AI-native products and companies are growing faster and engaging users more deeply than ever before. Over the coming decade, AI will likely underpin category-defining companies. This article examines data on the most popular generative AI products to uncover patterns on how consumers are using the technology. It looks at the rate of progress in the industry, emerging categories, hot niches, the mobile vs web split, and AI as a global pursuit.

Researchers have developed an AI network where one AI can teach another to perform tasks using natural language processing, a capability not previously demonstrated. The system uses a model called S-Bert that allows AI to perform tasks given via instructions and then communicate that knowledge to another AI. This breakthrough has potential applications in robotics and could further understanding of human cognitive functions.

Researchers have developed DiJiang, a new approach that transforms existing Transformers into leaner, faster models without the heavy burden of retraining.

Extracting information from datasets is critical for enterprise AI applications. These five new benchmark datasets can be used to measure general algorithmic performance for RAG applications.

OpenAI is expanding its Custom Model program with assisted fine-tuning and custom-trained models to help enterprise customers develop tailored generative AI models for specific use cases.

Midjourney’s new algorithm can use the same character across multiple images and styles without deviating too far from their original design. The feature isn’t designed to replicate real people from photographs and works best on characters generated via Midjourney.

Peter Deng, OpenAI's VP of Consumer Product, dodged a question about artist compensation for training data, reflecting the delicate legal position that OpenAI and similar companies find themselves in when using data to train generative AI.

A fast way to OCR a document for AI ingestion. You can pass in a PDF URL or a file itself, which is then transformed and returned as Markdown.

A new toolkit from Nvidia that allows checkpointing the CUDA state for transfer and restarting. It is useful for distributed training of very large AI models.

Apple is considering using Baidu's Ernie and other third-party AI engines to enhance its capabilities. This could lead to challenges in consistency and privacy concerns due to cloud reliance and Ernie's alleged ties to the People's Liberation Army cyber warfare division. It could also pose a significant hurdle in delivering a unique user experience in the competitive AI market.

Microsoft wants to make generative AI applications more secure and trustworthy with new tools in Azure AI that help product builders deal with prompt injection attacks, hallucinations, and content risks. These include Prompt Shields to detect and block direct and indirect prompt injection attacks, Groundedness Detection to identify text-based hallucinations, and Safety System Message Templates to help mitigate misuse and harmful content. Azure AI Studio lets organizations measure an AI application's susceptibility to many threats.

Oasis AI has introduced a distributed AI inference platform enabled by a distinctive browser extension in an attempt to entice a wider user base, irrespective of their technical expertise, to participate in the AI space. The platform facilitates decentralized computing, makes APIs easily accessible, and promotes AI inference at scale. In the upcoming weeks, an extension for providers, an inference platform, and enterprise APIs are expected to be released.

Three major players in the crypto AI space, SingularityNET, Fetch.ai, and Ocean Protocol, have announced a merger to create the ‘Artificial Superintelligence Alliance’, a decentralized AI network. The alliance merges $FET, $OCEAN, and $AGIX tokens into one - $ASI. This groundbreaking move aims to accelerate the path towards Artificial General Intelligence, provide an alternative to Big Tech's AI advancements, and facilitate more ethical AI development and utilization.

In his annual letter to shareholders, Amazon CEO Andy Jassy laid out a vision for how generative AI could become the company's next pillar of growth. The company has invested $4 billion in AI startup Anthropic and added prominent computer scientist and AI expert Dr. Andrew Ng to its board. Amazon is taking a three-pronged approach to AI innovation: it will focus on AI models, applications built on top of the models, and the chips that power the technology. The company is optimistic that a lot of world-changing AI will be built on top of AWS.

AI tools like Devin claim to replace software engineers. These startups face an uphill battle against established AI coding assistants like GitHub Copilot, so their claims need to be bold to get attention. For decades, there have been attempts to automate away the need for programmers, yet software engineers remain high in demand. The future likely lies in improved versions of existing AI coding assistants, not standalone "AI developers."

