AI News & Launches

Ken Kantzer shares insights from using over 500 million GPT tokens at his startup, highlighting the differences between community narratives and practical usage. Key lessons include the advantage of less detailed prompts, the sufficiency of simple chat APis over complex language tools, and the utility of streaming APIs for user experience. GPT struggles with hallucinations and generating null responses but shows reliable context analysis and task execution.

MIT scientists have utilized AI technology to identify a compound, semapimod, that targets dormant, metabolically inert bacteria which traditional antibiotics fail to kill. This discovery can potentially address one cause of recurring infections and combat antibiotic resistance, a global health threat.

Pastebin.ai is a feature-rich online pastebin where users can share text and code snippets with options for syntax highlighting, protection, and URL monetization. The platform includes over 150 syntax languages, customizable paste settings, and folders for organization.

Kazimir AI is an online platform that offers the ability to fill and validate PDF forms with their InstaFill AI feature, significantly reducing the time and errors associated with manual form entry. The site also hosts AI-generated images of actors, sports figures, and social media celebrities.

Azyri is an AI copilot for healthcare, offering a free bone age assessment tool intended for professionals, students, and AI enthusiasts. Accessible from mobile devices, it provides estimates without the intention of giving medical advice.

Expense Badger is an online service that converts PDF invoices and JPG files into spreadsheets for easier expense management. Users can upload their files, and the service extracts the billing information to create a detailed spreadsheet while renaming the files for better organization.

Interview with AI provides a unique tool that uses artificial intelligence to help individuals prepare for job interviews. It offers a personalized interview preparation roadmap based on the job description input, includes quizzes for self-assessment, and the ability to practice with AI through mock interviews.

Claud is an innovative AI-powered app that serves as a dream journal and interpreter, designed to help users record, understand, and explore their dreams. With a user-friendly interface and advanced AI for dream analysis, Claud offers a seamless experience for dreamers and oneirologists alike.

Everlyn AI offers 24/7 AI Tutors to personalize and enhance the learning experience for students, providing instant support and automated assessments for a more efficient classroom.

ChatSlide is an AI-powered communication platform designed to enhance team collaboration and streamline conversations. With innovative features and a user-friendly interface, ChatSlide is redefining how teams interact and manage their workflow.

This paper analyzes the capability of pre-trained large language models, such as Llama2, GPT-4, and Claude 3, in performing regression tasks without further training or updates. Findings suggest these models can rival traditional supervised methods on complex datasets, scaling well with the number of in-context exemplars.

A TikToker, Brianna Janel, was trapped inside her Tesla Model 3 for approximately 40 minutes during a software update in 115-degree heat. She followed Tesla’s warnings not to open doors or windows during the update to avoid damaging the vehicle. Concerns arose from this incident about the safety of occupants during Tesla's update process.

Artisan AI, a Y Combinator W24 batch company, seeks a Senior Backend Engineer with 3+ years of experience, offering $110K-$150K and equity options. Location is flexible, with options in San Francisco, New York, or Remote. Role involves Python development, API creation, and CI/CD pipeline maintenance.

Research paper 'Increased LLM Vulnerabilities from Fine-tuning and Quantization' examines how downstream tasks such as fine-tuning and quantization affect vulnerabilities in Large Language Models (LLMs), with tests on frameworks like Mistral, Llama, MosaicML. Outcomes show such tasks reduce jailbreak resistance, suggesting the need for enhanced security measures.

Apple is set to update its Mac lineup with M4 chips focused on artificial intelligence capabilities starting in late 2024. The new chips, expected to come in various models including Donan, Brava, and Hidra, will power the iMac, MacBook Pro, Mac mini, MacBook Air, Mac Studio, and Mac Pro through late 2024 and early 2025.

In an interview with Ben Thompson, Google Cloud CEO Thomas Kurian discussed Google’s enterprise AI strategy, highlighting the importance of choosing an open AI platform over individual models. He touched on how integrating AI into products and core processes can transform businesses and gave examples of use cases in various industries. Kurian also mentioned the advancements in Google Cloud's AI and infrastructure offerings, including improvements to Google’s models and the announcement of the million context window for AI.

The article 'You can't build a moat with AI' by Vikram Sreekanti and Joseph E. Gonzalez discusses how differentiating AI applications is a challenge and why simply using LLMs (Large Language Models) isn't enough. The authors argue that real differentiation comes from the data fed into models and stress the importance of data engineering to build a competitive edge in AI applications.

Holodeck is an innovative system created by researchers from the University of Pennsylvania, Stanford University, University of Washington, and Allen Institute for Artificial Intelligence. It automates the generation of 3D embodied AI environments based on language descriptions. Leveraging GPT-4 for commonsense knowledge and a vast collection of 3D assets, Holodeck efficiently tailors environments to specific prompts like style or objects. The system is validated by extensive human evaluations showing preference over manual designs.

STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking) is a system developed by Stanford University researchers. It uses Large Language Models (LLM) to curate knowledge by conducting internet-based research and generating Wikipedia-like articles from scratch. STORM aims to assist in the pre-writing stage of article creation, especially useful for experienced Wikipedia editors.

