Augmented Shelf | Issue 4 | April 3, 2024
Welcome to Augmented Shelf, a wrap-up of the week’s AI news, trends and research that are forging the future of work.
Street Fighting LLMs
Ready for a digital brawl like no other? In a groundbreaking experiment, cutting-edge Language Models (LLMs) are being tested in a novel way: real-time combat in Street Fighter III. The latest study pits LLMs like GPT-3.5 Turbo against Mistral in a digital showdown, assessing their speed, strategy, and adaptability far beyond traditional metrics. With over 342 matches fought in the LLM Colosseum, these AI gladiators are ranked using a sophisticated ELO system, providing a unique glimpse into the potential of AI in dynamic and high-pressure scenarios.
Databricks Unveils Its Open LLM, DBRX
Databricks has unveiled DBRX, a state-of-the-art open LLM, setting a new standard for AI models. DBRX eclipses existing models like GPT-3.5 and poses competition to Gemini Pro and GPT-4 Turbo, showcasing versatility and prowess in coding and beyond. With a fine-tuned MoE architecture, DBRX promises up to twice the inference speed of its peers, delivering top-notch results with remarkable efficiency. Embracing the open AI ethos, Databricks releases DBRX’s base and finetuned models under an open license on Hugging Face. Databricks customers gain privileged access to DBRX via APIs and can customize models using the same sophisticated tools behind its creation. DBRX is being woven into Databricks’ generative-AI powered products, and excels in tasks like SQL queries.
Stargate 2028
Microsoft and OpenAI are reportedly joining forces on a staggering $100 billion project to construct ‘Stargate,’ a supercomputer aimed at revolutionizing AI capabilities. With a timeline aiming for a 2028 launch, this ambitious venture is part of a five-phase collaborative strategy to push the boundaries of computing power. If realized, Stargate has the potential to become an unparalleled force in AI infrastructure, challenging the industry’s current operational scale and power consumption paradigms.
BBC Eyes the AI Arena
The BBC is stepping into the AI arena, eyeing the creation of their own proprietary AI models. The British news org is in talks with tech giants to unlock their vast archives for AI training, tapping into a treasure trove of source audio, video, and text data. The goal? To boost journalistic efficiency and unlock new commercial prospects and enable journalists to use models trained on this data to produce new content and stories. The BBC’s announcement comes in the wake of other highly visible AI content partnerships, notably Reddit’s recent licensing deal with Google, and Axel Springer and Le Monde’s deals with OpenAI.
AI’s Spending Dilemma
The AI industry is caught in a spending cyclone, investing in NVIDIA GPUs at a rate 17 times higher than their revenues. With titans like Microsoft, Google, and Meta sidelined by antitrust barriers, they’re pouring money into AI development instead. While this may offer tax benefits through GPU expense depreciation, the sustainability of such spending is questionable. Amidst this, Cognition Labs’ valuation soars on the back of its AI code tool, Devin, signaling a shift towards automation in software creation. As AI continues to consume capital, companies are racing to develop expansive models, resulting in a fierce debate on the industry’s economic trajectory and NVIDIA’s probable win in the generative AI hardware race.
Augment Yourself 🤖
- Discover why recurrent neural networks (RNNs) dominate sequential data analysis.
- Here are 5 attention mechanism insights every AI developer should know.
- Seamlessly link structured and unstructured data with a knowledge graph.
- Is your company’s data ready for Generative AI?
- How can you use Darwin Information Typing Architecture (DITA) to create smart, structured content for GenAI.
🔥 For more AI News brought to you via email subscribe to our newsletter here.
👀 Want to know more about Shelf’s suite of AI solutions? Check out our website here.
🎦 Subscribe to our YouTube channel for the latest updates and educational videos on all things AI.