Eight Google researchers made a breakthrough in 2017 with a paper titled "Attention Is All You Need," which introduced the Transformer architecture now powering AI systems like ChatGPT. The paper demonstrated the power of attention mechanisms and outperformed existing methods. The eight authors have since left Google to found successful AI startups. Their groundbreaking research has had a widespread impact.

OpenAI's Sora model for AI-generated video builds on diffusion models instead of operating on raw pixels. It works in a compressed 'latent space' and uses the Transformer architecture. Like with large language models, throwing more compute at Sora yields better results. Sora is good enough for real-world use, but it is expensive as the inference cost could require hundreds of thousands of GPUs at its peak.

Google designed the TPU v1 for fast, cost-effective inference using trained neural network models at scale. Its key feature is a focus on tensor operations, specifically matrix multiplications, which are core to neural network computations. The TPU v1 is 15-30x faster than contemporary CPUs/GPUs for inference. It has 25-29x better performance per watt than GPUs.

ChatGPT is now instantly accessible without signup.

This interview with Jackie Rocca, the VP of Product at Slack, delves into the role of generative AI in shaping modern collaboration platforms. She explains how Slack has evolved from a messaging tool to a comprehensive workspace, integrating AI to address user challenges such as information overload and knowledge retrieval. Slack offers AI-powered features like channel recaps, thread summaries, and search answers to increase productivity through relevant insights. The interview highlights the iterative nature of AI development and the importance of keeping the user needs in focus to drive innovation that really impacts their experience.

Paid search engine Kagi has spread itself too thin with multiple niche projects. The product is also starting to become over-reliant on AI, which leads to inaccuracy and bias that goes against what Kagi was in the beginning.

Devin AI has been promoted with claims that it can perform software engineering tasks by itself. However, it has been revealed that the company cherry-picked tasks and hid the true capabilities of Devin AI with flashy demos.

The “Joel Test” rates the quality of a software development team. It should be updated so that employers who allow the use of AI tooling while coding are given a boost.

Consumers now have subscriptions for everything — streaming services, software, and grocery delivery. App subscription fatigue is about to get even worse with AI. Major tech companies are using the AI boom to push consumers toward pricier subscriptions. They're also using consumers' data to help train various gen AI tools, meaning the consumers are effectively paying for these tools twice.

Apple has aggressively expanded its artificial intelligence capabilities by hiring numerous AI experts from Google and establishing a secretive AI research lab in Zurich, Switzerland. The lab, known as the "Vision Lab," focuses on developing advanced AI models that integrate text and visual inputs.

Diddo is a CV/AI API for streaming and media companies that can make their content instantly shoppable without the use of QR codes or second screens, opening up new revenue streams for them. Diddo's API keeps the purchase capabilities on platform and 100% native so it is truly an end-to-end commerce solution for video owners.

Cognition Labs, which is transitioning from crypto to AI, is seeking a $2 billion valuation for its code-writing AI tool, Devin. Amidst a trend of soaring valuations in AI startups, Cognition's success reflects the sector's growth and the importance of substantial investments in data and computational resources. Key players like Google and Microsoft continue to advance aggressively in AI with large language models.

Current AI leaderboards aren't useful anymore because they don't account for the cost of running AI agents. Researchers should instead use Pareto curves to visualize the tradeoff between accuracy and cost. Surprisingly, simple baseline agents can be just as accurate as the complex, costly agents that top leaderboards.

The AI-powered search engine Perplexity, which has tens of millions of users with less than 50 employees, uses AI internally to inform all aspects of building the company. It has small teams of 2-3 people that work in parallel to reduce any coordination costs. The company hires self-driven and very proactive engineers.

While Apple is investing heavily in AI across its product lineup, it won't be building massive data centers to support it. Instead, Apple will follow its typical 'hybrid' cloud approach, using third-party services with its own infrastructure combined. Apple is expected to announce more AI-powered features at WWDC in June.