news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Dream morning from Aofei Temple
Quantum Bit | Public Account QbitAI

ChatGPT has a new model!

The official account announced:The latest version GPT-4o has been installed since last week, we guess you’ll like it.

Then... it was awkwardThe comment section below was filled with the same tone: “Ah? What was changed? I didn’t notice.”

No improvement, I use it every day and there is no difference.

as well as“OpenAI is dead”It is speculated that the new model’s hard indicators cannot exceed those of the Claude-3.5-Sonnet next door, so it is not appropriate to release it officially, so it can only be released in a roundabout way.

Is the new version of ChatGPT more like Claude?

According to the official ChatGPT account’s response to netizens, this new model on ChatGPTThis is not the same as the previous API update gpt-4o-2024-08-06。

The good news isFree users can also experience(There is a limit on the number of messages).

Among the several popular large-scale IQ tests recently,How many r's are there in Strawberry?, ChatGPT's new model and gpt-4o-2004-08-06 can only be answered correctly under specific questioning conditions.

Currently, the only question that is known to be able to be answered correctly is "How many 'r's are there in the word Strawberry?", but no Chinese version has been found yet.

The same question can be asked for the API version. Version 05-13 cannot answer the question correctly, but version 08-06 can answer the question correctly.

To summarize more of the trial feedback shared by everyone, the changes are very subtle, mainly reflected in a more humane tone.

Entrepreneurs who study AI writing believe thatThe new version of ChatGPT feels a bit like Claude next doorThe specific changes started last Thursday, August 8.

Other changes may be made to the safety guardrails.。

AI jailbreak master Pliny found that he had to modify the jailbreak prompt words he had used for a long time in order to crack the new version of the model.

Emmm... In other words, the new version of ChatGPT has added protection but still failed to prevent it.

In addition, entrepreneurs who mainly use AI to generate regular expressions have discovered this.

ChatGPT's new model answering style and "anonymous chatbot" in the big model arenaanonymous-chatbotVery similar.

Ultraman leads the team, OpenAI employees collectively hype "Strawberry"

There is a large-scale performance art going on inside OpenAI around the next-generation model "Strawberry"/Q*, which outsiders cannot understand.

CEO Ultraman Benman took the lead and posted photos of strawberries as riddlers.

But this time is different from before, many OpenAI employees also participated collectively.

Tao Xu, an old employee who joined in 2019, arranged real strawberries into a Q shape, implying that strawberries are related to Q*.

Combined with the news that OpenAI's president is taking a long vacation and its co-founder is leaving, it seems like something big is brewing.

But everyone is really a little tired of performance art.