<p>Today I would like to share with you excerpts from a remote article by Humanloop, which is based on a conversation between Sam Altman and 20 developers about OpenAI's plans.</p>
<ol>
<li>OpenAI is currently limited by GPU resources, which leads to difficulties in scaling models, such as the context of 32,000 tokens. Scaling with O(n^2) complexity is still a problem. OpenAI is willing to provide dedicated resources if the client makes a prepayment of $100k.</li>
<li>In the near future, OpenAI plans to create a cheaper and faster GPT-4, increase the size of the context window, enable model finetuning, and introduce a Stateful API. Multimodality (images, etc.) has been postponed until 2024 due to resource shortages.</li>
<li>Plugins are unlikely to be introduced in the API due to low demand. Sam Altman expressed the opinion that most people would prefer to have ChatGPT integrated into their applications rather than the other way around.</li>
<li>OpenAI will try to avoid competition with clients, except for ChatGPT. The company's goal is to create the best AI and sell it as an API.</li>
<li>Regulation is needed, but at the current generation of models, it makes no sense. OpenAI is considering the possibility of releasing GPT-3 as Open Source.</li>
<li>Scaling patterns still exist: growth in model and dataset sizes, albeit at a rate of 2-3 times a year.</li>
</ol>
<p>🧐 One can only wonder why OpenAI specifically requested the original post to be removed.</p>
<p>The removed post can be found here: <a href="https://humanloop.com/blog/openai-plans">https://humanloop.com/blog/openai-plans</a></p>;
<p>#openai #sam_altman #gpt4 #AI #GPT3</p>
