Discover more from WorkWithAI.com's Daily Newsletter
Stability AI unveils a generative AI for coding
And IBM researchers hypnotize ChatGPT
⚡ Today’s Highlights
📰 News: Stability AI unveils StableCode, the latest AI coding assistant + IBM researchers find ChatGPT and Bard can be convinced to assist with cyberattacks
💰 Funding: Chargeflow, Catch+Release, MindsDB, Dropzone AI, Birdstop, and Kavida.ai
🦾 Tools: This ML-powered browser extension secures ChatGPT by detecting and redacting sensitive data
📰 Today's Top Stories
(5 min read) (Source: Stability AI)
TLDR: Stability AI, known for its Stable Diffusion text-to-image model, unveiled StableCode, an open large language model (LLM) to assist users in generating programming code. It has three versions to support various use cases and can manage up to 16,000 tokens.
The model's foundation lies in BigCode data, similarly used by HuggingFace and ServiceNow for their models. However, Stability AI extensively refined the data and employed successive training on specific programming languages for better accuracy.
Differing from many models that use the ALiBi approach, StableCode adopts the rotary position embedding (RoPE) technique. This ensures better attention to all parts of the code, benefiting developers by offering tailored code suggestions based on a broader context.
The Big Picture: The intersection of generative AI and coding promises a transformative shift in software development. As AI models like StableCode simplify and augment the coding process, the nature of work in software development is poised to undergo a significant evolution, enhancing productivity and democratizing programming.
(5 min read) (Source: Axios)
TLDR: IBM researchers have highlighted that generative AI, such as ChatGPT, can be hypnotized into assisting with scams or cyberattacks with minimal coding knowledge. The research indicates that understanding the English language and some background on AI training is sufficient to exploit these models for malicious acts.
This research is particularly timely as cybersecurity professionals gather at the DEF CON conference's AI Village in Las Vegas to assess the security of large language models.
IBM's Chenta Lee demonstrated how AI models can be tricked by presenting tasks as "games" with specific rules. This approach has led certain models, including GPT-3.5 and GPT-4, to give incorrect security advice or even generate malicious code. Notably, different AI models responded variably to these tactics.
There are mixed reactions in the cybersecurity industry to the proliferation of LLMs. While some companies are launching generative AI tools to aid cybersecurity efforts and address the worker shortage, many researchers and officials caution that LLMs can assist even amateur hackers in crafting sophisticated malware and deceptive phishing emails.
(6 min read) (Source: Financial Times)
TLDR: Google is negotiating with Universal Music to legally use artists' voices and melodies in songs created by AI. This aims to provide fans with a legitimate tool to craft these tracks while compensating copyright holders. Warner Music is also in similar talks with Google.
Generative AI's popularity has led to an increase in "deepfake" songs replicating established artists. Notable instances include mimicked voices of Frank Sinatra, Johnny Cash, and a YouTube user imitating rappers Tupac and Notorious B.I.G. Responses are mixed, with some artists like Grimes welcoming the technology while others, like Drake, are concerned about its misuse.
As competitors like Microsoft integrate AI capabilities, such as through their association with OpenAI's GPT-4, Google aims to bolster its standing in the AI domain. Google's AI-driven music software, MusicLM, which crafts tunes based on text descriptions, is a testament to its innovation in the space.
The Big Picture: The fusion of AI and the music industry highlights a pivotal moment for intellectual property rights and artistic expression. As the lines blur between human-created and AI-generated content, the industry grapples with challenges, potential, and a redefined artist-audience relationship.
(15 min read) (Source: VentureBeat)
TLDR: Nvidia collaborates with partners like BOXX, Dell, HP, and Lenovo. They're launching high-end RTX workstations, particularly for generative AI development and content creation, using their RTX 6000 Ada Generation GPUs.
Alongside the hardware, Nvidia emphasizes on tools like the NeMo framework for AI and the Omniverse platform for 3D workflow collaboration. These systems can deliver impressive AI performance and manage intensive graphics tasks.
Nvidia's new GPU lineup includes the RTX 5000, RTX 4500, and RTX 4000, boasting advanced AI, graphics, and rendering capabilities. Separately, the L40S GPU is introduced for compute-intensive applications, supported by big tech manufacturers.
