The Surging AI Job Market

Author

AI Breakfast
November 29, 2023

Good morning. It’s Wednesday, November 29th.

Did you know: Atari’s classic Pong turns 51 years old today.

In today’s email:

  • AI in Consumer Services & Products

  • AI Job Market

  • AI Startup Funding

  • AI Ethics and Governance

  • 9 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Interested in reaching 45,399 smart readers like you? To become an AI Breakfast sponsor, apply here. 

Today’s trending AI news stories

AI in Consumer Services & Products

> Pika Labs unveils Pika 1.0, an AI video generator capable of creating and editing videos in various styles like 3D animation, anime, and film. This version, following successful funding rounds totaling $55 million, enhances video creation accessibility. Pika 1.0, comparable to Midjourney and Stable Diffusion in images, now integrates logos and texts into videos. Competing with Runway and Stability AI's Stable Video Diffusion, Pika Labs reasserts its market position with Pika 1.0's launch.

> Stability AI introduces SDXL Turbo, a real-time text-to-image generation model using a novel Adversarial Diffusion Distillation (ADD) technique, drastically reducing generation steps from 50 to just one. Outperforming other multi-step models in quality with lower computational demands, SDXL Turbo is detailed in a research paper highlighting its innovative distillation technology. Available for non-commercial use on Hugging Face, it demonstrates real-time capabilities on Clipdrop’s image editing platform.

> Amazon unveils Amazon Q, a workplace-focused AI chatbot, challenging perceptions of its AI progress lag. Q assists with internal tasks like document summarization and policy inquiries, ensuring data security and privacy, crucial for enterprise use. Built on the Bedrock platform, integrating various AI systems, Q is priced competitively at $20/month per user. Amazon also strengthens its AI infrastructure, collaborating with Nvidia on a pioneering AI supercomputer project.

AI Job Market

> Demand for generative AI skills has surged, with Meta, Amazon, and Capital One leading in job postings. Companies across various sectors seek AI expertise, reflecting a paradigm shift in work requirements. Lightcast’s study highlights 10,113 job postings related to generative AI, from tech firms to education platforms. Major players like Meta, with 330 listings, and Capital One, with 305, offer roles paying over $200,000, indicating a competitive job market in AI-related fields.

> AI jobs pay $100000 or more—and many can all be done from home. LinkedIn data reveals the surprising leads in AI job hires: professional services, retail, and manufacturing outpace tech. Not just tech roles, demand extends to non-technical positions like sales managers and supply chain specialists, showcasing a growing need for AI literacy across diverse job functions.

> Wall Street banks, including Goldman Sachs, are fiercely competing for AI talent. In the past year, Goldman Sachs experienced a net loss of 60 AI specialists to rivals like Morgan Stanley and Citigroup. Conversely, Wells Fargo gained 130 staff, showcasing the dynamic talent movement. Banks like Citigroup and JPMorgan Chase are actively enhancing their AI capabilities, with JPMorgan Chase adding 325 AI-focused employees recently.

> A survey reveals a growing demand for AI skills among employers, with 73% prioritizing such hires. Despite this, 75% struggle to find qualified candidates. The survey indicates potential salary increases for AI-skilled workers across various departments, with hikes up to 43% in sales and marketing, 42% in finance, and 35% in human resources. Companies are exploring various strategies to address the AI talent shortage, such as educational partnerships and leveraging gig workers.

> Europe now surpasses the US in AI talent, with a 10x increase in AI professionals over the last decade, according to Atomico's research. This growth has fueled European tech startups, with AI firms raising 11 of 36 mega-rounds in 2023. The continent has also gained four AI-focused unicorns: DeepL, Helsing.ai, Synthesia, and Quantexa. Despite layoffs, Europe's tech workforce has grown, benefitting from net talent migration from the US. The rise in tech workers has sparked a surge in new startups, with 14,000 European founders compared to 1,000 fewer in the US. However, European startups face challenges in VC funding.

AI Startup Funding

> Cradle, an AI-driven biotech startup, secures $24M for protein programming, impressing drug development giants like Johnson & Johnson. Their generative approach streamlines protein design, potentially halving development time. Beyond drugs, their tech may impact food and industrial sectors. Despite EU challenges, Cradle finds talent and proximity to key clients advantageous. The funding aims to expand sales and team. Cradle's innovation reshapes biotech with AI-guided protein engineering.

> PhysicsX emerges from stealth with a $32M funding round focused on AI-driven engineering simulations. Co-founded by two theoretical physicists, including a Formula One engineering superstar, the London-based startup offers an AI platform for simulations in automotive, aerospace, and materials science manufacturing. Led by General Catalyst, the Series A funding aims to advance simulations’ accuracy and speed, targeting physical system optimizations across industries. The platform’s goal is to transform engineering in complex sectors by revolutionizing simulation engineering and machine learning.

> Dell collaborates with AI research company Imbue on a $150 million project to develop a high-performance computing cluster for training foundation AI models. Imbue focuses on creating foundation models with advanced reasoning abilities, like planning and self-critique. The cluster, powered by Dell PowerEdge servers and NVIDIA H100 GPUs, supports Imbue's development of AI agents for diverse applications. Dell's technology aids Imbue in exploring impactful AI innovations, with a focus on creating practical, autonomous AI agents for real-world tasks.

