Same here, I created a prompt enhancer gpt and a prose enhancer gpt, and I tend to chain all my prompts through them, then I use an extension to remove markdown + replace Unicode, and then add tabs and proper formatting to product a final version of all my prompts. This tends to result in prompts that perform 20-25% better for all difficult or multi-part tasks.
Logically this makes sense, the probabilites for next tokens the model produces follows the pattern it observes from the initial input, if your prose reflects that which individuals with higher intelligence tend to use, the model will continue this high in response and vice versa
I’m curious to see what happens as the % of training input using synthetic data generated by models tips the scales so that markdown reflects higher intelligence inputs - I wonder when this will occur
I think you did an amazing job of converting this into something that people can easily understand and use within their codebase - so props on that. It however would be the the right thing to give the proper credit to the source of idea - at least to thank them for their work and effort given this more of less is a direct copy of what they published https://www.intent-systems.com/ (not me and it’s something I use as well, but proper thing to do is give credit where credit is due)
How long do you think it will be until the “ai isn’t doing anything” people are going away
1 month, 6 months, I’d say 1 Year at the most, anyone who has used Claude code since Dec 1st knows this in their bones, so I’d just let these people shout from the top of the hill until they run out of steam…
Right around then, we can send a bunch of reconnaissance teams out to the abandoned Japanese islands to rescue them from the war that’s been over for 10 years - hopefully they can rejoin society, merge back with reality and get on with their lives
Have y’all tried Claude code using opus 4.5 - I believe it has fully joined the workforce, had my grandma build and deploy her own blog with a built in cms and an admin portal, post editor, integrate uploads with GitHub, add ci/cd and took about 2 hours mostly because she types slow
I don’t agree with the conclusions he draws from his own analysis - almost all the issues and shortcomings he points out are related to technological shortcomings he admits - are already being addressed by the new systems - or are primarily issues with how drones are being used in the field - i.e. tactical combat decisions.
These are not inherently valid arguments regarding the effectiveness of drones as a new weapons platform - but with the current state of the technology and with the decisions on the battlefield
It’s early days, the technology will improve and the tactics will be standardized with time and drones will prove to be a dangerously effective tool - which has the additional scary bonus of being cheap and easy to mass produce and deploy
Given the statement ‘we achieve SPDK-like throughput,’ I’m curious whether the performance is slightly worse than SPDK. If it is, do you have any comparison metrics for throughput?
Logically this makes sense, the probabilites for next tokens the model produces follows the pattern it observes from the initial input, if your prose reflects that which individuals with higher intelligence tend to use, the model will continue this high in response and vice versa
I’m curious to see what happens as the % of training input using synthetic data generated by models tips the scales so that markdown reflects higher intelligence inputs - I wonder when this will occur