Nvidia's Maxine software improves video quality in real-time communications using AI. Moreover, Nvidia's 3D video tech can convert standard 2D video into 3D, offering a more immersive video conferencing experience.
👀 More Reading:
Meet the hackers who are trying to make AI go rogue (The Washington Post)
AI Botched Their Headshots (The Wall Street Journal)
Why it’s impossible to build an unbiased AI language model (MIT Technology Review)
The Pope Doubles Down on AI Concerns (Futurism)
💰 Funding News
Chargeflow utilizes Generative AI and ML to help e-commerce merchants recover lost revenue by improving chargeback win rates.
Round led by OpenView, and Tom Holahan, a Partner at OpenView, will be joining Chargeflow's board of directors.
Funding is for enhancing its tech development efforts and strengthening its growth trajectory in the US market.
2. Catch+Release is a content licensing marketplace that is creating an AI-powered search engine to enable brands and creators to curate and license existing user-generated content for marketing materials. They raised $8.8M in funding.
Round led by return investor Accel (Eric Wolford, Partner), with participation from Cervin, Astia Fund, Stagwell, HarbourVest Partners, Rich Kleiman + Kevin Durant's 35V, and Nick Mehta.
Funding is for accelerating product features, enhancing user experience, and growing the creator community. Catch+Release plans to further develop their AI-powered search engine to facilitate content discovery and licensing.
Round led by NVentures (Mohamed Siddeek, Corporate Vice President and Head of NVentures), NVIDIA’s venture capital arm with participation from existing investors Mayfield (Navin Chaddha, Managing Partner), Benchmark, Y Combinator, OpenOcean, and Walden Catalyst.
Funding is for expanding platform reach, foster partnerships, and enhancing its innovative approach to integrating AI capabilities into applications.
Dropzone AI utilizes pre-trained AI security agents to investigate alerts from security systems, mimicking the techniques of elite analysts.
Round led by Decibel Partners, with participation from Pioneer Square Ventures Fund and angel investors including Oliver Friedrichs, Jon Oberheide, and Jesse Rothstein.
Funding is for expanding the engineering team and accelerate technology development to support various alert types and security stacks.
Round led by Lerer Hippeau VC.
Funding is for acquiring more drones for inspecting critical infrastructure and to enhance the company's AI capabilities.
6. Kavida.ai, a supply chain startup that offers an order management platform using AI, helping manufacturers and retailers manage procurement and detect issues in supplier orders, raised an additional $1.1M investment in a seed round.
Round led by Qima, a supply chain quality and compliance solutions provider.
Funding is for strengthening and developing its innovative supply chain ChatGPT and recruit AI experts and data scientists to enhance the company's team.
🦾 Cool Tools
1. Relume AI Site Builder is an AI-powered web design tool that accelerates website creation.
Features: Instant sitemap generation, multi-page wireframes with AI-generated copy, Figma and Webflow integrations.
Best for: Web designers, developers, and creative teams seeking an edge in website design efficiency and collaboration.
2. Personal Assistant by HyperWrite is an AI browser agent for seamless web browsing, automation, and research.
Features: Navigate, research, automate online tasks. Operate hands-free while the AI assists you with browser activities.
Best for: Individuals aiming to enhance their web browsing experience, automate tasks, and conduct research efficiently.
3. Modyfi is an AI-native design platform combining graphic design and image-making tools with AI-powered workflows. Collaborate and design effortlessly in your browser.
Features: Streamlined workflows, AI-powered tasks, real-time collaboration, versatile image generation.
Best for: Graphic designers and image-makers aiming for efficient, collaborative, and creative design processes.
4. SuperAGI Cloud is an open-source autonomous AI framework for building, managing, and deploying useful AI agents with ease.
Features: Open-source, dev-friendly, autonomous agent development, agent templates, concurrent AI agents, performance monitoring.
Best for: Developers aiming to streamline autonomous AI agent development, management, and deployment processes.
5. Nightfall for ChatGPT is an ML-powered browser extension that secures ChatGPT by detecting and redacting sensitive data like PII, PCI, and confidential information.
Features: ML-based sensitive data detection, real-time employee education, easy installation via Chrome Store.
Best for: Companies aiming to enhance ChatGPT's security, prevent data leaks, and maintain productivity.