Don’t be fooled, the first few years you get spoiled with “the first moments” of things. Then suddenly the “last moments” start creeping in, “the last nappy”, “the last car seat”, at first they seem like a god send but then they accumulate like an avalanche.
One day you will pick them up and, and most likely neither of you will know it, but it will be the last time you ever do.
> the more people you have using your models, the more training and fine-tuning data you're accumulating, so the faster you can develop the next frontier model
I’ve wondered this but then wouldn’t a large amount of input now just be AI output from a previous PR/client email/spec document/chat. Training of that would be an issue leading to distillation?
As an Australian, I’m not particularly surprised by this. From purely a capacity perspective, it seemed fair to reason, if AI is so powerful and capacity is an issue, why wouldn’t you prioritise domestic and restrict foreign usage.
It’s a massive betrayal for foreign entities and it would be silly to continue with all my eggs in anthropic basket but I get it.
Agreed. But I think it’s also important to realise if you sent this article back to 2020 people would say it was pure fantasy that a tool could do this. Hype aside, there’s a bit of cool magic here.
The author just wrote an anecdote about how a prompt to fix an issue played out. Their conclusion wasn’t about cost or gushing at its ability but that it’s dangerous:
> Fable is arguably smarter and hence more suspicious of potentially malicious instructions. But that smartness is very much a two-edged sword: if it does get subverted by instructions, the amount of damage it can do given its relentless proactivity is terrifying.
This might sound like a joke but it’s a surprisingly good take. RSCSS provides a great set of principles to structure and isolate against unintended change. Open Props can be added for design tokens and Open UI has some great examples of components.
From my 10 years in the .net, it seemed C# devs will pretty much do anything to avoid using the right tool for the job or solving the immediate problem at hand.
As the author notes in the end, it would be really interesting to do these again on more recent models. I wonder if the no context file being cheaper still stands. But then how much does the harness influence the results. It can be frustrating trying to gauge what’s influencing what and if something suddenly starts working against you.