LANDO AI specializes in creating AI-Generated App Landing Pages aimed at boosting app downloads through stunning designs. Easy start-up process with options to continue with Google or Apple.

Cohere introduces Rerank 3, a new foundation model designed to enhance enterprise search and retrieval systems. Rerank 3 offers capabilities like 4k context length, multilingual coverage, and improved latency. It's compatible with any database and can be integrated into legacy applications. Improved enterprise search, multi-aspect data ranking, code retrieval, and multilingual retrieval are some of its features. It reduces total cost of ownership and is supported in Elastic's Inference API.

The Verge's David Pierce reviews the Humane AI Pin, a screenless wearable aiming to replace smartphones by using AI and a voice-driven interface. For $699 and a $24 monthly fee, it promises efficiency without the distractions of a screen. However, Pierce finds the device unfinished and unreliable, with slow response times, limited functionality, and frequent errors that make it impossible to recommend.

eezyCollab is an affordable, AI-powered influencer marketing tool that simplifies the process of finding and collaborating with influencers. It offers features like AI-powered matching, bulk email campaigns, and reasonable rates. Users can easily describe their product and let the AI handle the matchmaking process to find the best influencers for their marketing campaign.

Google Workspace provides a suite of communication and collaboration tools, including email, calendar, file storage, video meetings, and word processing. It offers solutions tailored for businesses, enterprises, education, and more.

Slidey: presentation maker app is a mobile application designed for creating presentations quickly and easily. It offers pre-made templates and themes with simple slide creation tools, allowing users to make presentations offline on their mobile devices. Users can export presentations to PDF or JPEG formats.

Biliki AI is a travel platform dedicated to helping users plan sustainable trips effortlessly. With a focus on various travel experiences, such as cultural tours, luxury getaways, and family adventures, Biliki AI provides personalized recommendations to create memorable journeys.

Musho is an AI-powered creative assistant that integrates with Figma to help designers create stunning UI designs quickly and efficiently, boosting creativity and productivity.

Eugene Yan shares his experience building an AI coach named Tara to address his monkey mind, imposter syndrome, and the hedonic treadmill. Utilizing ChatGPT, voice interactions, and a platform called Vapi, he creates a 24/7 accessible AI coach for self-reflection, emotional clarity, and preparation for challenging situations.

HyperDoc offers a solution to streamline the sales process by generating AI-powered flashcards that summarize key selling points, helping to clear objections and close deals faster. With features like document summarization, clear insights, and success tracking, HyperDoc enhances productivity and organization for sales teams.

Captions.ai is your AI-powered creative studio for next-generation storytelling. Use AI to create studio-grade videos, scriptwriting, recording, production, visual editing, translation, and distribution. Trusted by 3M+ people worldwide.

Arena Chat is an advanced AI-powered chat platform tailored for e-commerce stores to engage visitors and convert them into customers. It offers features like smart product recommendations, automated customer service, and deep integrations with major e-commerce platforms. Available 24/7, it assists businesses in personalizing shopping experiences and managing buyer interactions with ease.

GoProfiles offers an AI-powered people platform designed to strengthen genuine connections within workplaces. It includes features such as employee directories, peer recognition, org charts, and workforce analytics to foster a positive company culture.

Aider is an AI-based command line tool for pair programming that uses GPT-3.5/GPT-4, allowing developers to edit code within local git repositories. Aider applies changes and commits with descriptive messages and can operate on existing codebases or new projects, utilizing tree-sitter to map out repositories for context-aware assistance.

RecurrentGemma by Google DeepMind is an open-weights language model based on the Griffin architecture, optimized for fast inference in generating sequences. This repository includes the model, examples, and instructions for running and fine-tuning it. With a focus on TPU optimization, it's an Apache-2.0 licensed project open to contributions.

Simon Willison reviews a flurry of activity in the Large Language Model space, including releases like Google Gemini Pro 1.5, OpenAI's GPT-4 Turbo, and Mistral's Mixtral 8x22B. Willison also discusses his work on various Datasette plugins integrated with these models and notes on system prompts and their implementations.

Persana 2.0 is an AI prospecting platform that enhances outbound efforts by leveraging over 75 data sources to find, enrich, and personalize outreach. With features like dynamic data enrichment and lead scoring, Persana is rated #1 in its space for 2024. It helps companies boost revenue, save time, and generate highly qualified leads.

Prelaunch.com offers a platform where users can validate their product ideas by creating a product page with AI assistance, sharing it with a global audience of over 200 million people, and gaining insights through tailored surveys to make informed decisions on whether to pursue their idea.

Infinity AI Studio is a platform offering advanced data analysis and machine learning capabilities, designed to automate processes and enable users to gain insights from their data.

OpenAI's recently unveiled GPT-4 Turbo with Vision model exhibits poorer performance on coding benchmarks compared to its predecessors, achieving a low 62% completion rate on aider's code editing benchmark. It has also been noted for its tendency towards 'lazy coding', where it provides incomplete code fragments with placeholders rather than functional code. Despite this, aider has integrated support for GPT-4 Turbo with Vision.

OpenAI is reportedly gearing up for a series of legal challenges as it faces scrutiny over its products and business practices. The company is said to be preparing for both defense and offense in a legal landscape that is becoming increasingly complex for AI technology providers.