• AI Breakfast
  • Posts
  • Unemployed Man Uses AI to Apply for 5,000 Jobs, Gets 20 Interviews

Unemployed Man Uses AI to Apply for 5,000 Jobs, Gets 20 Interviews

Good morning. It’s Friday, November 10th.

In today’s email:

  • AI Industry Expansion and Partnerships

  • AI Tools and Applications

  • AI in Hardware and Devices

  • AI and the Job Market

  • AI Advancements and Predictions

  • AI in Entertainment and Creative Rights

  • AI Data Management and Security

  • AI Disruptions and Challenges

  • 6 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.

Today’s trending AI news stories

AI Industry Expansion and Partnerships

Amazon dedicates team to train ambitious AI model codenamed 'Olympus' a large language model with 2 trillion parameters, aiming to compete with OpenAI’s GPT-4. Spearheaded by Rohit Prasad, the model unites AI efforts across the country, potentially enhancing AWS offerings to attract enterprise clients seeking top AI models. The project’s details remain confidential, and Amazon has not commented publicly.

Google Announces Expansion of AI Partnership with Anthropic The collaboration will utilize Google’s latest TPY v5e technology for efficient AI inference, ensuring responsible AI advancements. Anthropic will benefit from Google’s security services, safeguarding its deployment of AI models. This enhanced cooperation underscores a mutual dedication to the ethical development of AI technologies.

AI Tools and Applications

Hugging Face has a two-person team developing ChatGPT-like AI models with a focus on alignment, that is teaching language models to behave according to human or AI feedback. Starting in January, H4 has released open-source language models, including a chat-centric model Zephyr-7B-α, and improved another, Falcon-40B, for better natural language responses. H4’s work, although not directly monetized, supports Hugging Face’s enterprise program, contributing to the company’s AI solutions without the pressure of AI commercialization.

Microsoft plans to bring its AI Copilot to 1 billion Windows 10 users through an update that adds a Copilot button to the taskbar, similar to Windows 11. This move aims to tap into the vast market of 1 billion active Windows 10 devices and encourage developers to create Copilot plugins by increasing its user base to 1.4 billion devices. The experience on Windows 10 will mirror that on Windows 11, including plugin compatibility. This strategy marks a shift from quickly moving on from Windows 10 to keeping users up to date with new features.

GitHub teases Copilot enterprise plan that lets companies customize for their codebase at $39/month starting February 2024. The announcement came during GitHub Universe, where they also revealed plans for Copilot Chat’s general availability in December and future support for JetBrain’s IDEs. Additionally, GitHub is launching a Copilot Partner Program with third-party integrations and gave a sneak peek at Copilot Workspace, expected in 2024.

AI in Hardware and Devices

The Humane Ai Pin Launches Its Campaign to Replace Phones Humane Inc., initiated by ex-Apple personnel Imran Chaudhri and Bethany Bongiorno, introduced the Ai Pin, an integrated device that attaches to clothing. Retailing at $699, plus a $24 monthly subscription for limitless data and calls, the Ai Pin is positioned as a contemporary alternative to conventional phones. Collaborating with T-Mobile, Microsoft Corp., and OpenAI for technology support, the Ai Pun will be up for orders starting November 16.

Nvidia develops three new AI chips for China to comply with US export controls Nvidia has introduced three new AI chips tailored for the Chinese market to align with U.S. export controls. These chips, named H20, L20, and L2, are designed to fulfill China's AI demands without breaching U.S. sanctions. Despite reduced performance compared to previous models sold in China, the new GPUs aim to offer competitive capabilities within the restrictions. Nvidia's move comes as part of its strategy to maintain sales growth in China, navigating the complexities of U.S. geopolitical constraints.

AI and the Job Market

Unemployed Man Uses AI to Apply for 5,000 Jobs, Gets 20 Interviews Software engineer Julian Joseph used the AI-powered tool LazyApply to automate his job application process. This method, while showcasing efficiency, also underscores the deficiencies in the modern job application process. It reflects a shift toward an automated battleground where both job seekers and employers increasingly rely on AI for application and screening processes, suggesting a need to update and humanize the system for efficiency and fairness.

