PG's Founder Mode killed the advice to hire good people and give them room to execute.
Hiring has its own version "write the job description, hand it to a recruiter, give them room."
It fails for the same reason.
Recruiters match the brief. At early stage, founders choose the person they enjoy working with for years.
That is founder-engineer fit.
2
3
81
The level of sophistication. I would hire Will without an interview.
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
72
Founder-led hiring is like...
The job: judging fit, selling the role, deciding.
The pile: finding, verifying, DMing, scheduling.
Done well, you spend your time on the THE job.
Done badly, weeks in the inbox, 10 minutes deciding.
1
42
My favorite interview question is "tell me about something you built, and where it broke."
The engineers who end up getting hired usually answer it in about 90 seconds or so.
1
8
291
On your next job interview, ask "How do you think I would fit into the team?", and see the magic.
1
152
old software wasn's safe either. it's just nobody looked.
finding a hole used to take a skilled person weeks so most code never got checked
now that cost collapsed and decades of it are coming due all at once
In grade 6 I couldn't add fractions.
Teachers gave me B's anyway, for singing at an event and playing Grandfather Frost at New Year.
In grade 7 my parents moved me to a boarding school. I prepared the whole summer to get in. By grade 9 I had finished the entire school math program.
3
5
325
AI code that passes every test is the dangerous kind.
Green checks, and nobody on the team can explain why it works.
If nobody can explain the PR, nobody owns it.
4
222
two companies, solo. when I'm deep in a problem, nobody shows up. it's just me.
so, long ago, I stopped asking permission from people who won't be there.
6
11
241
Measuring engineers by how many AI tokens they burn is counting lines of code all over again.
Last time we got more code. This time we get bigger bills.
Count what shipped and what broke.
2
2
135
Bots don't come for the people. They come for the likes and followers.
Take those away and the bots leave on their own.
1
1
138
OpenCode breach came through the hosted dashboard where people saved their OpenAI and Anthropic keys.
The coding agent had nothing to do with it.
My rule: keys stay on your machine or your own gateway, each with a hard spend cap.
5
9
373
I replied to your product drops with the first question a buyer would ask.
None were about features.
"when my process changes next month, who updates the tool?"
"can I move all my clients without anyone getting double-billed?"
"what does the AI decide on its own and what it leaves to me?"
Most founders pitch the day it works.
Buyers decide based on the day it breaks.
3
7
371
Can stupidity be a moat? Like strangers sleep on strangers' mattress. I mean it worked for Airbnb.
3
10
390
Yesterday 130 founders dropped their product in one line.
I replied with the first question their customer would ask.
The best responses all did the same thing: they removed a step. None added a feature.
Top 3 were from: @JackHFIL @nonstabIe @georgebnov
👇Drop yours. Rnd 2!
This quoted post is unavailable.









