Apostrophe

Your content team wants visual editing. Your dev team wants a modern, flexible architecture. Why not both? ApostropheCMS now works seamlessly with Astro, giving you: ? A powerful headless CMS backend ? Live, in-context editing for rich page building and structured content ? Astro’s lightning-fast frontend performance and amazing DX We just launched a new tutorial series to show you how to integrate the two for a best-of-both-worlds web stack. ?? Get started here: https://hubs.la/Q039VyX-0 Are you using Astro for client projects? Let’s talk! ??

ChatGPT 4.5 has dropped and yes, it's better. But the costs show we've hit a real plateau with general-purpose LLMs. OpenAI says ChatGPT 4.5 "excels at tasks requiring creativity, empathy, and broad general knowledge, including writing help, coaching, brainstorming, and nuanced communication." But, this comes at an enormous increase in cost. The API will be 30 times as expensive as ChatGPT 4o. If they even continue to offer that API. For chat window users, it will be a "research preview" for now in the $200/month tier. If it graduates to the $20/month tier, I expect to see strict limits on usage per day, similar to how they handle the o1 "reasoning" model. And looking at the attached chart from the ChatGPT blog, I think the real message is that users prefer it just a little more often than ChatGPT 4o — at vastly greater expense. OpenAI has hit the long-predicted plateau. I don't expect their foundation models to get better until there's a breakthrough in the software, or a breakthrough in hardware costs. What about the competition? The biggest news with deepseek was that they trained their model for less. They could do that again by leveraging ChatGPT 4.5 in the training process. But the problem with ChatGPT 4.5 isn't just the cost of training, it's the high cost per query after it's trained. So how does vanilla, non-R1 deepseek stand up in terms of the cost per query? They charge about 10x less for the regular deepseek chat API, compared to ChatGPT 4o's chat API. That suggests progress in operating costs as well. But, it also has about one-third as many parameters as ChatGPT 4o. And I haven't found many sources that suggest it is actually better at "tasks requiring creativity, empathy, and broad general knowledge." That would help explain the reduced cost without any actual game-changing improvement in the basic technology. My gut feeling is that deepseek itself is just a "pretty good" general-purpose foundation model and their real innovations lie in the R1 reasoning model, which provides an o1-like improvement in reasoning ability for less. But, isn't suitable for every application. And, it is largely matched by ChatGPT's more recent o3-mini. So all in all, I think we've hit a broader plateau around the industry when it comes to foundation models. But, we're seeing rapid advancement in "reasoning" models and other niche solutions that build on that foundation. In "The Bitter Lesson," Rich Sutton wrote about a recurring theme: trying to teach machines to "think the way we think we think" never works as well as simply training them on input, as we do with foundation models. But, the success of reasoning models shows that we can have better luck applying those high-level ideas if we use foundation models as a building block. "A poet, a psychologist and a programmer walk into a bar..." it's a setup for a joke, but the punchline could be a successful, affordable model for more empathetic reasoning.

Did you know you can watch a walkthrough video of our releases? We are pumped about this one and would love to hear your thoughts on how it's packaged. Do you prefer to watch a video, read through a list in an email or read on discord/other platforms? Watch it here if you're a video person: https://hubs.la/Q034kvlv0

Hello Apostrophe Community! This month we're excited to bring you ApostropheCMS 4.13.0 which includes enhanced localization, improved imports, and better document management! Peek the teaser https://lnkd.in/eX7paj_U and read all the details on the blog https://lnkd.in/eiwHkPFK

Metadata and translations can can suck up your time. Our SEO Assistant and Automatic Translation tools help you optimize SEO and localize content faster using AI, so you can focus on what matters—creating smooth experiences for your clients and their clients. Our customers have been loving these features - try them out! Read more on our blog: https://hubs.la/Q034j_Rm0 #SEO #Metadata #AIinWebDev #ApostropheCMS

?? Managing hundreds of career sites for global brands requires serious scalability and flexibility. iCIMS powers its Career Sites platform with Apostrophe Assembly, enabling teams to efficiently manage content across employer sites—all while delivering a tailored, brand-consistent experience and deep integration with their other products. See how iCIMS leverages Assembly’s multitenant capabilities to scale and streamline their platform.

?? Choosing the right CMS for your agency clients isn’t just about features—it’s about delivering scalable, flexible, and maintainable solutions. Apostrophe Assembly is built for agencies that need to manage multiple sites, balance customization with efficiency, and provide a seamless experience for their clients. See why it stands out as a top multisite CMS for agencies.

Does HTMX live up to the hype? We think so. Its goal is to keep interactivity simple and efficient, so it cuts out much of the complexity of traditional web development frameworks. We have a blog article (https://hubs.la/Q033n8GY0) to help you dive in : ?? What HTMX is and why it’s worth checking out ?? How it’s different from JavaScript frameworks ?? The best use cases for it ?? Instructions for setting it up with ApostropheCMS #WebDevelopment #HTMX #opensource #ApostropheCMS

Good tools make work feel like magic. Apostrophe is rolling out updates and new tools designed to give editors a firmer grip on templated content and site design. Read about what's coming to Palette and two new features: Sections and Widget Tools today on our blog https://lnkd.in/e3exQ2DQ

Palette, our in-context style editor, has been getting a steady stream of UI/UX improvements. Get reacquainted with the latest in Palette by checking out this short overview of what we've been up to https://lnkd.in/ernrT4CZ