IBL News

Author: IBL News

  • Training a Custom Model with OpenAI’s GPT-4 Takes $2-$3 Million

    Training a Custom Model with OpenAI’s GPT-4 Takes $2-$3 Million

    IBL News | New York

    Training a custom model from scratch using OpenAI’s GPT-4 may take several months, with pricing starting at $2 to $3 million, according to the company.

    This high price sparked a discussion among practitioners on Twitter, now known as X. Many users agreed that a much smaller pre-trained base model with fine-tuning on top of it would cost ten times less.

    OpenAI justifies the price by stating:

    “The Custom Models program gives selected organizations an opportunity to work with a dedicated group of OpenAI researchers to train custom GPT-4 models to their specific domain.”

    “This includes modifying every step of the model training process, from doing additional domain-specific pre-training to running a custom RL post-training process tailored for the specific domain.”

    “Organizations will have exclusive access to their custom models. This program is particularly applicable to domains with extremely large proprietary datasets—billions of tokens at minimum.”

    On the other hand, OpenAI announced Data Partnerships, an initiative intended to work with organizations to produce public and private datasets for training AI models as a way to combat models that contain toxic language and biases.

    To work with data and PDFs in those large-scale datasets, OpenAI says that it uses world-class OCR technology and automatic speech recognition (ASR) to transcribe spoken words.
    .

     

  • Grok, Elon Musk’s Chatbot, Tries to Raise $1 Billion

    Grok, Elon Musk’s Chatbot, Tries to Raise $1 Billion

    IBL News | New York

    X.AI, the AI start-up launched by Elon Musk in July, filed with the SEC to raise up to $1 billion in an equity offering.

    The company, which released a chatbot called Grok, already brought in $135 million from investors.

    “Grok is designed to answer questions with a bit of wit and has a rebellious streak, so please don’t use it if you hate humor!” X.AI wrote on its website, adding, “It will also answer spicy questions that are rejected by most other AI systems.”

    Grok, X.AI has real-time knowledge of the internet, including access to all of X, giving it a leg up on all other chatbots.

    It aims to compete directly with ChatGPT, Bard, and Claude. However, Musk said that Grok has been designed to be anti-woke and lacks the political correctness built into other chatbots.

    Grok started rolling out to X Premium Plus users this month at $16 a month, per month.

    “We are a separate company from X Corp, but will work closely with X (Twitter), Tesla, and other companies to make progress towards our mission,” X.AI says on its website.

    People working on X.AI include alumni of DeepMind, OpenAI, Google Research, Microsoft Research, Twitter, and Tesla.
    .

  • Mistral, a French AI Start-Up, Was Valued at $2 Billion in Funding Round

    Mistral, a French AI Start-Up, Was Valued at $2 Billion in Funding Round

    IBL News | New York

    Mistral AI, the 22-people French start-up founded seven months ago by researchers from Google and Meta, raised $415 million (or 385 million euros) at a valuation of $2 billion.

    Investors include the Silicon Valley venture capital firms Andreessen Horowitz and Lightspeed Venture Partners.

    Also, this month, Mistral opened beta access to its platform services so people can build their own chatbots.

    The start-up also released Mixtral 8x7B, a high-quality sparse mixture of expert model (SMoE) with open weights, handling a context of 32k tokens, and licensed under Apache 2.0.

    It handles English, French, Italian, German and Spanish and shows strong performance in code generation.

    Mistral AI stated:

    “Mixtral — pre-trained on data extracted from the open Web — outperforms Llama 2 70B on most benchmarks with 6x faster inference. It is the strongest open-weight model with a permissive license and the best model overall regarding cost/performance trade-offs. In particular, it matches or outperforms GPT3.5 on most standard benchmarks.”

    Rivals like OpenAI and Google said that Mistral, which releases its technology as open-source software, can be dangerous, arguing that the raw technology could be used to spread disinformation and harmful material.

    The Mistral platform serves three endpoints: Mistral-tiny, which serves Mistral 7B Instruct v0.2; Mistral-small (Mixtral 8x7B); and Mistral-medium.

