? Why not just ask? TLDR members get EXCLUSIVE 3 months free*