Welcometo theNeuralAI is moving fast. We’ll help you keep up.OpenAI officially announced GPT-4.5 as the latest version of ChatGPT, the fastest model. OpenAI stresses that GPT 4.5 excels in many areas, including accuracy and emotional intelligence. However, GPT 4.5 is not the most advanced model at this time.
OpenAI is aware that its models and names can be confusing.
GPT-4.5 has become smarter, warmer and not yet available to everyone. GPT-4.5 won’t help. OpenAI claims that despite GPT-4.5 being the newest release, its o3 reasoning remains its most powerful. It describes GPT 4.5 as a research model and not a new frontier.
OpenAI expects GPT-5 to be released later this year as a single model that can handle both quick chat requests and reasoning requests.
GPT-4.5 will replace GPT-4o in the near future with greater accuracy, less hallucinations and a higher emotional intelligence. Anthropic’s Claude, a chatbot that is more natural than ChatGPT, has been praised for this last quality.
GPT-4.5, for example, may invite users to continue the conversation rather than just provide detailed information. The new model is more concise and conversational. Openaisays that.
GPT-4.5 will be available to Pro subscribers who pay $200/month starting today. The new model will be available to more users next year. OpenAI will roll out GPT-4.5 next week to ChatGPT Plus ($20/month), Team subscribers, and Enterprise and Edu users.
Watch a 13-minute livestream below:
OpenAI’s 30 page system card for GPT-4.5 is available. Here
Beyond Openai
While OpenAI is still advancing, former key players who left the company are launching new startups to raise money and build new AI tools. Mira Murati, former Chief technology officer at OpenAI and founder of Thinking Machines Lab, announced her AI startup,Thinking Machines Lab,
recently. Murati has brought in more than 30 experts, including OpenAI, Meta and Google DeepMind.
Lilian Weng, co-founder of OpenAI, was previously Vice President for AI safety and robots as well as applied Research. John Schulman, a co-founder of OpenAI, joined Thinking Machines after working as a researcher at Anthropic.
Thinking Machines Lab focuses on multimodal AI products and models that are accessible, with practical applications.
This is how Thinking Machines Lab works The focus of is as follows:
AI capabilities have advanced dramatically but key gaps remain. The scientific community’s knowledge of frontier AI systems is behind rapidly evolving capabilities. The top research labs have a monopoly on the knowledge of how these systems are taught, which limits both the public’s understanding of AI and their ability to use AI effectively. These systems are difficult to customize for individual needs and values, despite all their potential. To bridge these gaps, we are building Thinking Machines Lab in order to make AI systems better understood, customizable, and more capable. […]
The emphasis is on human-AI collaborative work. We are excited to create multimodal systems that collaborate with people, rather than focusing on fully autonomous AI systems.
AI systems that are more flexible, adaptable and personalized. We see AI as having enormous potential to help in all fields of work. We’re building AI to adapt to a wider range of human expertise.
According to reports, the company is looking to raise $1 billion with a valuation of $9 billion. Business Insider
Ilya Sutskever, former OpenAI Chief Scientist, co-founded Safe Superintelligence with former OpenAI researcher Daniel Levy, and former Apple AI Projects Lead Daniel Gross.
This is how SSI works explains the purpose of its existence:
The most important technical challenge of our time is to build safe superintelligence.
The world’s first straight shot SSI laboratory has been established with a single goal and product: a superintelligence. Safe Superintelligence Inc. is the name of our company.
SSI, our mission, is also our name and our entire product road map. Our team, investors, business model, and all are aligned with SSI.
Safety and capabilities are treated as technical problems that can be solved by revolutionary engineering and scientific breakthroughs. We will advance capabilities as quickly as possible, while ensuring that safety is always first.
Another AI startup that you should be aware of is Eureka Labs,which is creating an AI native educational platform. Andrej Karpathy, a former OpenAI researcher and founding member has been building Eureka Labs in 2024. He is currently the director of AI for Tesla. The startup’s
is the following: announcedits mission and first product-in-development in July 2024:
We are Eureka Labs and we are building a new kind of school that is AI native.
What is the ideal learning experience? In the case of physics, one could imagine working with Feynman to work through high-quality course materials. He would be there to guide you at every step. Unfortunately, subject experts who are passionate, excellent at teaching, infinitely tolerant, and fluent in every language of the world are very rare and cannot tutor us all on demand.
Recent advances in generative AI have made this learning experience more tractable. The teacher designs the course materials but an AI Teaching Assistant is able to support, leverage and scale them. This Teacher + AI symbiosis can run an entire course curriculum on a single platform. If we succeed, it will be simple for anyone to learn something, expanding education both in reach (a large group of people learning something), and extent (anyone learning a wide range of subjects, beyond what is possible today without assistance).
The world’s best AI course will be our first product. LLM101n. This is a class for undergraduates that guides students through the process of training their own AI. It’s similar to a smaller version AI Teaching Assistant. The course materials will be online, but we plan to run digital and physical cohorts that go through it together.
We are currently working on LLM101n but we are looking forward to a world where AI will be a key technology in the future. Eureka Labs has more to come. Karpathy shares his knowledge regularly in YouTube course-length videos:
Meta is also reportedly working on a standalone AI application that will be released in Q2 of this year. CNBC reports
According to people familiar with this matter, the company plans to debut a Meta AI stand-alone app during the second half of the year. The people who requested anonymity because the project was confidential said that it is a major step towards Meta CEO Mark Zuckerberg achieving his goal of making his company the leader of artificial intelligence by the year’s end.
OpenAI’s Sam Altman joked on X that “ok fine, maybe we’ll make a social app” in response to news of Meta’s AI efforts being focused into a standalone application and not just within Facebook and WhatsApp.
The next edition of Neuronalwill cover the latest AI developments. Only on 9to5Mac. The previous issue is available here. Add 9to5Mac’s Google News feed.FTC: we use auto affiliate links that earn income. More.