AI Advancements and Predictions

OpenAI's Altman says today's AI will be "quaint" by 2024, in line with first GPT-5 rumor OpenAI CEO previewed the future of AI at the company’s developer conference, suggesting that current AI, including GPT-4, will seem “quaint” compared to what’s being developed for 2024. Amid speculation, GPT-5 is rumored to be a multimodal model codenamed “Gobi,” potentially capable of handling text, images, and video, with an expected early 2024 release. Despite challenges and the need for more computational power, OpenAI is aiming for GPT-5 to work for “most things you want to build.”

NVIDIA's Eos supercomputer just broke its own AI training benchmark record NVIDIA has accelerated the training of large language models with its new Eos supercomputer, which is equipped with over 10,000 H100 Tensor Core GPUs. This advancement reduces the time to train a 175 billion-parameter GPT-3 model on 1 billion tokens to under four minutes, tripling the speed of previous benchmarks. Eos boasts 40 exaflops of AI processing power and offers cloud services for AI expansion at $37,000 per month. The new records set include training times for various AI models, highlighting both the supercomputer’s efficiency and the optimization of software.

NVIDIA Shows Intel Gaudi2 is 4x Better Performance Per Dollar than its H100 NVIDIA's display of MLPerf Training v3.1 results highlighted Intel Gaudi2's cost-efficiency, claiming four times the performance per dollar compared to NVIDIA's H100. While NVIDIA's new Eos supercomputer with 10,752 H100 GPUs set records, the company's comparison inadvertently showcased Gaudi2's value. Despite using fewer GPUs, Gaudi2 delivered competitive performance, suggesting a strong market position, especially when system simplicity and costs are considered.

AI in Entertainment and Creative Rights

Hollywood Actors Strike Ends With a Deal That Will Impact AI and Streaming for Decades SAG-AFTRA’s new tentative agreement sets a precedent in Hollywood, focusing on AI’s role in performers’ rights. The deal outlines terms for the use of actors’ AI-generated images and performances, establishing consent and compensation frameworks, especially posthumously. It also addresses residuals for streaming, aiming to modernize pay standards for the digital era. This agreement marks a significant shift in the industry, balancing technological advancements with creative rights.

AI Data Management and Security

Decentralized AI platform Ritual lands $25M in seed funding led by Archetype. Set to launch in early 2024, Ritual is positioned to leverage AI for enhanced computational integrity and privacy across both enterprise and Web3, promising to bring the transparency and open-source ethos of Web3 to critical AI technology applications.

OpenAI has announced the establishment of Data Partnerships, a new venture calling for collective efforts to create both open and proprietary datasets designed to refine AI training. This project is dedicated to broadening AI’s comprehension by assimilating an extensive range of data forms, such as text, imagery, audio, and video, particularly those that encapsulate the essence of human purpose. These collaborative partnerships are set to shape the trajectory of AI, offering advantages from tailored models, demonstrated by OpenAI’s recent projects with the Icelandic authorities and the Free Law Project.

AI Disruptions and Challenges

ChatGPT Blames Outages on DDoS Hackers but has now restored functionality for all users, overcoming the outages caused by the cyberattack. The assault, which the hacker group claimed was in support of Palestine, targeted OpenAI’s servers post-DevDay amidst a spike in user activity due to newly introduced features. OpenAI’s CEO had initially attributed the service instability to the heavy usage following the DevDay announcements, not revealing the cyberattack aspect.

ElevenLabs Turbo v2 is Here - An advanced model delivering speech synthesis at around 400ms latency, significantly faster than its predecessor while maintaining high-quality audio output. The new model supports mulaw 8khz output, optimizing speed for VoIP users, with multilingual capabilities in development. Further details and updates can be found in the latest product documentation.

5 new AI-powered tools from around the web

Attio Automations transforms CRM workflows with intelligent, AI-driven automation, featuring smooth integration with various tools and over 20 ready-to-use templates for Go-to-Market strategies, making complex systems both powerful and easily manageable.

BeeDone energizes goal achievement with a gamified, AI-enhanced productivity app that makes task completion fun and rewarding, transforming daily routines into an enjoyable game.

Shader app is an intuitive AR tool, enabling instant look customization with AI, equipped with voice, type, swipe controls, video recording, sharing, and collaborative features for enhanced user engagement and creative fun.

Dubbing AI offers a swift AI voice changer with a vast array of tones, compatible across devices and languages, enhancing gaming and streaming experiences with customizable and emotionally expressive voice cloning.

