• NATURAL 20
  • Posts
  • Cognition Labs introduces DevIn: The First AI Software Engineer

Cognition Labs introduces DevIn: The First AI Software Engineer

PLUS: Amazon Enhances AI for Sellers, EU Adopts Groundbreaking AI Act and more.

Sponsored by

Today:

How do you stay up-to-date with the insane pace of AI? Join The Rundown – the world’s fastest-growing AI newsletter with over 500,000+ readers learning how to become more productive using AI every morning.

1. Our team spends all day researching and talking with industry experts.

2. We send you updates on the latest AI news and how to apply it in 5 minutes a day.

3. You learn how to become 2x more productive by leveraging AI.

World's First AGI Agent SHOCKS the Entire Industry! (FULLY Autonomous AI Software Engineer Devin)

Cognition Labs introduces Devin, the first AI software engineer, capable of autonomously tackling complex tasks like benchmarking API performance, debugging code, and building websites. Devin's abilities stem from advancements in reasoning and long-term planning.

With impressive capabilities like learning from blog posts and autonomously generating code, Devin redefines efficiency in software development. Promising advancements hint at a future where AI agents like Devin become integral in businesses, streamlining processes and driving innovation. The transition towards widespread adoption may occur within a decade, transforming the landscape of work and productivity.

A generalist AI agent for 3D virtual environments

DeepMind researchers introduce SIMA, a versatile AI agent trained on various video games to understand and execute tasks based on natural-language instructions. Unlike previous AI systems focusing on single games, SIMA demonstrates the ability to navigate diverse gaming worlds, showcasing potential for broader applications beyond gaming. 

Developed in collaboration with eight game studios, SIMA's training spans nine games, including No Man’s Sky and Teardown. The agent's effectiveness relies on language comprehension, emphasizing its adaptability and potential for real-world usefulness. By advancing general AI capabilities, DeepMind aims to create agents capable of performing a wide range of tasks efficiently and safely.

Anthropic releases Claude 3 Haiku, an AI model built for speed and affordability

Anthropic, a San Francisco startup, launches Claude 3 Haiku, a lightning-fast and budget-friendly AI model. Haiku joins its siblings, Opus and Sonnet, offering top-tier visual processing and speed, making it ideal for high-volume and time-sensitive tasks like customer support chats. With the ability to handle 21,000 tokens per second, Haiku accelerates document analysis, such as legal cases or contracts, saving valuable time. 

Anthropic prioritizes security, ensuring robustness against harmful outputs. The release meets rising demand for efficient AI solutions in diverse enterprise settings, offering versatility within the Claude 3 family. Haiku is available via Anthropic’s API and for Claude Pro subscribers, with plans for Amazon Bedrock and Google Cloud Vertex AI integration.

Microsoft Copilot is now more accurate and powerful, thanks to a free upgrade to GPT-4 Turbo

Microsoft Copilot receives a significant upgrade, now powered by GPT-4 Turbo, offering a more recent knowledge base and a larger context window. The free version now utilizes GPT-4 Turbo, enhancing its capability to handle information and generate responses. Copilot Pro subscribers retain the option to use GPT-4. 

The change promises better performance and accuracy, catering to users' evolving needs. GPT-4 Turbo boasts a knowledge cutoff of April 2023 and a 128k context window, distinguishing it from its predecessor. Microsoft combines Bing search data with OpenAI's technology to power Copilot, delivering improved AI-generated responses.

OpenAI’s Sora text-to-video generator will be publicly available later this year

OpenAI's Sora, a text-to-video generator, will soon be accessible to the public, potentially revolutionizing content creation. Initially aimed at visual artists and filmmakers, Sora creates hyperrealistic scenes from text prompts. Future plans include incorporating audio for added realism and enabling users to edit generated content. However, concerns regarding data sourcing and potential misuse persist, with Sora likely restricted from generating images of public figures. 

Despite its promise, Sora's operational costs pose a challenge, though efforts are underway to ensure affordability upon release. As generative AI tools advance, so do the accompanying ethical and practical considerations.

Amazon selling partners can now access even more generative AI features to create high-quality product listings

Amazon is empowering sellers with enhanced generative AI tools to streamline the creation of high-quality product listings. These tools utilize sparse text or images provided by sellers to automatically generate compelling product titles, descriptions, and attributes. Over 100,000 selling partners have already utilized these features, reporting time savings and improved listing quality. 

Now, sellers can further simplify the process by providing a URL to their existing product listings, which will be parsed by Amazon's AI to create engaging listings. This innovation aims to boost seller efficiency and customer satisfaction, ultimately enhancing the shopping experience on Amazon.

