2 min

Tags in this article

, , ,

Paying users of ChatGPT can now use a new version of GPT-4 Turbo. In addition to more accurate answers than before, the chatbot will waste fewer words on simple answers.

The new model (gpt-4-turbo-2024-04-09) also has a more recent dataset than its predecessors. OpenAI has provided this LLM with information up to December 2023, a substantial step forward from the previous cut-off date of April 2023.

Another improvement, however, is more striking. The new model lets ChatGPT provide shorter responses. This solves a common problem of the chatbot: its output is often more elaborate than necessary. For example, answers to brief questions often contain extensive explanations of the terms discussed, or the chatbot would close off its answers with added disclaimers that the user had not asked for.

The updated GPT-4-Turbo variant is available through the paid services ChatGPT Plus and Enterprise.

More direct

The summary responses are reflected in many ways. OpenAI indicates that the improvement can be applied to all written commands. An example on X shows how this efficiency leap plays out:

Twee screenshots van sms-berichten: de ene is een vriendelijke herinnering om te reageren voor een verjaardagsdiner, en de andere is een vriendelijke follow-up op zoek naar een reactie.
Source: X

OpenAI has continuously refined the chatbot’s functioning since the introduction of ChatGPT in November 2022. The biggest update to date remains the expansion to GPT-4 in March 2023, a model exclusive to ChatGPT Plus subscribers. GPT-4 Turbo later increased the efficiency of the LLM, which is especially beneficial to organizations that receive outputs from GPT models via OpenAI’s API.

Incidentally, the shorter responses are also likely beneficial to OpenAI’s power consumption. Shorter outputs cost less computing power, while more efficient models also have fewer requirements. In addition, the more accurate answers should make users feel less compelled to re-post the same question to ChatGPT.

Tip: ChatGPT’s water consumption is astronomical: 40 prompts demand one litre

Human behavior

Although ChatGPT is far from an “Artificial General Intelligence” with the cognitive ability of humans, human traits are more often attributed to the chatbot as changes occur under the LLMs’ hood. For example, users have noted that ChatGPT seemed to have gotten “dumber” during 2023, while an earlier modification to GPT-4 Turbo was supposed to result in a conversation with a less “lazy” virtual conversation partner.

Also read: ChatGPT gets smarter, cheaper and less lazy