OpenAI's GPT Next will be trained with approximately 100x the compute load of GPT-4. It will be released later this year. The true estimates suggest some of the compute load changes are also due to algorithmic improvements. The article is in Japanese.

OpenAI has launched fine-tuning for GPT-4o, allowing developers to customize the model for specific use cases with their own datasets. It is offering 1 million free training tokens per day through September 23.

All future versions of Redis will be released with source-available licenses starting with Redis 7.4. Redis will no longer be distributed under the three-clause Berkeley Software Distribution and will be instead dual-licensed under the Redis Source Available License and Server Side Public License. The new licenses enable Redis to sustainably provide permissive use of its source code, allowing it to continue to be freely available to developers, customers, and partners through Redis Community Edition as Redis moves on to its next phase of development as a real-time data platform with a unified set of clients, tools, and core product offerings.

Neuralink has shared a video of its brain-computer interface in action inside a human patient. The patient, a 29-year-old man named Nolan Arbaugh who is paralyzed from the neck down, is able to control a cursor on a screen using Neuralink's device. He is able to play games online for about eight hours before the implant needs to recharge. Arbaugh reports that his experience with the implant has so far been positive, despite some initial issues. The video is available in the article.

Scientists from the University of Amsterdam claim they have successfully eliminated HIV from infected cells using CRISPR gene-editing technology. The work remains a proof-of-concept and will not become a cure for the disease any time soon. The study's findings are still being scrutinized.

OpenAI is expected to release a major AI model, possibly GPT-5, sometime in mid-2024, likely during the summer. The new model will likely be a multimodel large language model with similar capabilities to GPT-4, but better. OpenAI is still reportedly training the model, after which it will go through internal safety testing to identify any issues before public release. Other issues, besides those that may come up from testing, may delay the launch.

GPT-5 will be significantly more intelligent than GPT-4. It is still at least a year and a half from release. The model will have advanced memory and reasoning capabilities and Ph.D.-level intelligence in specific tasks. A short video of OpenAI's Chief Technology Officer Mira Murati comparing the different GPTs is available in the article.

OpenAI has announced a new model called GPT-4o (o is for omni) that is a natively multimodal model, with superior performance to GPT-4 on text and state-of-the-art performance on a variety of modalities. It also announced a new desktop app, a near real-time audio interface, and a variety of improved reasoning features.

OpenAI has released GPT-4o, an AI model that can reason across audio, vision, and text in real time. Developers can access the GPT-4o API as a text and vision model. It's 2x faster, half the price, and has 5x higher rate limits compared to GPT-4 Turbo.

GPT-4's dominance in AI benchmarks has been challenged by four new models from different vendors, each showing the potential to surpass GPT-4's capabilities. However, concerns arise as, amidst growing legal and ethical considerations, none of these models are open-source or transparent about their training data. The push for models trained on public domain or licensed content continues, highlighting the complexity of creating competitive AI without proprietary data.

OpenAI announced a new AI model yesterday called GPT-4o that can converse using speech in real time, read emotional cues, and respond to visual input. It will roll out over the next few weeks for free to ChatGPT users and as a service through API. Paid subscribers will have five times the rate limits of free users. The API will feature twice the speed, 50% lower cost, and five times higher rate limits compared to GPT-4 Turbo. A 26-minute long video that introduces GPT-4o and demonstrates its abilities is available in the article.

OpenAI has released o1 and o1-mini, the first in a series of reasoning models that have been trained to answer more complex questions faster than a human can. The model is better at writing code and solving multistep problems than previous models, but it is more expensive for developers and slower to use than GPT-4o. The release is still in preview to indicate how nascent it is. ChatGPT Plus and Team users should already have access to the model, while Enterprise and Edu users will get access early next week. OpenAI plans to bring o1-mini access to all free users, but it hasn't set a release date yet.

OpenAI has launched a new AI model called GPT-4o mini. It is available to free users. The 'o' in the name stands for 'omni', hinting at the model's audio, video, and text capabilities. The model will be made available to ChatGPT Enterprise users next week.

OpenAI appears to have accidentally published a blog post that was indexed by Bing and DuckDuckGo before it was quickly taken down. The post was an announcement for GPT-4.5 Turbo, a new model that surpasses GPT-4 Turbo in speed, accuracy, and scalability. The cached description mentions a knowledge cutoff date of June 2024 and a 256k context window. OpenAI has not commented on the leak.

