The Generative AI Market Roundup: Fine-Tuning on GPT-3.5 Turbo, Code Llama, Lilli and $383 Million

The Generative AI market is moving at lightspeed, from Generative AI startup funding rounds, LLM advances and releases, and Generative AI startup exits. The market is dominating the tech world and we have you covered with a breakdown of the top headlines over the past two weeks.

Fine-Tuning LLMs for the Enterprise, Courtesy of OpenAI

First and foremost, OpenAI is offering new built-in support for enterprises to fine-tune its GPT-3.5 Turbo Large Language Model (LLM), allowing enterprises to train on proprietary data and deploy it at scale. A true defining moment for the Generative AI market. The fine-tuning process involves three main steps: preparing the data, uploading the files, and creating a fine-tuning job, with safety features maintained through OpenAI's Moderation API. The startup plans to extend fine-tuning capabilities to its flagship GPT-4 model later this fall and will also launch a fine-tuning interface to simplify the process. Data labelling startup, Scale AI, was named its "preferred partner" for fine-tuning its GPT-3.5 Turbo LLM.

Nvidia's Momentum Doesn't Slow

GPU king, Nvidia, reported record Q2 revenue of $13.5 billion for 2024, exceeding Wall Street expectations and marking a 101% YoY increase. The company's data center segment was a significant driver, accounting for $10.3 billion (76% of total revenue), and growing 171% annually. Nvidia is reinvesting its increased profits, which saw a gross margin rise to 71%, into rapid product development, including its Hopper architecture, to maintain its lead in AI and LLM technologies. The company plans to launch a new product or platform every six months and reported $2 billion in R&D spending for Q2 FY2024.

Text-to-Image Generator for Fonts with Ideogram

Earlier this week, Ideogram, a new AI image generator startup founded by former Google Brain researchers, launched with $16.5 million in Seed funding from VCs a16z and Index Ventures. The Canadian startup sets itself apart by solving a long-standing issue in AI image generation: reliable text generation within images, making it particularly appealing for tasks like creating company logos or signage. Dalle and Midjourney don't fare well, or at all, when prompted for text/font creation. While the platform offers a variety of preset styles, including 3D rendering and cinematic effects, it shines in its typography capabilities.

Tack Another $350 Million to Anthropic and Hugging Face

AI startup Anthropic, developer of its LLM Claude, has secured an additional $100 million in funding from South Korean telecom giant SK Telecom (SKT). Claude differs from GPT models in that is allows for a 100,000 token limit, compared to the 32,000 and 48,000 token limits of GPT's models. The strategic investment aims to develop a custom LLMs tailored for the telecom industry's specific needs, supporting multiple languages including Korean, English, German, Japanese, Arabic, and Spanish. Another angle that Anthropic distinguishes itself by focusing on "constitutional AI," which adheres to ethical guidelines to avoid toxic or discriminatory outputs. The custom LLM will be integrated into a Telco AI Platform being developed by the Global Telco AI Alliance, aiming to serve as the foundation for new AI services in the telecom industry.

Hugging Face secured $235 million for its Series D funding round led by Salesforce with participation from Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM, and Sound Ventures. Their valuation now stands at $4.5 billion. Hugging Face plans to use the funds to expand its team and invest in open-source AI and collaboration platforms. The startup is integral to the AI community, boasting over a million repositories, including 500,000 models and 250,000 datasets.

Generative AI Acquisitions: OpenAI, Aivo and SpeakGPT

OpenAI acquired New York-based startup Global Illumination, founded by former Facebook and Instagram employees. The startup has experience in leveraging AI for creative tools and digital experiences, and its team has also contributed to other major companies like YouTube, Google, Pixar, and Riot Games. While the financial terms of the deal weren't disclosed, OpenAI plans to integrate Global Illumination's talent into its core products, including ChatGPT. The acquisition suggests that OpenAI may be looking to expand into more multimedia features, augmenting it's text-to-image creation features natively in ChatGPT's web application, and potentially adding text-to-video generation.

Generative AI Solutions Corp purchased the intellectual property assets, including source code and a fully functional prototype, of SpeakGPT for only $60K. It's an AI-powered virtual assistant mobile application from Metachain Technologies.

Engageware acquired conversational AI startup Aivo to bolster its cloud-based customer engagement platform. The acquisition aims to automate customer service, reduce call center workloads, and improve customer experience across various industries.

AI Writing Assistants: Lex and Writerly AI

Writerly AI raised $2 million to further develop and expand its latest product, EKOM, a digital asset creation and optimization engine for e-commerce. EKOM automates the process of creating and optimizing product assets for SEO, aiming to enhance sales and buyer acquisition channels for e-commerce businesses.