    The French Finance Minister pointed to Mistral as providing the European continent a chance to challenge U.S. tech giants.

    In the sphere of open-source, the American company Meta has been at the forefront. This year, it released an LLM called LLaMA.
    .

  • Google Makes ‘Gemini Pro’ Available to Developers and Enterprises

    Google Makes ‘Gemini Pro’ Available to Developers and Enterprises

    IBL News | New York

    Google announced yesterday that it was making its new AI model, Gemini Pro, available to developers and enterprises. It has been released via the Gemini API, and it’s free for now. SDKs are also available for Gemini Pro to help to build apps.

    Last week, Google started to roll out Gemini, which comes in three sizes: Ultra, Pro, and Nano.

    “The faster way to build with Gemini is with Google AI Studio, a free, web-based developer tool that enables you users to develop prompts and then get an API key to use in their app development,” said the company.

    Google intends that users migrate from AI Studio to Vertex AI. Early next year, the company plans to launch Gemini Ultra, its largest and most capable model for complex tasks.

    Yesterday, Google announced that Duet AI for Developers will be launched in Q1 of 2024. It will compete with GitHub GPT-based Copilot.

    Google’s partners adding support for coding with Duet AI for Developers include Confluent, Elastic, Grafana Labs, Hashicopr, MongoDB, Neo4j, Pinecone, Redis, and SingleStore.

    In addition, Google Cloud’s image-generation capabilities have been upgraded with Imagen 2. This feature is now generally available for Vertex AI customers on the allowlist.
    .

     

     

  • IBM, Meta, and 40 Top Organizations Create the ‘AI Alliance’ to Develop Open Source Technology

    IBM, Meta, and 40 Top Organizations Create the ‘AI Alliance’ to Develop Open Source Technology

    IBL News | New York

    IBM, Meta, and over 40 companies and organizations are creating the AI Alliance, an industry and academic group dedicated to open-source AI technology.

    This coalition said in a statement that it will focus on the responsible development of AI technology, including safety and security tools.

    AI Alliance will look to increase the number of open-source models rather than the proprietary systems favored by companies such as OpenAI, Google, Anthropic, and Microsoft.

    It will also develop new hardware and team up with academic researchers.

    The AI Alliance consists of a broad range of organizations that are working across aspects of AI education, research, development and deployment, and governance.

    Among them, there are creators of some of today’s most used open models including Llama2, Stable Diffusion, StarCoder, Bloom, as well as application developers like MLPerf, Hugging Face, LangChain, LlamaIndex, and other open-source AI toolkits will participate.

    Frameworks that drive platform software, including PyTorch, Transformers, Diffusers, Kubernetes, Ray, Hugging Face Text generation inference, and Parameter Efficient Fine Tuning, joined the initiative.

    Partners and collaborators include:

    • Agency for Science, Technology and Research (A*STAR)
    • Aitomatic
    • AMD
    • Anyscale
    • Cerebras
    • CERN
    • Cleveland Clinic
    • Cornell University
    • Dartmouth
    • Dell Technologies
    • Ecole Polytechnique Federale de Lausanne
    • ETH Zurich
    • Fast.ai
    • Fenrir, Inc.
    • FPT Software
    • Hebrew University of Jerusalem
    • Hugging Face
    • IBM
    • Abdus Salam International Centre for Theoretical Physics (ICTP)
    • Imperial College London
    • Indian Institute of Technology Bombay
    • Institute for Computer Science, Artificial Intelligence
    • Intel
    • Keio University
    • LangChain
    • LlamaIndex
    • Linux Foundation
    • Mass Open Cloud (MOC) Alliance, operated by Boston University and Harvard
    • Meta
    • Mohamed bin Zayed University of Artificial Intelligence
    • MLCommons
    • National Aeronautics and Space Administration (NASA)
    • National Science Foundation (NSF)
    • New York University
    • NumFOCUS
    • OpenTeams
    • Oracle
    • Partnership on AI
    • Quansight
    • Red Hat
    • Rensselaer Polytechnic Institute
    • Roadzen
    • Sakana AI
    • SB Intuitions
    • ServiceNow
    • Silo AI
    • Simons Foundation
    • Sony Group
    • Stability AI
    • Together AI
    • TU Munich
    • UC Berkeley College of Computing, Data Science, and Society
    • University of Illinois Urbana-Champaign
    • The University of Notre Dame
    • The University of Texas at Austin
    • The University of Tokyo
    • Yale University

