- Dev Notes
- Posts
- OpenAI Launches Next-Gen AI and Model Updates
OpenAI Launches Next-Gen AI and Model Updates
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/3f382e8d-dc40-4f24-8cd7-41179fb0d4fe/newsletter_header.png?t=1703072155)
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e2e91d89-f933-4752-817f-adfd037e4a33/header.png?t=1706228032)
Good Morning! OpenAI launched major updates including new semantic text embeddings, reduced GPT-3.5 Turbo pricing, and granular management tools for APIs. Google rolled out iOS and Android mobile previews in Project IDX, its AI-powered web dev platform, enabling cross-platform testing. To comply with the EU's Digital Markets Act, Apple opened up the iPhone app ecosystem in Europe to allow third-party stores and reduced commissions.
OpenAI Launches Next-Gen Embeddings and Model Updates
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/6b627cba-08f8-4312-b421-9770cabcd265/chatgpt.png?t=1706226658)
OpenAI announced several major updates last week, including new embedding models, reduced pricing on GPT-3.5 Turbo, and management tools for APIs.
Embeddings are vector representations of text that capture semantic meaning. OpenAI released two new embedding models - text-embedding-3-small and text-embedding-3-large - which outperform previous models on benchmarks. For example, text-embedding-3-large improved multi-language retrieval scores by over 23 percentage points. The smaller model is also 5 times cheaper.
OpenAI explained that the models support "shortening" embeddings to reduce costs, giving developers flexibility to balance performance and affordability. The company also updated GPT-3.5 Turbo with 50% lower input pricing, 25% lower output pricing, and accuracy improvements.
The GPT-4 Turbo preview saw upgrades as well, with better task completion and bug fixes. OpenAI's moderation API, which flags harmful text, now boasts the "most robust" model yet in text-moderation-007.
Finally, the new API management tools provide finer-grained control over keys. Developers can assign restricted permissions to keys and enable tracking to monitor usage metrics for features, teams, and projects separately.
Read More Here
Google Rolls Out iOS and Android Previews in Project IDX Web Development Platform
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/6178db82-0bb8-4649-8645-bdc9e4f24940/porjectidx.png?t=1706227039)
Google announced several major updates to Project IDX, its AI-powered web development platform, including the addition of iOS and Android mobile previews. Now, developers can access iOS and Android simulators directly in the browser instead of requiring a Mac or mobile device.
This allows developers building cross-platform apps with frameworks like Flutter to easily see how their changes impact the iOS, Android, and web versions. The iOS simulator is still experimental but allows spot-checking layout and behavior. The Android emulator enables full testing and debugging.
In addition to the mobile previews, Google has expanded Project IDX's support for more languages and frameworks by adding project templates for Astro, Go, Python, Node.js and others based on user requests. There are also package manager upgrades through Nix that make it easier to customize starter templates and recover from issues.
Additional improvements include the ability to auto-configure firewall settings, run command-line tools within Project IDX instead of locally, and simplify Docker configuration. The updates streamline multi-platform development so developers can build and iterate faster.
Project IDX entered early access last year as an ambitious cloud-based development environment aiming to provide a full coding workspace and preview testing in a single web-based IDE. It uses AI for features like autocomplete and comes pre-loaded with templates for many popular web frameworks.
Read More Here
Apple Opens Up iPhone App Ecosystem in EU to Comply with Digital Markets Act
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/8163c01e-161f-4d10-bfbb-be6b937bf376/appstore.png?t=1706227339)
As part of complying with the EU's Digital Markets Act (DMA), Apple has announced significant changes coming to the iPhone app ecosystem in Europe. The most notable change is that third-party app stores will be allowed on iPhones for the first time, ending the App Store's monopoly as the sole distributor of iOS apps.
These "alternative app marketplaces" will go through Apple's approval process and once installed, users must explicitly grant them permission to download apps. Apps from third-party stores won't be subject to App Store guidelines and could include things like emulators and clipboard managers which are currently prohibited.
On the developer side, the 30% "Apple tax" will be reduced to 17% for apps using alternate payment systems, and further reduced to 10% for small developers. A new annual $0.54 per user fee kicks in after a million installs. So most devs will pay less while the most popular apps generate new revenue for Apple.
Other technical changes include allowing alternate browsers like Chrome and Firefox to be set as defaults, opening up NFC access for payments in third-party apps, and allowing game streaming services globally.
These are significant concessions from Apple's walled garden approach, albeit just in the EU for now. It shows the impact tighter regulations can have on Big Tech's business models. For iOS users it means more choice, but Apple warns it also means more potential for malware and scams outside the App Store. Technical users should understand these tradeoffs as they explore the newly opened iOS ecosystem.
Read More Here
SnapLogic Introduces No-Code Tool for Building AI Apps
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/9635ae48-94d1-4be7-89e5-1e4ce4a6dc0a/snaplogic.png?t=1706226044)
SnapLogic has rolled out a new no-code platform called GenAI Builder that lets enterprises create custom AI applications without coding. Part of SnapLogic's AI suite for integrating AI capabilities into data workflows, GenAI Builder utilizes large language models like GPT-3 to draw insights from all kinds of structured and unstructured data.
GenAI Builder gives business users and citizen developers an intuitive visual interface for building conversational AI apps. It comes with templates for common needs like customer service chatbots, market research analysis, and document review automation. Users can also connect GenAI Builder to existing databases, APIs, and even legacy systems to build powerful retrieval-augmented generation tools.
As per SnapLogic, GenAI Builder meets the fast-growing demand among enterprises for swiftly deploying new AI-powered services. It helps different business teams to seamlessly add intelligent features into both new and existing apps without involving data scientists. This no-code approach boosts time-to-market while overcoming resource limitations.
Early adopter Lumeris intends to leverage GenAI Builder for expediting development of generative AI solutions to improve healthcare outcomes. "By streamlining our integration needs, SnapLogic is assisting us to rapidly construct generative AI systems that will meaningfully help our members and partners," said Lumeris Director Gaurav Bajpai.
Read More Here
Youtube Spotlight
How Apple accidentally broke my Spotify client
Software developer Roberto Frenna was frustrated when his Spotify client abruptly stopped playing music after he returned from some time off. He soon realized the issue was only occurring on his work laptop, not his phone. After lots of debugging and log analysis, he found that the root cause was a recently introduced bug in the macOS mDNSResponder daemon, which handles DNS requests. The bug was in logic intended to fix incorrect search domain usage when following CNAME records. It ended up sometimes telling APIs like SCNetworkReachability that hostnames had "no addresses" when search domains were enabled but not actually used. This caused Spotify's connectivity check to fail. Roberto dug deeply into the mDNSResponder code, staring at incorrect versions for a while before realizing the latest was on a separate GitHub branch. Once finding the true culprit commit diff, he proposed a simple fix to only reapply search domains if used before. After much frustration from Apple's logging redactions and failed debugging attempts, he finally solved his Spotify playback problem.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/92f7d28f-33c2-404e-9ce8-4174bcc6f439/credits.png)
Was this forwarded to you? Sign Up Here
Reply