Lex, an AI-powered writing platform, raised $2.75 million in a seed round led by True Ventures. The tool aims to help writers overcome blocks by offering AI-generated suggestions when the user slows down or halts in their writing process. While the tool aims to integrate AI into the writing process, it also places a strong emphasis on user privacy, stating that it does not use user content for training its AI models. With strong initial interest and a user base of 25,000 in its first 24 hours, Lex plans to introduce a paid tier but intends to keep its team small to manage costs effectively. Their website crashed a few minutes are the announcement.

Meta Releases Code Llama to Compete with Codex

With great reception from the developer community, Meta Platforms launched Code Llama, an open-source LLM specifically designed for programming tasks, positioning it as a major competitor to OpenAI's Codex and other coding-specific LLMs. Code Llama aims to assist software engineers across various sectors and can generate code, complete code, create developer notes, and assist in debugging. It supports multiple programming languages including Python, C++, and Java. The model family includes versions with 7-billion, 13-billion, and 34-billion parameters, each designed for different computational needs, and they allow for up to 100,000 tokens in their prompts for more context-relevant code generation.

Mckinsey Debuts Generative AI Tool

After announcing their efforts in building Generative AI for the enterprise with Quantum Black, the consulting giant's plans finally culminate. McKinsey and Company developed its own Generative AI tool called Lilli, designed to assist its employees by providing information, insights, data, and expert recommendations for consulting projects. Created by McKinsey's "ClienTech" team, Lilli has been in beta since June 2023 and is set to roll out across the company this fall. The tool has already been used by approximately 7,000 employees and has significantly reduced the time spent on research and planning. Lilli's interface allows users to ask questions and receive answers sourced from McKinsey's extensive corpus of documents and transcripts, as well as from LLMs. The tool also cites its sources, providing full attribution for the information it delivers. McKinsey is considering the possibility of making Lilli available to external clients in the future.

IBM Announces COBOL Upgrade and Partners with Microsoft

On the heels of a number of research surveys about enterprise usage of Generative AI, IBM Consulting is partnering with Microsoft to accelerate the adoption of Generative AI for businesses. The collaboration will focus on implementing and scaling Azure OpenAI Service, a managed AI service that allows the application of LLMs for various business use cases, including automating procurement processes, content summarization, and streamlining healthcare operations. The partnership aims to provide enterprise customers with the expertise and technology needed to innovate their business processes and effectively scale Generative AI solutions.

Last week, they announced a new initiative to modernize COBOL applications using its Generative AI LLMs called watsonx. The watsonx code assistant for Z service aims to help clients migrate their COBOL applications to more modern Java application code, addressing the talent gap in COBOL and reducing risk. The assistant can analyze, refactor, and transform COBOL code into Java, generating automated tests to ensure the new Java code is semantically equivalent to the original COBOL.

Text-to-Speech Startup Eleven Labs Adds Multi-Lingual Model

ElevenLabs, a startup specializing in voice cloning and synthesis, has launched a new text-to-speech model that supports 30 languages, exiting from its beta phase. The startup's platform allows users to clone their own voices or use pre-made AI voices to convert text into speech, targeting diverse markets and languages. With applications ranging from voicing video game characters to customer service avatars and audiobooks, ElevenLabs aims to break down linguistic barriers and make content universally accessible.

Generative AI Funding Rounds

Ikigai Labs raised $25 million in Series A funding to expand its no-code platform that utilizes proprietary graphical models for Generative AI applications in tabular data. The platform aims to assist enterprises in efficiently handling tabular data for tasks such as forecasting and scenario planning. Led by Premji Invest, the funding will help Ikigai grow its team and compete in the enterprise AI market, specifically in the realm of Generative AI for tabular data.

Irreverent Labs, a startup specializing in Generative AI for video creation, secured an undisclosed amount of funding in a round led by Samsung Next. Previously backed by Andreessen Horowitz with $45 million in funding, the company initially focused on blockchain and gaming but has pivoted to enable users to generate various types of short videos, including 3D. The funding will help Irreverent Labs collaborate with Samsung's device units for broader distribution and offer its API to Samsung Next portfolio companies. The startup aims to democratize short-form video production, targeting both individual content creators and enterprises.

Portkey.ai, a San Francisco-based startup specializing in Generative AI app-building tools, raised $3 million in a Seed funding round led by Lightspeed with participation from angel investors from AWS, OpenAI, Cloudflare, Postman, and Asana. Founded in January 2023 by Rohit Agarwal and Ayush Garg, the company provides tools for monitoring large language model operations (LLMOps), connecting to multiple LLMs, and managing prompts. The funding will be used to help engineering teams build and launch Generative AI apps more quickly. Portkey.ai already serves Generative AI companies like Postman, Jio Haptik, and Springworks through its full-stack LLMOps solution.