    In addition to bringing together leading developers, scientists, academics, students, and business leaders in the field of artificial intelligence, the AI Alliance will plan to partner with initiatives from governments, non-profit and civil society organizations working in the AI space.



  • Google’s Promotional Video of Gemini Was Fabricated

    Google’s Promotional Video of Gemini Was Fabricated

    IBL News | New York

    Google’s main Gemini AI demo video released — see it below — last week was fabricated, misrepresenting the model’s capabilities. The company admitted it in a report issued by Bloomberg.

    “Google’s video made it look like you could show different things to Gemini Ultra in real time and talk to it. You can’t,” Bloomberg’s reported wrote in a tweet.

    Google researchers interacted with Gemini through text, not voice, then picked the best interactions and edited them together with voice synthesis to make the video.

    As a result, the Internet giant is now facing controversy among AI experts.

    Running still images and text through LLM is computationally intensive and makes real-time video largely impractical. That was one of the clues that led experts to detect that the video was misleading.

    As Ars Technica reports, “OpenAI has embarrassed Google in generative AI over the last year.”

    On December 6th, Gemini was introduced as the first true rival to OpenAI’s GPT-4.

    The stock of the company went up 5% amidst comments about Google’s model “sophisticated reasoning capabilities,” supported by multiple benchmarks.

    The contested video, titled “Hands-on with Gemini: Interacting with multimodal AI,” shows that the AI model sees, accompanied by the AI model’s responses on the right side of the screen. The viewer hears a voice, apparently of Gemini Ultra, responding to the questions.

    Additionally, on the next day, TechCrunch reported that users were very disappointed with Gemini Pro’s performance on Bard.

    ArsTechnica‘s tests showed that Google’s AI system improved in the eight months but “ChatGPT is still the winner”.
    .

  • What Happened at OpenAI: Board Members Believed that Altman Was Dishonest

    What Happened at OpenAI: Board Members Believed that Altman Was Dishonest

    IBL News | New York

    Before firing Sam Altman on November 17, Ilya Sutskever, the chief scientist of OpenAI, and three board members had been whispering behind his back for months, The New York Times revealed in an insightful story this Saturday.

    They believed Sam Altman had been dishonest and should no longer lead the company.

    They were worried that ChatGPT’s success was antithetical to creating safe AI.

    In September, Altman met investors in the Middle East to discuss an AI chip project. The board was concerned that he wasn’t sharing all his plans with it.

    He also believed that Mr. Altman was bad-mouthing the board to OpenAI executives. Other employees have also complained to the board about Mr. Altman’s behavior.

    The ouster was the culmination of years of tensions and divisions at OpenAI.

    Microsoft, which had committed $13 billion to OpenAI, weighed in to protect its investment. Many top Silicon Valley executives and investors, including the CEO of Airbnb, also mobilized to support Altman.

    From his $27 million mansion in San Francisco’s Russian Hill neighborhood, Sam Altman, driven by a hunger for power more than by money, lobbied through social media and voice their displeasure in private text threads, according to the NYT, who interviewed more than 25 people with knowledge of the events.

    On Nov. 21, Sam Altman returned as CEO of OpenAI.

    The San Francisco OpenAI lab was founded by Elon Musk, Sam Altman, Ilya Sutskever, and nine others. Its goal was to build AI systems to benefit all of humanity.

    Unlike most tech start-ups, it was established as a nonprofit with a board that was responsible for making sure it fulfilled that mission.

