Linus Tech Tips Discussion

 GPT Atlas: Sam Altman takes chromium, adds OpenAI API, and claims he revolutionized browsing Summary I feel a demo video says more than a thousand words.   In this demo from OpenAI, the devs ask a sidebar to look at a recipe and order the ingredients, it takes a lot of typing, and it adds the wrong quantities to the cart (instead of ordering 6 green onions, it orders 6 bounches of green onion).   Quotes   Quote AI gives us a rare moment to rethink what it means to use the web. Last year, we added search in ChatGPT so you could instantly find timely information from across the internet—and it quickly became one of our most-used features. But your browser is where all of your work, tools, and context come together. A browser built with ChatGPT takes us closer to a true super-assistant that understands your world and helps you achieve your goals.   My thoughts Sam Altman took chromium, and packaged an OpenAI sidebar hooked to OpenAI APIs, a tab screenshot functionality and click functionality to let the user interact with open tabs. All by chatting with the sidebar.   It's unironically something that I personally am confident I can build with local AI hosting. And looks to me like it would be a terrible user experience.   LLMs are incredible, at a narrow range of tasks. E.g. creating a doxygen comment blocks, creating an example snippet to learn a new library, letting you rewrite a mail or phrase with different styles and emphasis, sentiment analisys, recap/translate a transcript, OCR and so on.   Accurately filling forms without error is NOT one of those tasks, even basic arithmetic is hard because of the temperature, the model can just... count wrong. It's statisticaal in nature, not deterministic. I do use LLMs to generate json structures, but if the number are importants, I need to double check the work, and that negates productivity gains. Deterministic tasks are usually a bad fit for LLM automation.   Sam Altman released a product that leverages an amility LLMs are notoriously bad and inconsistent at, so much that the DEMO got it wrong, and the OpenAI devs didn't notice.   There is a clear trend of AI assist products from venture capital funded companies getting worse, not better   Someone famous once said, "tell me the incentive and I'll tell you the outcome". Sam Altman won when he got venture capital to give him hundreds of billions of dollars. His incentive, is to string investors along. This I feel explains why products that are BAD at what the product should do keeps being released by companies filled with venture capital.   K SHAPED DEVELOPMENT   This contrast heavily with enormous, and vast improvement in model functionality both in open and clo