OpenAI unveils GPT-4 Turbo with enhanced code generation, addresses ‘laziness’ complaints

In a recent blog post, OpenAI announced the release of GPT-4 Turbo, a revamped version designed to excel in tasks like code generation, addressing concerns raised by some users regarding incomplete task execution. OpenAI emphasized that GPT-4 Turbo is specifically tailored to mitigate instances of 'laziness' observed in its predecessor.

Despite the acknowledgment of the update, the company remained tight-lipped about the specific modifications made to GPT-4 Turbo. 

Critics of the existing GPT-4 model have recently voiced their grievances, attributing frequent refusals to complete tasks to the lack of updates. OpenAI clarified that the update is exclusively for GPT-4 Turbo, a preview version trained on information as recent as April 2023. Users relying on GPT-4, trained on data available before September 2021, may continue to experience similar issues.

OpenAI revealed that over 70 percent of users accessing GPT-4 via its API have migrated to GPT-4 Turbo, citing its more updated knowledge base. The company assured users that additional updates to GPT-4 Turbo are on the horizon, with the promise of general availability, including vision capabilities, in the coming months. This advancement will empower users to engage in multimodal prompts, such as text-to-image generation.

In a further development, OpenAI introduced smaller AI models named embeddings. Described as "sequences of numbers that represent concepts within content," embeddings support applications using retrieval-augmented generation. This type of AI retrieves information from a database rather than generating answers independently. OpenAI launched two variants of these models - text-embedding-3-small and the more robust text-embedding-3-large - both of which are now accessible to users.

Also Read

ChatGPT maker OpenAI revises AI usage policy to allow 'military and warfare' applications

‘I wish I could say ChatGPT’: OpenAI CEO Sam Altman reveals his most used app on Bill Gates’ podcast