iFlytek Races to Match GPT-4 Turbo by 2024
iFlytek aims to significantly upgrade its Spark large language model to match the performance of OpenAI's GPT-4 Turbo model by first half of 2024. The goal is to establish leadership in key areas like language comprehension, mathematical abilities, coding while catching up in complex reasoning.
Summary
- iFlytek's current Spark V3.5 model is close to GPT-4 Turbo in performance, surpassing it in some areas like language and math. But lags in complex reasoning.
- To boost Spark, iFlytek partners with Huawei to develop dedicated computing infrastructure "Feixing No.1" using Huawei's Ascend platform.
- Focus is on developing vertical capabilities in Spark for key industries while striving for parity with general large models like GPT-4 Turbo.
- Computational resources and data remain key challenges in further development of large language models.
- Domestic tech firms aim to surpass ChatGPT. But marketing hype makes claims suspicious despite rapid progress.
- Tencent, ByteDance also testing own internal AI chatbot and large model projects in competition with OpenAI.