Recently, there has been a lot of turnover and big moves in the field of artificial intelligence (AI), especially two star companies, OpenAI and Anthropic.
Just as OpenAI held its DevDay developer conference, Anthropic announced the onboarding of OpenAI co-founder Diederik Kingma ·.
The differences between the two companies in terms of AI development direction and security philosophy seem to be reshaping the industry landscape.
Jinma's career has spanned academia and industry. He received his Ph.D. from the University of Netherlands Amsterdam in 2017 with honors, with research on deep learning and generative models.
Back in 2015, Jin Ma, who had not yet received his Ph.D., became a member of OpenAI's founding team, and later left in June 2018 to join Google.
Since then, he has been a research scientist at DeepMind, where he has led several research projects focused on text, image, and video generation models.
Jinma is also one of the inventors of the highly influential Adam algorithm in the field of deep learning. In addition, his contributions in this field include the development of variational autoencoders, Glow generative models, and variational diffusion models, among others.
In the field of AI, whether it is academic attainment or influence, Golden Horse ranks first. There's no doubt that his addition is a great addition to Anthropic's team.
Figure | Jinma's personal experience posted on his homepage (source: Jinma)
"Anthropic's approach to AI development is very much in line with my philosophy." "I look forward to contributing to Anthropic's mission to responsibly develop robust AI systems and can't wait to work with this talented team, which includes many wonderful former colleagues from OpenAI and Google," Kim Ma wrote on the X platform. ”
He will work remotely mainly in the Netherlands, but will occasionally travel to the San Francisco Bay Area in the United States.
This is the latest "talent war" between the two companies. The addition of Golden Horse continues the trend of high-level talent flowing from OpenAI to Anthropic.
Earlier this year, Anthropic recruited Jan Leike, · former head of security at OpenAI, and John Schulman, co-founder · co-founder.
In addition, in May, Anthropic appointed Mi·ke Krieger, co-founder of Instagram and Artifact, as its first product owner.
In fact, Anthropic's founders, Dario Amodei, and Daniela Amodei·· also employees of OpenAI were also employees of OpenAI but left in 2019 due to disagreements over the company's direction, particularly its deepening relationship with Microsoft.
Under their leadership, Anthropic has built a brand image that is more focused on safety.
OpenAI Technology Innovation: Developer Day 2024 Highlights
Despite the upheaval and major funding changes in the past few weeks, OpenAI hosted its 2024 DevDay as scheduled.
At the conference, the company announced several new tools, including prompt word caching, visual fine-tuning, model distillation, and a public beta version of the Real-Time Application Programming Interface (API) for building applications with low-latency, AI-generated speech responses.
There are no new models, no new products, and behind the somewhat monotonous launch is OpenAI's efforts to shift its strategic focus: it wants to build a more mature AI development infrastructure and developer ecosystem.
Figure | Some of the new products announced by OpenAI (Source: OpenAI official website)
Ke·vin Weil, OpenAI's chief product officer, said that the departure of Mira Murati, chief technology officer, and Bob McGrew, ·· chief research officer, will not affect the company's growth course.
"I would say that Bob and Mira are great leaders, and I've learned a lot from them, and they've been a big part of what we're to get to where we are today," Vail said. But at the same time, we're not slowing down. ”
OpenAI says more than 3 million developers have used its AI models to develop. The company has cut the cost of accessing APIs for developers by 99% over the past two years.
However, this is likely due to pressure from competitors such as Meta and Google to keep cutting prices.
The most eye-catching new tool is the real-time API. It allows developers to build near-real-time voice conversational experiences in their apps, with a choice of six voices from OpenAI.
These voices are different from those provided by ChatGPT, and developers cannot use third-party voices to avoid copyright issues.
Romain ·Huet, Head of Developer Experience at OpenAI, showcased a travel planning app built using a real-time API.
The app allows users to talk to an AI assistant to discuss an upcoming trip to London and plan a trip. In the demo, the user gets a response really fast.
It's worth noting that while these AI-generated voices sound fairly real, OpenAI doesn't add AI recognition claims for the AI to identify itself during the call.
OpenAI has also introduced visual fine-tuning capabilities into its API, allowing developers to fine-tune GPT-4V's applications using images and text. This should theoretically help developers improve GPT-4V's performance on tasks involving visual understanding.
Olivier Godement, OpenAI's API product ·lead, said developers will not be able to upload copyrighted images, images with violent elements, or other images that violate OpenAI's security policies.
Finally, no new AI models were announced at this year's Developer Day. Waiting for the developers of GPT-5 or the video generation model Sora will have to wait a little longer.
Resources:
https://techcrunch.com/2024/10/01/openais-devday-brings-realtime-api-and-other-treats-for-ai-app-developers/
https://techcrunch.com/2024/10/01/anthropic-hires-openai-co-founder-durk-kingma/
http://dpkingma.com/
Operation/Typesetting: He Chenlong