I'm writing this essay a few weeks late (like the last ones) and only part way through Week 4 as I've been falling behind due to having too much other stuff going on. These weekly essays are supposed to be about the prior week's work and discuss what I learned and what I still need help on, but this one is going to be different. I'm going to talk about Claude Mythos and Claude Fable, the "too dangerous" models that Anthropic released the week I was doing Week 3 and which the government decided to hit with an export control restriction a few days later.
I was able to use Claude Fable for only about a day as I had been busy with other things and did not have the time. I had previously built a few web-apps with Opus 4.8: two websites for my armored combat team and a blog / tracker for this coursework that you're currently reading this on. To say I was blown away would be an understatement. It was night and day the abilities of Fable compared to Opus and I have been having withdrawals ever since. I pointed Fable at all three sites and asked it to look at the code, see what could be improved, and suggest anything that might need fixing. It found vulnerabilities that Opus had created / missed and suggested a few new features that were very useful. It also redesigned the mountain and tracking in the background to actually look like a mountain and track normally. The detail and level of coding was absolutely insane. I have since built about six more sites / apps with Opus that I can't wait to point Fable at when it's back, assuming it ever comes back online.
In regards to the export control restrictions that the government decided to place on Anthropic and their two latest models, I personally think it's an extreme overreach and has nothing to do with the models themselves and is instead payback for Anthropic not caving in to the government's demands a few months ago. It also doesn't paint a good look that Sam Altman and OpenAI were recently working with the government on certain things. Now in the present, the government has now expressed some concerns with OpenAI's latest model (which is in no way as powerful or useful as Mythos and Fable were) but I think it's to save face and not look like it was a political hit.
Either way, the models were amazing and I'm hoping they come back soon. I'm not sure how that will work, if you'll be required to upload your ID or verify via those phone apps that some sites use (thinking KYC on crypto sites) or what, but I'm not sure I'm comfortable with that. I guess I'll see when it comes to that point and in a future post I will detail the really cool ecosystems that I have built since in the past two weeks that bridge a few of my hobbies together into a cohesive environment.