Microsoft’s AI Copilot for Security launches next month with pay-as-you-go pricing

Microsoft is launching Copilot for Security, an AI chatbot for cybersecurity professionals, powered by OpenAI's GPT-4. Scheduled for release on April 1st, the service will feature pay-as-you-go pricing of $4 per hour, allowing businesses to scale usage as needed. 

Copilot for Security provides real-time information on security threats and includes collaboration tools for cybersecurity teams. Microsoft's move towards AI-driven cybersecurity coincides with recent attacks on its systems, prompting a renewed focus on software security.

EU Parliament officially adopts AI Act — landmark regulation likely to become law in May

The European Parliament greenlit the AI Act, regulating high-risk AI systems and AI-human interactions. It’s set to take effect in May, impacting global firms, notably US companies. The Act aims for trustworthy AI, with hefty fines for non-compliance. Experts stress the need for swift action in assembling AI compliance teams. 

IBM and Salesforce commend the EU for its risk-based approach, pledging support for compliance. While major players like OpenAI and Google remain silent, other tech giants endorse the Act. The EU's move sets a standard for responsible AI, urging global adoption.

New York Times: OpenAI’s claim we 'hacked' its products both 'irrelevant' and 'false'

The New York Times fires back at OpenAI's accusations of "hacking" its products, labeling them both irrelevant and false in their latest legal exchange. Amidst claims of copyright infringement, NYT argues that OpenAI's defense is more spin than substance. 

OpenAI's alleged use of NYT content in training its models without permission remains a central issue. The legal tussle stems from NYT's accusations against Microsoft and OpenAI for using its content without consent, with OpenAI countering that NYT manipulated its ChatGPT. The dispute underscores broader concerns about copyright and AI's role in media reproduction.

🧠RESEARCH

MoAI is a new model that enhances scene understanding by incorporating data from computer vision (CV) models like object detection and OCR into language and vision tasks. Unlike previous methods that rely solely on large language models, MoAI uses a unique mix of visual, auxiliary, and language features to excel at understanding real-world scenes, all without needing bigger models or extra data. It's a smart mix, giving it an edge in tasks like identifying objects and their relationships.

"Chronos" revolutionizes forecasting by tokenizing time series data for use with transformer-based language models, enhancing prediction accuracy. It's trained on a vast array of public and synthetic datasets, showing stellar performance across 42 benchmarks. This method stands out for its ability to adapt and predict new datasets with little to no prior exposure, simplifying forecasting processes significantly.

Branch-Train-MiX (BTX) is a method for training Large Language Models (LLMs) to excel in various specialized domains like coding and math reasoning. BTX begins with a base model, branches out to train domain-specific experts in parallel for efficiency, and then integrates their expertise into a Mixture-of-Experts (MoE) model. This approach, which also includes an MoE-finetuning stage for optimal token-level routing, offers a superior accuracy-efficiency balance compared to previous methods. 

"Synth^2" tackles the bottleneck in developing Visual-Language Models (VLMs) by creating synthetic image-text pairs using Large Language Models and image generation technologies. This innovative approach allows for efficient VLM training with synthetic data, showing comparable or superior performance to models trained on human-annotated datasets, and significantly speeding up the process. This method opens up new possibilities for generating customizable image datasets, enhancing VLM applicability across various fields with better data and resource efficiency​.

FAX is a JAX-based library that supports scalable, distributed, and federated computations both in data centers and across devices. By integrating federated computing primitives directly into JAX, FAX allows computations to be efficiently executed on TPUs and modern JAX runtimes. This integration also simplifies federated automatic differentiation, making it easier to implement federated computations. FAX aims to provide a highly programmable, high-performance framework for federated learning tasks, available on GitHub for further exploration​.

🛠️TOP TOOLS

Repeto- upload your document. Chat with it. Take smart notes. Generate interactive quizzes. Visualize hard topics.

Emojifyer- Turn anything into Emoji

Summate - sums up your newsletter subs in a snap

Rio - catch up on news with an AI news anchor

You - Use Smart Mode to accomplish tasks quickly, draft unique content, and access real-time web results

Ultimate - Automate +60% of your support requests across all digital channels with the latest generative AI technology

Attention - offers AI for sales in real-time, eliminating over 50% of administrative tasks for sales reps and reducing sales cycles for leaders

Tavus - AI video personalization platform that allows users to create a single video template which can then be automatically customized into hundreds, thousands, or even millions of unique videos

📲SOCIAL MEDIA

🗞️MORE NEWS

No, this isn’t CG.

