The show of solidarity of the OpenAI team is inspiring. I am proud to develop on OpenAI. Just a short while ago, we risked losing the team, uncertainty about this platform and the API, and more.
No other company has done more for developers and equal access to the very best AI models than GPT. I don’t trust any group of people more with achieving AGI and then rolling it out in a way that shapes the future of humanity in a responsible way.
That requires both capital and safety, and I believe that Sam understands this better than most. Neither a doomer, nor a blind optimist, and with a company mission to ensure AGI benefits all humankind.
That said, I do have some requests / suggestions:
Please make a production environment. You can’t break production 5 times a week without giving your developer community and their customers heartburn. For example, the APIs fall back to lesser models at certain times / loads without notice, breaking our apps that depend on specific model traits, or ChatGPT releases functionality that then breaks. This happens multiple times a week. Why not have a canary program like Google Chrome, where we can vet & break things before you roll it out?
Please consider a basic roadmap so you don’t compete with your developers. It doesn’t have to give away secrets and timelines can be vague. That way we build things that are synergistic with you. It’s not healthy to shock your developers overnight with things that wipe out their work. Communication from OpenAI has not been excellent.
Please consider a lightweight open source model of GPT 3.5 turbo at the right time. Open source is getting close to 3.5 already, so there is no risk to humanity. Alternatively, keep it more cost competitive with open source models running on cloud infrastructure. LLAMA2 70B (not as capable - yet) on a certain provider is 6X more affordable and much faster. If you don’t keep pace with this, you’ll lose some of your developers. They will continue to be here for your best models, but not everyone needs the best models for every task.
Please consider giving Logan more resources to support the community. I’m astounded at how thin he is spread in terms of responsibilities.
So Sam, Greg, and others: I’m glad you’re back, I’m glad OpenAI will thrive, and congratulations to the OpenAI team who achieved this in solidarity with one another. This is a day for celebration.
Please, restore the GPT-4 model from March. Do not downgrade models and performance. Only UPGRADE.
Please, return the old output token limits and restore the Continue Generating button if exceeded.
If you cannot sustain a good service, admit it, offer a more expensive solution for people who want a high-grade product and want to pay more. I would pay double if I have the quality and I would pay ZERO for the joke ChatGPT has become.
Best Regards
ex-subscriber who cancelled his plan and looks forward to Claude becoming available in Europe.
Hello Zarko,I read your message regarding your recent experiences with ChatGPT and the decision to cancel your subscription. Your perspective on the updates being a downgrade piqued my curiosity, and I would like to understand more about what specifically led you to this conclusion.
Could you share some examples or specific aspects of the service that made you feel it has been downgraded? This would be helpful for a clearer understanding, as experiences can vary from one user to another.
On my end, I haven’t perceived a significant decline in the service’s quality. In fact, with the introduction of new tools like the GPTs and the assistant feature, it seems to me that these changes are not so much a downgrade but rather an adaptation. The AI field is rapidly evolving, and each update brings its own set of challenges and improvements. I’m optimistic that, over time, these updates will only enhance the overall user experience.
Your detailed feedback is valuable, as it might shed light on aspects of the service that could be improved or adjusted. Looking forward to hearing more about your specific experiences.
You make a valid point. The effectiveness of AI models like ChatGPT often depends heavily on the context provided by prompts. If the prompts are unclear or lack specific details, it can lead to less accurate or relevant responses.
Regarding your observation of the model seeming ‘dumber’, subjective experiences can vary, but they are important feedback for ongoing development and refinement of these AI systems.
I completely agree with your sentiments regarding the solidarity and achievements of the OpenAI team. It’s truly inspiring to see how they’ve navigated challenging times and maintained a strong focus on advancing AI technology responsibly.
Your trust in OpenAI’s commitment to AGI development and ethical rollout reflects a widespread confidence in their capabilities. The balance between innovation and safety is crucial, and it seems OpenAI is indeed approaching this with the necessary seriousness and thoughtfulness.
Your suggestions are very pertinent. A stable production environment, transparent roadmaps, and effective communication are essential for fostering a healthy developer community. The idea of a canary program for testing, as you mentioned, could be a game-changer in ensuring stability and continuous improvement. Also, the point about open-source models and pricing competitiveness is particularly relevant in today’s rapidly evolving AI landscape.
Acknowledging the hard work and dedication of individuals like Logan in supporting the community is important. More resources in this area could indeed make a significant difference.
Here’s to celebrating the team’s success and looking forward to the innovative strides OpenAI will continue to make in the field of AI.