Inflection AI, maker of the AI-powered Pi chatbot, claims to have built a new LLM that is competitive with Google’s PaLM-2 and **Meta’s **LLaMA 2.
Dubbed Inflection-2, the LLM reportedly matched up with - and in some instances - outperformed PaLM-2 and LLaMA 2 across multiple benchmark tests. As reported by Search Engine Journal, Inflection-2 held its own in math and code reasoning tests, despite not being specifically trained to handle math challenges.
Inflection AI notes the LLM was trained on 5,000 Nvidia H100 GPUs (which have become an industry standard), and that it has plans to develop an even larger model using 22,000 GPUs.
Inflection AI says it will incorporate Inflection-2 into Pi.
Why it matters: OpenAI’s *ChatGpt and Google’s Bard have been grabbing headlines over the last year, but Inflection AI’s announcement shows the AI race still has plenty more miles to go yet.
*Search Engine Journal | Link to story.
**The New Yorker **profiles Jensen Huang, co-founder and CEO of Nvidia.
Huang explains that, in his view, AI deep learning is not to be feared. “I know how it works, so there’s nothing there,” says the executive. “It’s no different than how microwaves work. All it’s doing is processing data.”
The piece explores Huang's history, the founding of Nvidia, how the company almost went bankrupt, and its eventual breakthroughs in the GPU space. It goes on to explain that Nvidia paid little attention to AI until industry pioneers such as Geoffry Hinton, Alex Krizhevsky, and Ilya Sutskever revealed to the company the capabilities of its GPUs for training neural networks.
Huang notes Nvidia’s long-term goal is to unify its GPU unit with its generative AI division. Huang suggests this could ultimately lead to the creation of immersive worlds, consisting of almost-real graphics and responsive to human speech. Huang also posits the technology could improve the development of driverless cars, the creation of next-level entertainment experiences, and more.
Why it matters: *As the CEO of arguably the most influential hardware company in the AI space, Huang’s vision of the future will go some way to determining what the future is going to look like.
*The New Yorker | Link to story (paywall).
AI startup PhysicsX has emerged from stealth having raised a $32 million Series A funding round.
London-based PhysicsX has developed an AI-powered platform that allows simulated testing of physical objects. Aimed at the engineers and scientists working in materials, automotive, and aerospace industries, PhysicsX’s technology promises to expedite the development process by using AI in place of certain manual processes.
The founders of the company, Robin Tuluie and Jacomo Corbo, have backgrounds in the vehicle industry, with both having been involved with Formula One racing.
General Catalyst led the Series A, with participation from Standard Industries,** NGP Energy**, and others.
**Why it matters: **By its very nature, the material sciences space has always required a lot of physical testing. The introduction of an AI platform that can replace some of that development could be a boon to many industries.