OpenAI officially announced a major open source announcement! The most powerful open source model is coming? Ultraman revealed that 1 million new users were added in the past hour
OpenAI has released its first open-source model after GPT-2, and the countdown for developers to celebrate has begun!
Core content:
1. OpenAI announced the open-source of a new model, focusing on strong reasoning capabilities
2. Offline events in San Francisco, Europe, and Asia-Pacific will be launched soon, and developers can experience the prototype in advance
3. Safety comes first, OpenAI has triple protection and will never release high-risk models
Sam Altman personally wrote: Open source for the first time since GPT-2, developers countdown to the carnival
Key points :
1️⃣ OpenAI announced that it will release an open-weight model , which is the first open-source model since GPT-2. 2️⃣ The new model focuses on strong reasoning ability . Sam Altman said frankly that "it will be very, very good." 3️⃣ Developers can participate in feedback in advance. Offline activities in San Francisco, Europe, and Asia Pacific will be launched soon. 4️⃣ Security is the core: Special tests such as adversarial fine-tuning have been launched, and high-risk models will never be released.
OpenAI CEO Sam Altman just published a message throwing out a "nuclear bomb": "We plan to release a powerful open source large model in the next few months . "
It’s been 26 months since ChatGPT exploded onto the scene, one of the craziest viral phenomena Altman has ever witnessed, when OpenAI attracted a million users in just five days.
And just in the past hour, OpenAI has added another million users
Altman certified "the craziest viral moment in history", and this time OpenAI chose to return to its original intention of open source - this is the first time OpenAI has open sourced a language model since GPT-2 in 2019.
Why open source now?
Altman "frankly" said: I have wanted to do this for a long time, but it has never been a priority. Now that the time is right, the team thinks "this matter has become important", so we must go crazy @ DeepSeek ?
The new model has two main features:
• Open -weight • Strengthen reasoning skills
Although the specific parameters were not disclosed, the official used two "very" to emphasize the goal: "to make it a very, very good model"
Developers, please take note!
OpenAI has started global warm-up:
1️⃣ San Francisco first offline event2️⃣ Europe and Asia Pacific sessions will follow3️⃣ Early registration channel is open (link at the end of the article)
The purpose is to collect feedback and allow developers to experience the prototype in advance . The official statement is clear: we want to see how large companies and governments deploy models independently.
Playing the safety cards
From pre-training to release, OpenAI offers triple protection:
? Use the self-developed "Preparedness Framework" evaluation framework ? Focus on studying the risks unique to open source models (such as adversarial fine-tuning) ? Internal testing + third-party expert double insurance
Altman was categorical: “Never release a model that could create catastrophic risks.”
Summary in human language:
OpenAI's open source is not only an olive branch to developers, but also a reassurance to the industry - "open source and security can coexist ." As for how powerful the model is? Let us know your expectations in the comments section!
Attachment: Official feedback link
https://openai.com/open-model-feedback
full text:
Sam Altman
OpenAI
When ChatGPT launched 26 months ago, it was one of the craziest viral stories I’ve ever witnessed, as we reached 1 million users in just five days. And just in the past hour, we added another 1 million users.
Long story short: We are excited to release a powerful new open source weighted language model with inference capabilities in the coming months. We hope to engage with developers on how to get the most out of this model: https://openai.com/open-model-feedback We are very excited to make this a really, really great model!
We are planning to release the first open source weighted language model since GPT-2.
This is something we've been thinking about for a while, but other things took precedence, and now we feel it's the right time (or important enough) to do it.
Before publishing, we will evaluate the model against our Preparedness Framework like any other model. And, given that we know the model may be modified after publication, we will do additional work. We still have some decisions to make, so we will host developer events to collect feedback and let people try out early prototypes later. Events will start in San Francisco (SF) in a few weeks, and will be held in Europe and Asia Pacific (APAC) later. If you are interested in attending, please sign up through the link above. We are excited to see what developers create and how large enterprises and government agencies that prefer to deploy and run the model themselves will use it.
Security was a core concern as we developed this open source weight model, from pre-training to release. While open source models present unique challenges, we follow our security readiness framework and will not release models that we believe could pose a catastrophic risk. We are particularly focused on investigating adversarial fine-tuning and other risks unique to open source models. As with all model releases, we are conducting extensive security testing before public release, both internally and in partnership with trusted third-party experts.