  • Quora Creates a Revenue Program for Chatbot Creators

    Quora Creates a Revenue Program for Chatbot Creators

    IBL News | New York

    Quora, the company behind the Poe.com chatbot, said it intends to create a new market allowing bot creators in areas such as tutoring, knowledge, therapy, entertainment, analysis, storytelling, roleplay, image, video, and music to generate revenue.

    With this stated goal, this month, Poe.com launched this month a Creator monetization program to support both prompt bots and server bots created by developers who write code and integrate with the platform API.

    “For anyone training their own models and running their own inference, or using a third-party AI service through an API, operating a bot can entail significant infrastructure costs, and we want them to be able to operate sustainably and profitably,” said Adam D’Angelo, CEO at Quora, the company behind Poe.com.

    Poe.com’s monetization structure, currently available only in the US, has two components:

    • If a bot causes a user to subscribe to Poe (measured a few ways), the company will share a cut of the revenue they pay.
    • The bot generator can set a per-message fee and Poe will pay that on every message.

    Users can get started at poe.com/creators, or learn about creating a bot at developer.poe.com.

     

     

  • The ChatGPT App Gets Nearly $30 Million in Revenue

    The ChatGPT App Gets Nearly $30 Million in Revenue

    IBL News | New York

    The ChatGPT app topped 110 million installs, and its paid version of $20 per month reached $28.6 million in revenue, according to TechCrunch and data.ai.

    Competitor AI chatbot Anthropic’s Claude had 1.18 million monthly active users as of September 2023, according to Apptoppia.

    The top chatbot by revenue was Ask AI, which ranges from one week up to one year, as well as multiple tiers, like “premium” and “elite.”

    Other successful generative AI apps with lots of downloads, as shown in the graphics, are Character AI, Chai, Open Chat, Nova, ChatBot, AI Mirror, Imagine, Artimind, and ChatBox.
    .

  • Google Introduced Its Multimodal Technology ‘Gemini’ and Added It to Bard

    Google Introduced Its Multimodal Technology ‘Gemini’ and Added It to Bard

    IBL News | New York

    Google introduced yesterday its long-awaited answer to ChatGPT, a multimodal, natively designed, and pre-trained AI technology with reasoning capabilities named Gemini.  

    While other multimodal offerings — meaning it can analyze text, audio, video, images, and code —  exist, Gemini was described by Google’s CEO Sundar Pichai as the company’s “most capable and general model yet.”

    “Our first version, Gemini 1.0, is optimized for different sizes: Ultra, Pro, and Nano.”

    Demis Hassabis, CEO and Co-Founder of Google DeepMind, explained that “Gemini Ultra’s performance exceeds current state-of-the-art results on 30 of the 32 widely-used academic benchmarks used in large language model (LLM) research and development.”

    “This makes it especially good at explaining reasoning in complex subjects like math and physics.”

    Gemini can understand, explain, and generate high-quality code in Python, Java, C++, and Go. “Its ability to work across languages and reason about complex information makes it one of the leading foundation models for coding in the world,” said Demis Hassabis.

    Google said that Gemini 1.0 was now rolling out across a range of its products and platforms.

    For example, the chatbot Bard was upgraded with Gemini Pro, while Gemini Ultra will applied early next year in a new experience called Bard Advanced.

    Google was also bringing Gemini to Pixel. Pixel 8 Pro will be engineered to run Gemini Nano, powering new features like Summarize in the Recorder app and rolling out in Smart Reply in Gboard, starting with WhatsApp.

    In the coming months, Gemini will be available in more of our products and services like Search, Ads, Chrome, and Duet AI.

    Starting on December 13, developers and enterprise customers will be able to access Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI.

    (Google AI Studio is a free, web-based developer tool to prototype and launch apps quickly with an API key.)
    .

    A chart showing Gemini Ultra’s performance on common text benchmarks, compared to GPT-4 (API numbers calculated where reported numbers were missing).

    A chart showing Gemini Ultra’s performance on multimodal benchmarks compared to GPT-4V, with previous SOTA models listed in places where capabilities are not supported in GPT-4V.

     

    • The New York Times: Google Updates Bard Chatbot With ‘Gemini’ A.I. as It Chases ChatGPT