NVIDIA Unveils Its Most Affordable Generative AI Supercomputer

NVIDIA has unveiled the Jetson Orin Nano Super Developer Kit, a compact generative AI supercomputer designed for hobbyists, developers, and students. Priced at $249—half the cost of its predecessor—the kit offers a significant 1.7x gain in generative AI inference performance, boasting 67 INT8 TOPS and a 50% increase in memory bandwidth to 102GB/s. Equipped with an NVIDIA Ampere GPU and a 6-core Arm CPU, it supports up to four cameras and multiple AI pipelines, making it ideal for prototyping edge AI applications and projects in generative AI, robotics, and computer vision.

The Jetson Orin Nano Super caters to the rapidly growing field of generative AI, offering compatibility with transformer-based models and foundational AI frameworks. NVIDIA’s AI software ecosystem, including Isaac for robotics and Metropolis for vision AI, streamlines development with tools like the TAO Toolkit and Omniverse Replicator. Developers benefit from extensive support through the Jetson AI lab, open-source tutorials, and a thriving community. Jetson ecosystem partners also provide additional tools, sensors, and carrier board design services, enabling a versatile platform for innovation.

In addition to the new hardware, existing Jetson Orin Nano users can access a performance boost through a software update to the JetPack SDK. The update, dubbed "Super Mode," delivers up to 1.7x enhanced generative AI capabilities across the Orin Nano and Orin NX product lines. This ensures the latest advancements in performance are accessible to both new and current Jetson users, reaffirming NVIDIA’s commitment to democratizing AI technology.

Finally, a developer tool that makes digging around for the answer a thing of the past.

Other developer tools can’t tell you how your codebase works and why. Unblocked can. We augment your source code with context from Slack, Confluence, Jira (and more), so your team gets helpful answers instantly.

Unblocked gives you:

• Instant answers about every aspect of your codebase 

• Relevant, historical context for any file open in an IDE

• A powerful Slack integration that answers questions in channels, removing interruptions to you

⌛Engineers who use Unblocked save an hour or more a day.

“Unblocked empowers every developer with the insight once reserved for the seasoned few. ” Mathew Spolin - SVP Engineering at AppDirect

ChatGPT’s AI search engine is rolling out to everyone

OpenAI is expanding the reach of its AI search engine within ChatGPT to all users starting today. Initially launched for paid subscribers in October, the search engine is now accessible to users at the free tier, though an account and login are required. This announcement came during OpenAI’s “12 Days of Ship-mas” livestream, where the company also revealed improvements to the feature on mobile devices. Among the updates, users can now search with an optimized interface on mobile and access an advanced voice mode for enhanced interaction, the latter remaining exclusive to paid users.

The mobile updates make ChatGPT's search engine function more like traditional search tools. When searching for local spots, such as restaurants or attractions, the app now presents a detailed list complete with images, ratings, and operational hours. Selecting a location reveals additional information, and users can access a map with directions directly within the app. These refinements aim to provide a more seamless and visually engaging search experience.

Additionally, ChatGPT is introducing faster ways to find certain types of websites, like hotel booking platforms. Instead of generating lengthy responses upfront, the AI quickly displays links to relevant sites before delving into further details. Advanced Voice Mode also integrates real-time web searches, enhancing its utility for obtaining current information, though this feature is limited to paid plans. In earlier livestreams, OpenAI unveiled other developments, including the Sora text-to-image model and a $200 per month ChatGPT Pro subscription plan.

OpenAI brings its o1 reasoning model to its API — for certain developers

OpenAI has introduced its advanced reasoning AI model, o1, to its API, but initially limited access to developers in its “tier 5” usage category. To qualify, developers must spend at least $1,000 with OpenAI and have an account active for over 30 days since their first payment. O1 replaces the o1-preview model and offers enhanced features such as function calling, developer messages, and an image analysis capability. The model includes a new parameter, “reasoning_effort,” that adjusts the time it spends formulating responses. OpenAI claims this iteration, named “o1-2024-12-17,” improves its performance in programming and business-related queries and reduces instances of refusing valid requests.

In parallel, OpenAI unveiled updates to its GPT-4o models within the Realtime API, a tool for building low-latency voice applications. The new versions, “gpt-4o-realtime-preview-2024-12-17” and “gpt-4o-mini-realtime-preview-2024-12-17,” boast improved efficiency, reliability, and affordability. The Realtime API has also gained features like concurrent out-of-band responses for background tasks and support for WebRTC, enabling smooth real-time voice interactions even under fluctuating network conditions. This aligns with OpenAI’s recent hire of Justin Uberti, the creator of WebRTC, to spearhead these advancements.

Additionally, OpenAI expanded its fine-tuning capabilities by introducing preference fine-tuning, allowing developers to train models to prefer specific types of responses. The company also launched an early access beta for software developer kits in Go and Java, further enhancing its offerings for developers. OpenAI continues to roll out access to o1 incrementally, working to expand availability and refine rate limits for broader adoption.

Other stuff

• OpenAI co-founder Ilya Sutskever believes superintelligent AI will be ‘unpredictable’

• Meta asks the California attorney general to block OpenAI’s conversion to for-profit

• Radio LLM - Off-grid, disaster-proof LLM platform using Meshtastic

• OpenAI blames its massive ChatGPT outage on a ‘new telemetry service’

• David Sacks, from ‘PayPal mafia’ to Trump’s AI and crypto tsar

• Suit charges SF’s Scale AI with misclassifying workers, wage theft

• OpenAI fires back against Musk, claims he wanted an OpenAI for-profit

• Sam Altman reckons with a growing threat to OpenAI: Elon Musk

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.