If you can find a way to get the model to spit out bangers like “the next model will be Claude Shitpost,” please do let me know so I can spend less time writing.
(Hint: you’re also gonna find an absolutely massive number of AI tells in blog posts I published around 2019.)
On the contrary, I've been reading your AWS economics newsletter for years now (not since 2019 but for a while), and I've always appreciated both the humour and the insights. My favourite is the one where you talked about how AWS is like a train on rails.
However, in THIS newsletter specifically, some parts but not all are either AI generated or you've become conditioned by AI humour, both of which are quite bad. And I'm quite sure grok could come up with Claude Shitpost if you gave it a detailed enough prompt.
The damning part is that the key thing you focus on is wrong, a user doesnt pay fable rates when it falls back to opus. They actually specifically accounted for this issue and made it cheaper! Based on what I know about you I find it incredibly hard to believe that you would make a mistake like this and I can only assume its an AI hallucination.
Corey, I find it hard to believe that you'd give up the snarky opinions just because an AI could do it better. But I don't think that will be put to the test. AI specializes in predicting the most likely next word in a sentence, not the unexpected new phrase.
am I wrong, or are they charging the full cost of Fable 5 output tokens, and a 10% increase on input tokens, when you encounter a fallback? Also, I guess all developers of AI enhanced applications and agentic harnesses will have to code in fallback monitoring, and storage of fallback credits to a database for future use. Will azure, openai, etc. use the same fallback API, or invent their own (e.g. its more like a loyalty card to a grocery store).
do you think we cant tell that this is AI generated? very disappointing, thanks for wasting my time.
If you can find a way to get the model to spit out bangers like “the next model will be Claude Shitpost,” please do let me know so I can spend less time writing.
(Hint: you’re also gonna find an absolutely massive number of AI tells in blog posts I published around 2019.)
On the contrary, I've been reading your AWS economics newsletter for years now (not since 2019 but for a while), and I've always appreciated both the humour and the insights. My favourite is the one where you talked about how AWS is like a train on rails.
However, in THIS newsletter specifically, some parts but not all are either AI generated or you've become conditioned by AI humour, both of which are quite bad. And I'm quite sure grok could come up with Claude Shitpost if you gave it a detailed enough prompt.
The damning part is that the key thing you focus on is wrong, a user doesnt pay fable rates when it falls back to opus. They actually specifically accounted for this issue and made it cheaper! Based on what I know about you I find it incredibly hard to believe that you would make a mistake like this and I can only assume its an AI hallucination.
Source: https://platform.claude.com/cookbook/fable-5-fallback-billing-guide
Not mad, just disappointed. 💔💔💔
They even made sure that the fallback rates are billed as a cache hit, even though it would really break cache.
Corey, I find it hard to believe that you'd give up the snarky opinions just because an AI could do it better. But I don't think that will be put to the test. AI specializes in predicting the most likely next word in a sentence, not the unexpected new phrase.
" model was so good at finding software vulnerabilities..."
Um, not really, not better than other models. What it IS good at is writing exploits for them. Frighteningly, oh shit good.
am I wrong, or are they charging the full cost of Fable 5 output tokens, and a 10% increase on input tokens, when you encounter a fallback? Also, I guess all developers of AI enhanced applications and agentic harnesses will have to code in fallback monitoring, and storage of fallback credits to a database for future use. Will azure, openai, etc. use the same fallback API, or invent their own (e.g. its more like a loyalty card to a grocery store).