Figure's humanoid robot now utilizes OpenAI's large vision language model (VLM) for reasoning and understanding language. Demonstrated in a video, the bot adeptly identifies and interacts with objects upon request, such as asking for food. It showcases advanced capabilities beyond mere computer-generated imagery (CG). THE VERGE

Locus Robotics’ success is a tale of focusing on what works

Locus Robotics, led by CEO Rick Faulk, emphasizes its software prowess despite its robot-centric appearance. The firm's latest offering, LocusHub Engine, showcased at Modex, leverages AI for predictive analytics, optimizing warehouse operations. Faulk's strategic focus on client needs fuels Locus' steady growth, solidifying its position as a market leader in warehouse automation. TECHCRUNCH

Physical Intelligence Is Building a Brain for Robots

Physical Intelligence, a startup founded by robotics and AI experts, aims to revolutionize robotics by developing software that adds advanced intelligence to various machines. Unlike traditional AI systems, which often accompany physical robots, Physical Intelligence focuses on software that can be applied universally across different types of robotics, expanding their capabilities beyond specific tasks. BLOOMBERG

Pinterest launched an AI tool that lets users filter search results by body type

Pinterest has introduced an AI tool enabling users to filter search results by body type, initially for women's fashion and wedding items in the U.S., with plans for expansion to men's fashion. Despite aiming for inclusivity, the tool's 16 choices may not fully represent all body types, and a prompt to save selections was reportedly missing in tests. THE VERGE

UK chips in $44M for a piece of Europe’s $1.4B pot for semiconductors

The UK invests $44 million in Europe's $1.4 billion semiconductor initiative, aiming to bolster its chip industry post-Brexit. Joining the EU's project, it hopes to boost R&D and compete globally. With applications open until May 14, UK firms eye the funds for AI and tech advancements, recognizing the power of collaborative innovation. TECHCRUNCH

Akamai and Neural Magic team up to accelerate AI workloads on edge CPU servers

Akamai partners with NeuralMagic to enhance AI processing on CPU servers, boosting deep learning capabilities. NeuralMagic's software enables efficient AI workloads on cost-effective CPU servers, extending Akamai's cloud infrastructure. This collaboration aims to deliver advanced AI applications with low latency, particularly beneficial for edge computing. The move aligns with Akamai's transition to edge-based cloud infrastructure, offering cost efficiencies and high performance for customers. SILICON ANGLE

Together AI valued at $1.25 billion in Salesforce-led funding round

Together AI, backed by Nvidia, secures $106 million in funding led by Salesforce Ventures, valuing it at $1.25 billion. The startup aims to leverage Nvidia's GPUs to offer AI model customization and deployment services. The funding will fuel expansion, focusing on cloud computing and hiring researchers. This reflects investors' interest in AI startups amid technological advancements. YAHOO!

Larry Ellison and Elon Musk teaming up to bring AI to farming

Larry Ellison and Elon Musk are collaborating to develop an AI-powered mapping application for farms. Announced during Oracle's earnings call, the tool aims to assist farms in planning and increasing agricultural output by assessing factors like water and nitrogen levels. SpaceX's involvement expands its agricultural ventures, following a partnership with John Deere. BUSINESS INSIDER

Elon Musk says ‘AI will probably be smarter than any single human by next year’. How close are we to AGI?

Elon Musk predicts AI surpassing individual human intelligence by next year and outmatching collective human intellect by 2029. Disagreement among tech leaders persists regarding the definition and arrival timeline of Artificial General Intelligence (AGI), with differing views on its potential impact. AGI, the stage where AI can perform tasks on par with or better than humans, remains contentious in the tech community. Some argue current AI systems, like chatbots, are insufficient for AGI, while others anticipate unprecedented benefits from its emergence. MINT

OpenAI Made AI Videos for Us. These Clips Are Good Enough to Freak Us Out.

OpenAI's Sora AI video tool, capable of generating realistic scenes from text, raises questions about its training methods. In an interview with OpenAI's CTO Mira Murati, details about Sora's functionality are shared while avoiding inquiries about its training process. Concerns over AI's capabilities continue to emerge. THE WALL STREET JOURNAL

The road map to AI's next level could be nature

A new firm, Verses AI, proposes a nature-inspired approach to achieve Artificial General Intelligence (AGI), diverging from current large-scale models like OpenAI's GPT-4. Verses advocates for distributed intelligence, focusing on autonomy and efficiency akin to biological systems. With industry giants like NASA and Volvo testing its Genius operating system, Verses challenges conventional AI development. AXIOS

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.