OpenAI's GPT-4o and GPT-4o mini can now be fine-tuned and customized by developers for business use. Developers can now use their own datasets to enhance the model's knowledge base with proprietary information and control how the model responds to specific questions. It costs $25 for every 1 million tokens used to fine-tune GPT-4o - $3 for GPT-4o mini. 1 million tokens is roughly equivalent to 2,500 pages in a standard-size book.

OpenAI anticipated spending around $4 billion on Microsoft's Azure servers for ChatGPT inference in 2023, potentially causing significant financial losses. Although OpenAI is profiting approximately $2 billion annually from ChatGPT, it may require additional funding within a year to address a projected $5 billion shortfall. It currently utilizes the equivalent of 350,000 Nvidia A100 chip servers, mainly for ChatGPT, with discounted rates from Azure.

xAI's next model is slated for August. It reportedly took 20k H100s to train. Grok 3 is rumored for the end of the year and potentially requires 100k H100s.

OpenAI and Google have introduced advanced AI models that enable real-time multimodal understanding and responses and promise improved AI assistants and innovations in voice agents. OpenAI's GPT-4o boasts double the speed and half the cost of its predecessor, while Google's Gemini 1.5 Flash delivers a significant reduction in latency and cost. Both tech giants are integrating AI across their ecosystems, with OpenAI eyeing consumer markets, which could potentially reach up to a billion users, with its products and partnerships.

OpenAI and Meta are teasing the next iterations of their AI models, expected to feature enhanced reasoning and planning capabilities. Dubbed GPT-5 and Llama 3, the models aim to advance toward artificial general intelligence, with vague release timelines and application details. The tech community remains skeptical given the history of overhyped AI promises with limited substantive evidence.

OpenAI has announced a live stream event scheduled for May 13 to present updates related to ChatGPT and GPT-4, possibly including the launch of an AI-powered search engine.

There are 3 prominent models in the AI landscape: GPT-4, Claude 3 Opus, and Gemini Advanced. Each has distinct characteristics and capabilities, catering to different needs such as coding, writing, or streamlining basic tasks. This article delves into concepts like context windows and agents, highlighting how these features enhance AI systems capabilities. Even though GPT-4 remains the benchmark and standard, new emerging features in other models could make them solid alternatives in the near future depending on the specific use case.

GPT-4o Mini is a new, smaller, model from OpenAI aimed to replace GPT-3.5. It scores 82 on MMLU, which is reasonable for cheap models.

OpenAI has trained a model based on GPT-4 that can catch GPT-4 mistakes and help humans spot mistakes during the RLHF process.

GPT-4o, OpenAI's latest AI model, bridges real-time communication between humans and machines, extending capabilities beyond text to include vision and audio. The AI revolution introduces a new wave of human-to-AI and eventual AI-to-AI interactions, likely impacting the dynamics of our social behaviors and business models. As this technology progresses, its effect on human communication will unfold, potentially catalyzing the creation of innovative companies and software solutions.

In this article, OpenAI's Evan Morikawa provides insights into ChatGPT's inner workings, covering input text processing and tokenization to prediction sampling using large language models. ChatGPT operates by turning tokens into numerical vectors (embeddings), multiplying them by a weight matrix of billions, and selecting the most probable next word. The tech is grounded in extensive pretraining to predict text based on vast internet data.

Google's new AI chip, Cloud TPU v5p, is now available. It boasts nearly triple the training speed for large language models compared to its predecessor, TPU v4. This release underscores Google's position in the AI hardware race alongside competitors like Nvidia. Google has also introduced the Google Axion CPU, based on Arm's chip infrastructure, promising better performance and energy efficiency.

Researchers have demonstrated that OpenAI's GPT-4 model can autonomously exploit security vulnerabilities detailed in CVE advisories with an 87% success rate, far outperforming other models and tools like vulnerability scanners.

OpenAI's ChatGPT has varied performance in code generation, with success rates ranging from less than 1% to 89% depending on factors like task difficulty and programming language.

xAI is planning to build a supercomputer to power the next version of its Grok AI chatbot. Elon Musk says he wants to get the proposed supercomputer running by the fall of 2025. xAI could be partnering with Oracle for the project. When completed, the supercomputer will be at least four times the size of the biggest GPU clusters that exist today. Musk estimates that running the Grok 3 model and beyond will require 100,000 Nvidia H100 chips.

Despite GPT-4o's advanced imaging capabilities, OpenAI is still actively enhancing DALL-E 3, focusing on refining text rendering and visual accuracy. Facing stiff competition from Midjourney and Ideogram, OpenAI's strategy underscores the continuous evolution and challenges in AI-driven visual technologies.