AI Ethics and Governance

> Sports Illustrated has been accused of publishing AI-authored pieces and fake contributor profiles surfaced on SI and TheStreet.com, fueling concerns about journalistic integrity. The SI Union demanded clarity, disassociating from unethical practices. Arena Group denied the AI content allegation, stating it was licensed product reviews. Subsequently, they terminated their partnership with AdVon Commerce, the alleged source of AI-generated content.

> The California Privacy Protection Agency (CPPA) proposed comprehensive AI regulations allowing opt-out rights, access to data use info, and limits on AI-based profiling. Modeled partially on EU rules, the regime aims for tighter control over AI’s impact, possibly challenging firms like Meta reliant on user profiling for ads. CPPA’s risk-based approach echoes the EU’s AI Act, signaling California’s potential influence as a key AI rulemaker.

> OpenAI is not expected to offer board seats to major investors including Microsoft, Khosla Ventures, and Thrive Capital, despite their significant financial backing. This decision follows the recent turbulence involving the temporary ouster and reinstatement of CEO Sam Altman, who did not regain his board position. The new nine-person board will prioritize safety and ethical AI practices, with the initial three directors - Bret Taylor, Larry Summers, and Adam D'Angelo - to be confirmed soon. D'Angelo is the only holdover from the previous board that fired Altman. This approach indicates OpenAI's commitment to independence and responsible AI development over investor influence.

5 new AI-powered tools from around the web

Vector to 3D enables users to generate customizable 3D models by adjusting settings such as materials, lighting, resolution, and more. Create 3D effects from different perspectives and viewing angles with ease.

Claude 2.1 marks a significant leap for enterprises, offering a 200K token context window, reduced model hallucination rates, and API tool integration.

Hyper offers AI-powered live data chat integration for streamlined productivity. Consolidate apps into a unified interface, access live data, execute tasks with AI agents, and interact seamlessly across platforms for enhanced productivity.

Malloy is an AI-powered video transcription that excels with slangs, accents, and jargons. Swift, accurate, and error-free transcription of videos and podcasts, even with complex linguistic elements.

Zixflow is an AI-driven CRM platform offering a unified workspace for customer management, workflow automation, and engagement through Email, SMS, and WhatsApp, featuring a sleek UI, customizable options, and robust security.

ob1 by Outerbase automates backend creation, generating APIs, databases, and functions from simple prompts. It offers instant schema generation, function creation, and deployment capabilities, streamlining backend development for apps and businesses.

Walling AI transforms ideas into visually rich project walls instantly, offering features like custom domains, password protection, and slide sharing for efficient idea organization and presentation.

BounceBan GPT, a specialized ChatGPT, offers unlimited, free email verification, including catch-all emails, streamlining both single and bulk processes, essential for email marketing and productivity tasks.

arXiv is a free online library where researchers share pre-publication papers.

The study introduces "Style Tailoring" for refining Latent Diffusion Models (LDMs) in sticker image generation. Employing multi-stage fine-tuning, it adapts a base model like Emu with millions of sticker-like images and human-in-the-loop datasets to enhance prompt alignment and style conformity. The method employs a new training technique, balancing prompt and style alignment, demonstrating notable improvements—14% in visual quality, 16.2% in prompt alignment, and 15.3% in scene diversity compared to prompt-engineered Emu.

The paper introduces ToolTalk, a benchmark aimed at evaluating Large Language Model (LLM)-empowered assistants through intricate user intents requiring multi-step tool usage within conversational contexts. ToolTalk encompasses 28 diverse tools grouped into 7 plugins, facilitating the automated assessment of assistants reliant on execution feedback. Distinguishing itself by focusing on tools that affect the external world, rather than solely referencing information, ToolTalk evaluates GPT-3.5 and GPT-4, achieving success rates of 26% and 50% respectively.

The paper addresses the challenge of generating seamless 360-degree panoramic images using diffusion models, crucial for maintaining the continuity between the rightmost and leftmost sides. Present diffusion methods struggle to create such images flawlessly. It proposes a circular blending approach, optimizing both denoising and VAE decoding stages to sustain geometric continuity. The models cater to Text-to-360-panoramas and Single-Image-to-360-panoramas tasks. Despite improvements over prior studies like MVDiffusion, StitchDiffusion, and PanoDiff, their limitations motivated the researchers’ innovative circular blending strategy, enhancing image quality and continuity. The code is available on GitHub.

The paper explores LoRA’s resource efficiency and performance in adapting LLMs for specific tasks. While ChatGPT excels in various tasks, LoRA’s low-rank limits its effectiveness in complex multi-task scenarios, dominated by a few top singular vectors. MultiLoRA horizontally scales LoRA modules and alters parameter initialization, mitigating dominance, and achieving more balanced unitary subspaces. Experimental results demonstrate MultiLoRA’s superior performance over LoRA and full fine-tuning across diverse benchmarks and model scales, significantly reducing dependency on top singular vectors, indicating a more equitable contribution from unitary subspaces.

Adapters, an open-source library, revolutionizes transfer learning in large language models by unifying 10 diverse adapter methods into a cohesive interface. This framework empowers users with versatile, efficient, and modular transfer learning. Offering flexibility through composition blocks, Adapters enables complex setups while preserving parameter efficiency. Evaluated across various NLP tasks, Adapters showcases competitive performance against full fine-tuning paradigms. Integrated seamlessly into the Transformer ecosystem, this library addresses conventional fine-tuning challenges, fostering efficient and adaptable transfer learning practices.

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.