OpenAI launches compensation program for GPT developers
Advertisement
OpenAI is testing a new program that will allow developers in the US to earn money from the use of the GPT models they create. The program was first announced alongside Custom GPTs in November 2023. For this initial test, the AI company partnered with a small group of developers and aims to work with the developer community to determine the best approach to the pay-as-you-go model.
Details about the participating developers, the exact terms of revenue sharing and the schedule for a wider rollout are not yet known.
Amazon invests in Anthropic
Amazon is investing a further $2.75 billion in the AI startup Anthropic, increasing the total investment to the $4 billion announced in fall 2023 and thus securing an important partner in the field of generative AI. Anthropic is best known for the chatbot Claude and competes with OpenAI's ChatGPT. In return, Anthropic will move its software to Amazon Web Services (AWS) cloud infrastructure, also using Amazon's Trainium and Inferentia chips.
The collaboration between AWS, Anthropic and Accenture aims to help companies in highly regulated industries responsibly launch and scale tailored generative AI solutions. As part of the investment, Amazon will receive a minority stake in Anthropic, but no seat on the supervisory board. The company is now valued at $18.4 billion.
Adobe shows new AI functions
Adobe is presenting new functions for its Firefly image generator at Adobe Summit 2024. With the “Structure Reference” prompt, users can transfer the structure of an existing image to newly generated images to create multiple variants with the same layout. The combination of “Structure Reference” with the existing “Style Reference” function allows the structure and style of an image to be referenced in order to reduce the random factor in image generation.
Adobe also announces Custom Models for Firefly, which enable individual training of the AI model for your own corporate design language. According to Adobe, 10 to 20 training images are sufficient for good results. Firefly Services also offers over 20 APIs, tools, and services to automate content creation and editing that can be integrated into production workflows and Adobe Creative Cloud applications.
An expanded partnership also announced between Adobe and Microsoft will see elements of Adobe Experience Cloud and Firefly's generative AI integrated directly into Microsoft applications.
X.com is rolling out Grok to all premium users
Elon Musk, founder of xAI and owner of X (formerly Twitter), announced that the chatbot Grok will be available to all premium users this week. Previously, Grok was only available to Premium+ users and availability varied regionally. It remains unclear whether Grok will also be available in Germany.
Grok is based on the open source language model Grok-1, which was developed in just three months and is said to be comparable to OpenAI's GPT-3.5. Grok-1 is a Mixture of Experts model, where eight experts become active for each input token two to increase the efficiency of the AI model.
In contrast to other chatbots such as Google's Gemini or Adobe's Firefly, Grok has fewer barriers. Musk emphasizes that Grok is intended to be “of maximum benefit to all of humanity” and advises only using the chatbot if you have a sense of humor, as it responds “with a little humor.”
New open language model shows high performance
Databricks introduces DBRX, a powerful open language model that performs better in benchmark tests than established models such as GPT-3.5, Grok, Mixtral and Llama 2. The model comes close to GPT-4 in some tasks and outperforms Google's in others Gemini 1.0 Pro. Like GPT-4 or Mixtral, DBRX is a Mixture of Experts model and 132 billion parameters in size, with only 36 billion active at any given time, enabling high efficiency.
Databricks enables its customers to customize DBRX on the Databricks platform and train their own models on private data. The open source community can access DBRX via the Databricks and Hugging Face GitHub repository. Databricks wants to use an open approach to promote innovation and transparency in the AI industry.
Intel is expanding its AI PC Acceleration Program
Intel plans to accelerate the distribution of AI software for Windows PCs, particularly those with its own Core Ultra processors. To this end, the company is expanding its “AI PC Acceleration Program,” which was launched in October 2023, in two directions. Firstly, cooperation with more PC and device manufacturers will be expanded, with both large and small companies being included in the partner program. Intel engineers provide support, among other things, in optimizing hardware designs.
Secondly, the AI PC Developer Program is being created to develop software tools that use AI algorithms. It provides access to tools, workflows, AI implementation frameworks and developer kits that include the latest Intel hardware with Core Ultra processor. Such a dev kit is being developed in collaboration with Asus, which, after taking over Intel's NUC division, is contributing the hardware in the form of a mini PC with a Meteor Lake processor. The dev kit is expected to be available in the summer, with Intel and Asus still finalizing pricing and logistics plans.
How intelligent is artificial intelligence actually? What consequences does generative AI have for our work, our leisure time and society? In Heise's “AI Update” we, together with The Decoder, bring you updates on the most important AI developments every weekday. On Fridays we examine the different aspects of the AI revolution with experts.
(mack)