Today, Anthopic revealed Fable 5, a mythos-class model, and it is insanely powerful. This is arguably the most important AI news ever, because there is no model even close to this model's power. It even blows Anthropics' second-most-powerful model, opus 4.8, out of the water. Though on subscription, you will only be able to use Fable for about 2 weeks until June 22, which is disappointing, but it points out how expensive the model is to run. The input cost is $10, and the output is $50, which means Fable is about twice as expensive as Opus 4.8. The Fable model is so powerful that this mythos-class model completed a coebase-wide migration of a 50-million-line Ruby codebase in just one day. This is unbelievable. This task would have taken a team a few months; this model was completed in a day. This raises the question of when a model this powerful should be released to the general public, because its ability to code extremely well can easily uncover codebase errors. This poses extreme cybersecurity risks and biological risks related to the ability to create pathogens. Antorpic decided to make any requests related to cybersecurity, biology, chemistry, and distillation revert to opus. This is because it poses extreme risks and secrecy issues in biology and chemistry. With this model, it could help people make chemical weapons. What is distillation? It is when a company uses a much more powerful model from another company to train its AI model. Ensuring no distillation occurs prevents bad actors from creating a mythos class model without guardrails. It is a good way to provide the public with safe access. I believe Anthropic has tested the girdrails to make sure they work before retraining the model. Since it came out, I have been testing it and playing around with it, and it is extremely powerful. It found errors in my codebase that Opus wrote and did not see. Also, now you can give Fable a prompt, and it will complete it with goals and loops, which I will touch on later. It noticeably takes up an extremely large amount of tokes and should only be used for very complex tasks. Also, fable takes up an extremely large amount of time, though it is more token efficacy than opus. Now you might be wondering how to use this. I will tell you now. First, to use it most of the time, only use high or extra because it is already so powerful; above that, you will be using an extremely large number of tokens. I tried using Ulta code on the 20x pro plan, and it used up a quarter of my 5-hour limit in 9 minutes, which shows how many tokens it uses, especially with dynamic workflows. Next, you need to treat this model differently and work with it, and be a coworker or partner with it, because it is so strong that you do not have to worry whether it is coding correctly or doing a task correctly; rather, you only need to worry whether it is doing the correct task. This leads to my next advice: you should have it ask you a ton of questions so it knows what you are doing. When you tell it to do that, it will tell you exactly what it will do. Also, tell it to set up goals and loops for you, which are extremely important in this model because you can easily one-shot a prompt (getting a very usable product only after one prompt). Lastly, you should not need approval, as that will waste time; use auto or bypass if you are comfortable, though remember you are responsible for what the AI does. Overall, I am super excited for this model and will be testing it in the next few weeks. Next, OpenAI is going public after much anticipation. This follows an S-1 registration from Anthropic before OpenAI, and Anthropic gained a higher valuation than OpenAI. This means we will see the two largest AI companies go public in the foreseeable future, with a SpaceX IPO this Thursday. OpenAI will be forced to disclose its financials quarterly once it goes public. Traditionally, OpenAI is like other AI companies and will raise capital and generate revenue through hype in the AI space rather than through true profit. This is important because going public will likely bring in more profit for both AI companies, which are both in dire need of it. As AI models become increasingly powerful at an exponential rate, earlier models are used to help create the next versions. It will bring big changes to the space, with two humongous IPOs from AI companies, in addition to the largest ever IPO from SpaceX. I am very excited about this; it might also lead to cheaper models due to the extra revenue. At WWDC, Apple announced Siri AI, which, now, after a long-awaited period, actually makes Siri smart. It was developed in partnership with Google and Gemini, though Apple has not directly mentioned it. Siri will now be able to give answers that are not just a web search. Siri can now also tell what is on your screen. The most interesting part, though, is that Siri AI uses a mix of local and cloud AI models and does not run on Google's servers. It runs on Apple Cloud Servers when the task is extremely complex, ensuring the user's maximum privacy. The impact that Fable is going to have on the AI space is tremendous, and it has proven to be the best model out there now, though it costs an extreme number of tokens. OpenAI and Anthropic are both planning IPOs, which could bring more competition. Siri AI is now almost agentic and works very well. I am most excited about anthropic and cannot wait to see what they do next, especially once they go public.