Giskard offers a scalable ML testing framework to detect biases and vulnerabilities in ML models, including LLMs. It provides automated testing, integration with major ML tools, and is open-source, supporting AI quality and compliance efforts.

Pinokio is an AI-enhanced browser that automates the installation, running, and control of any application. It encourages community engagement by allowing users to discover and share scripts, enhancing functionalities like voice cloning and illusion artwork generation.

arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.

The Large Reconstruction Model (LRM) is a pioneering 3D reconstruction framework that generates detailed 3D models from single images using a scalable transformer architecture. Trained on a diverse dataset of 1 million objects, LRM leverages neural radiance fields to deliver high-quality, category-agnostic 3D reconstructions efficiently and is poised to revolutionize applications in design, gaming, and AR/VR.

NExT-Chat is an innovative Large Multimodel Model that enables chat-based interaction, detection, and segmentation using pixel2emb for versatile location modeling. It shows superior performance on tasks requiring visual comprehension, utilizing embeddings for localization to support multiple output formats like bounding boxes and segmentation masks. NExT-Chat excels in multi-task scenarios, including visual grounding and complex reasoning.

mPLUG-Owl2 is an advanced multi-modal large language model that enhances performance across both text and visual tasks by employing modality collaboration. It features a modularized design with a language decoder as a universal interface and introduces shared functional modules for effective modality integration, along with a modality-adaptive module that maintains modality-specific characteristics. Through extensive experiments, mPLUG-Owl2 demonstrates superior generalization across various benchmarks, marking a pioneering advancement in multi-modal functional models.

3DiffTection is an innovative 3D object detection method that enhances 2D diffusion model features with 3D awareness using geometric and semantic tuning. It outperforms existing benchmarks on Omni3D-ARkitscene dataset, showing robust data efficiency and generalization. However, it requires accurately posed image pairs and is computationally intensive which may limit its real-time application.

GENOME is a neuro-symbolic visual reasoning model that grows and reuses modules to perform tasks like visual question answering and referring expression comprehension. Unlike models that generate code for each task instance, GENOME gradually builds a library of reusable modules, inspired by human learning. It involves module initialization, generation, and execution, with LLMs translating language into module descriptions. GENOME excels in standard tasks, transfers modules to new tasks, and adapts to new tasks with few examples.

Latent Consistency Models (LCMs) expedite text-to-image generation, effectively condensing latent diffusion models into high-fidelity image producers. The LCM-LoRA module uses LoRA distillation, substantially reducing memory overhead while maintaining quality. This approach produces a universally applicable acceleration module, enabling swift, high-resolution image generation across different Stable Diffusion architecture and styles with minimal training resources, showcasing remarkable efficiency and generalization in generative AI tasks.

LLaVA-Plus is a multimodal assistant that enhances large multimodal models with a unique end-to-end training approach. It features a skill repository for dynamic tool activation and integrates a variety of pre-trained models to process and respond to multimodal inputs. This new system demonstrates superior task performance by composing tool outputs on-the-fly, showcasing new capabilities and significant advancements in real-world task fulfillment.

GPT4All is an initiative aimed at democratizing access to large language models (LLMs) like GPT-4. It provides an open-source alternative to proprietary models, addressing issues of cost, accessibility, and censorship. GPT4All began as a single model and has grown into a robust ecosystem offering technical transparency, diverse language support, and user-friendly APIs for wider community use. The project’s evolution showcases the demand for commercially usable, open-source LLMs and sets a precedent for future AI accessibility.

Neural MMO 2.0 is an open-source platform for reinforcement learning, featuring a new task system for diverse objectives and rewards. It supports research on generalization across novel tasks and opponents, with procedurally generated maps and up to 128 agents. The rewrite offers enhanced performance and CleanRL compatibility, fostering an active community and simulating research through a concurrent NeurIPS 2023 competition.

TEAL proposes a unified approach for Multi-modal Large Language Models (MM-LLMs) to handle inputs across modalities by tokenizing them into sequences and embedding them into a joint space. This enables existing text-based LLMs to perform understanding and generation tasks for non-textual modalities like images and audio without complex architecture changes. Experiments demonstrate TEAL’s effectiveness in multi-modal understanding and its simplified scheme for generating non-textual outputs, marking a significant step in multi-modal AI development.

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.