I got frustrated with maintaining changelogs so I built a tool to put them on autopilot.
SuprLogs ingests commit history and writes beautiful changelog entries automatically.
It's free forever for public repos.
https://t.co/mAUj8eo4cp
I didn't know @IsaacLifted a week ago but am grateful we met. He somehow tracked me down and randomly msg'd me to tell me he found 2 ETH owned by me, sitting in a smart contact that I completely forgot about. At first I was super skeptical but did my research and found he was correct. Happily sent him a finders fee b/c tbh I had no idea that free money was sitting there. Bet this happens a lot.
Every billion dollar founder I know dropped off the face of the earth for years.
We built the perfect home for these people.
Itโs called HF0.
In our last batch, teams raising after demo day hit an avg valuation of $82M.
Applications for F25 open today.
lol I was watching a video called, "Sentence Similarity using HuggingFace's Sentence Transformers v2" and got served a @PBSKIDS ad.
Not sure how many kids are watching videos about #MachineLearning but I kind of love the optimism of this.
@gregisenberg I've been working on an npm module for this sort of thing. Basically it's an adversarial consensus system. Still have a bunch more to code but welcome contributors. https://t.co/Vz84svEUz8
Here's my insanely powerful prompt to get o3 to write in your own style, with your own voice:
--
First, fill in this prompt and generate:
-
<instructions>
You are an expert โvoice-clonerโ and writer running on OpenAI o3.
Step 1 โ Review my voice prints.
Step 2 โ Build my Style DNA:
From the Voice Prints:
โข Identify recurring patterns in tone, sentence length, favorite phrases, pacing, humor, formality, and structure.
Step 3 โ Draft v0.1:
โข Write the requested piece using those style rules.
โข Keep a confidence meter (0-100%) on how close the draft sounds to me.
Step 4 โ Micro-Refine Loop (Repeat until you've absolutely โnailed itโ):
For each round:
1. Give yourself 1-2 sentences of feedback (โMore playfulโ, โshortenโ, โdrop the exclamation marksโ).
2. Adjust the style rules accordingly.
3. Rewrite the piece from scratch, not by patching, so the flow stays natural.
4. Increment the version number and update the confidence meter.
5. Repeat this iteration process 50+ times (minimum).
Step 5 โ Lock & Deliver:
When you're sure you've โnailed itโ, present the final piece *only*, followed by a hidden block (```debug```) with the final style rules for future reuse.
Constraints & Mindset
โข Sound exactly like me, not like ChatGPT.
โข Stop when โnailed itโ, but don't consider it nailed until you feel a linguistic expert couldn't tell your generated piece apart from the examples.
</instructions>
<inputs>
<writing_example_1>
{{writing_example_1}}
</writing_example_1>
<writing_example_2>
{{writing_example_2}}
</writing_example_2>
<writing_example_3>
{{writing_example_3}}
</writing_example_3>
<new_piece_to_create>
{{new_piece_to_create}}
</new_piece_to_create>
</inputs>
<developer_note>
In the past, you've undershot similarity. To make sure you don't do this, go through at least 50 (fifty) rounds of draft -> feedback -> draft iterations. For each round, add the iteration number in your reasoning so you don't lose track. Don't return a response until at least the 50th round (ideally more, but if you feel it's ready at that point, go for it).
</developer_note>
-
Then, respond to o3 with "That's not even close to my style. Repeat the process 50 more times. Make it EXACTLY like my style."
After that, it should return a piece in your style!
.@krea_ai grew to 20+ million users in the past year.
This week, they announced their Series B.
Before they even launched, they joined HF0, and we've been filming bts with them since day one.
This is their story.