OpenAI Promises to Fix 'Laziness' in GPT-4 Model, Offers Discounts on GPT-3.5 Turbo

OpenAI will be making some big changes to its GPT-4 Turbo model with its latest update meant to cure its "laziness."

The ChatGPT developer announced on Thursday that it will be fixing the bug causing the chatbot to fail in completing a task, as well as an update to remove the issue affecting non-English UTF-8 generations.

OpenAI did not disclose the cause of the problem. The company earlier said it had not updated the model since November, a probable reason why the chatbot is bugging out.

The company has previously been criticized for being unable to perform well despite the company's claims that AI will be an "all-for-one" answer to the tech industry.

The issue also happened while the company is facing copyright infringement charges from publishers and authors for allegedly using their licensed work as training data without consent.

The update comes in as OpenAI prepares to launch the model to the public very soon. As of writing, GPT-4 Turbo is only available on preview.

According to OpenAI, over 70% of its customers have already transitioned from the GPT-4 API to GPT-4 Turbo due to its "updated knowledge cutoff, larger 128k context windows, and lower prices."

OpenAI Cuts Down Prices for GPT-3.5 Turbo

GPT-4 customers are not the only ones that will benefit from the recent announcement.

OpenAI is also set to debut the GPT-3.5 Turbo next week at a much lower price.

The model is touted as a less effective but more affordable version of the standard GPT-3 chatbot for API and third-party extensions.

The input prices are set to be reduced by "50% to $0.0005 /1K tokens," while output prices are cut down by "25% to $0.0015 /1K tokens."

Customers that own the current version of GPT-3.5 Turbo will receive the update automatically two weeks after release.

OpenAI Launches New Embeddings at 'Lower Pricing'

In addition to GPT models, OpenAI will also roll out a new feature called "embeddings" to make it easy for machines to "understand the relationships between content."

According to the AI firm, the embeddings are a sequence of numbers that represent key concepts within an image, video, or text that can be analyzed by the AI.

The feature is expected to improve performance tasks of ChatGPT and Assistants API.