Keep seeing people say LLM writing is bad. I agree. You have to put in work to have it generate something usable
Here's the blueprint. Subagents are key
Here's what I did and what worked best for me. This is for generating writing close to your voice not generic
1. Collect a few long form pieces of writing you did pre AI. Doesn't need to be perfect but needs to be something you would say reflects your voice
2. Drop it into claude and ask it to create a voice profile for you based on the writing. Tell it to focus on sentence structure, literary techniques used, length and depth of paragraphs, analogies, construction
3. Tell Claude "kick off a subagent that is instructed to follow the voice profile we just generated and have it write a similar length piece on one of the topics i uploaded. Do not bias the subagent with anything beyond the voice profile and the topic and format. After that is done kick off another subagent that gets the written piece from subagent 1, the voice profile and the original piece and it's job is to grade the prose of the written piece. Focus on how the new piece is written not necessarily details of the content"
4. Use the second subagent's feedback to improve the voice profile
5. Repeat step 3 until the grading is high and you agree
Then finally tell Claude to package it up into a "my-voice-skill" that you can reuse from here on out
Tibo let this be a lesson to stop giving out early access to influencers instead of real senior level engineers who can provide real actual feedback
Most people you are giving early access to are just thinking about what to put in a video or tweet because they are algorithm chasers
@elie2222 Yeah this seems extremely fishy. For no models ever, have these cretins done this. Usually they just glaze it coz OpenAI is paying them
Maybe this go around, OpenAI decided to not fund these online cretins and they don't like that
@elie2222 Fwiw, i think they are actually providing an unfiltered opinion which is good because it's honest. For a change. Usually the influencer campaign calling it "The best model ever" is so strong it's nauseating. Very odd lack of such a campaign this time
They don't need to be on bad terms. Early access is a given to most popular online influencers. But i suspect it's either:
1. OpenAI decided not to do some usual compensation in the form of direct payment or slashed discounts or unlimited tokens
2. OpenAI decided to just leave it at early access whereas before they may have extended more perks like onsite visits, featuring them on launch videos etc
3. Ant is running a counter influencer campaign
I have just simply never seen a coordinated dunk quite like this
The problem is actual purveyors of AI slop outnumber "slop for good" people. I am in the same boat where i have seen so many lazy meat proxy responses back to me that my first instinct is to treat it as slop
There may come a time where the inflection point happens and more people are AI literate but until then it is slop by default unless proven useful
Blake these are amazing ideas and in a society with high agency, reasonable people this should be executed with utmost urgency
Unfortunately there are senile idiots everywhere in the positions of power who would stop reading just because you said "Starlink" which would mean "mUsK bAD"
MKBHD has turned out to be like the Theo of his world. Confidently wrong. Lacking humor and good vibes. Unwilling to take some heat. Nitpicking on basic things to keep up the facade of being right
It must be something about YouTube fame in tech that fries your brain like this
The memes are everywhere, so let’s go back over the bet: Cybercab publicly available for under $30k by the end of 2026
My hair is safe for several reasons
1: No deliveries yet. But they’re estimating starting in December so that’s neat
2: Invite only, not actually publicly available yet
3: You can’t actually use it as a cab(??!) It’s limited to private use only. Inside the existing geofenced locations. It’s a paperweight if I take this thing home. That’s one’s not actually part of the bet but yeah not exactly what was promised in this first wave
Hey @grok, set a timer for 48 days from now
Wow. So the malicious Claude extension on cursor marketplace and the fireworks hack turn out to be related. Attacker pivoted from fireworks into cursor and placed malicious extension on their marketplace
Some truly sophisticated attacks are happening
We’ve published an update on our recent security incident. See a detailed overview of what happened and our remediation steps here: https://t.co/2lf6lRJPlL
@goelrishabh "It's not random". But also
"We won't tell you the reason"
Brother. That means it's random to the end user if no one knows WHY or HOW they can end up in this position themselves. It doesn't matter that you know. The users need to know for it to not be random
@willdepue Anything for engagement I guess. Neoclouds and cloud provider can already provide indirect access to Nvidia compute. They could literally just have bought licenses to use those. Doesn't mean they physically had access to the GPUs