OpenAI aims high with 'Strawberry'

New from Erin Woo, Amir Efrati and me: OpenAI is racing to release its 'Strawberry' AI in a product, as it looks to raise more capital. And, it's using the AI, which it's shown to national security officials, to generate synthetic data for its next-generation 'Orion' LLM. More details on Strawberry and the efforts here: https://lnkd.in/e65vKt7h

This is the kind of article why I subscribe to The Information (https://lnkd.in/dgS9dNrE) Summary: - OpenAI is developing a new AI model called Strawberry (formerly Q*), focused on improving reasoning capabilities. - Strawberry is expected to launch in fall 2024, potentially integrated into ChatGPT as a smaller, distilled version of Strawberry. - The new model, Orion, will be OpenAI's next flagship large language model and is currently being trained using synthetic data generated by Strawberry. - Strawberry was demonstrated to national security officials in summer 2024. - OpenAI plans to release Orion after Strawberry. Few things became clearer - there are definitely two separate efforts, Strawberry is more of a new class of models, I'm not expecting it to be great yet, so unlikely to matter too much in the short term. Then Orion is the GPT-5 class model which I'm guessing we should expect after November, so maybe Q1 next year or so? This would be the biggest test for the field of LLMs - would capabilities keep emerging as we scale?

This is the kind of article why I subscribe to The Information (https://lnkd.in/dgS9dNrE) Summary: - OpenAI is developing a new AI model called Strawberry (formerly Q*), focused on improving reasoning capabilities. - Strawberry is expected to launch in fall 2024, potentially integrated into ChatGPT as a smaller, distilled version of Strawberry. - The new model, Orion, will be OpenAI's next flagship large language model and is currently being trained using synthetic data generated by Strawberry. - Strawberry was demonstrated to national security officials in summer 2024. - OpenAI plans to release Orion after Strawberry. Few things became clearer - there are definitely two separate efforts, Strawberry is more of a new class of models, I'm not expecting it to be great yet, so unlikely to matter too much in the short term. Then Orion is the GPT-5 class model which I'm guessing we should expect after November, so maybe Q1 next year or so? This would be the biggest test for the field of LLMs - would capabilities keep emerging as we scale?

This is the kind of article why I subscribe to The Information (https://lnkd.in/dgS9dNrE) Summary: - OpenAI is developing a new AI model called Strawberry (formerly Q*), focused on improving reasoning capabilities. - Strawberry is expected to launch in fall 2024, potentially integrated into ChatGPT as a smaller, distilled version of Strawberry. - The new model, Orion, will be OpenAI's next flagship large language model and is currently being trained using synthetic data generated by Strawberry. - Strawberry was demonstrated to national security officials in summer 2024. - OpenAI plans to release Orion after Strawberry. Few things became clearer - there are definitely two separate efforts, Strawberry is more of a new class of models, I'm not expecting it to be great yet, so unlikely to matter too much in the short term. Then Orion is the GPT-5 class model which I'm guessing we should expect after November, so maybe Q1 next year or so? This would be the biggest test for the field of LLMs - would capabilities keep emerging as we scale?