God-man | slightly techie 🤏🏽 Lead @thedsgnforge
Joined May 2020
- Tweets703
- Following537
- Followers111
- Likes1.6K
VĪ retweeted
Since we're talking about mundane things we find endearing.
I absolutely love it when your woman's walking a few steps behind you and you stretch your hand back, and she does that little run to grab it.
I couldn't find a term for it, so I dubbed it the Mjolnir call.
VĪ retweeted
Your National Service year can be more than just a requirement to fulfil.
We are opening two seats at Printmote, come discover what you enjoy doing and build something you can take into your next chapter.
Applications are open: printmote.com/careers
VĪ retweeted
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.
VĪ retweeted
I assure you, you would change your position if you were placed in a different environment. You are in a country where engineers are doing the work of technicians because there are few engineering jobs, most of which do not even pay well.
Just an internship in a proper engineering company will show you why some of what you think is mere theory is actually worth learning.
Bro I don’t understand why there should be BSc certificates for engineering majors at all.
Engineering is something that should be practical or technical only. I envy ND and HND engineering students fr, they’re the ones getting to learn the right way.
Just finished a class now, and despite the lecturer being very very good, I only managed to pick up just a few things. Omo
VĪ retweeted
all metal parts have been ordered!
big thanks to @sendcutsend for sponsoring the build
VĪ retweeted
“akristofuor our day” is back.
let’s go outside in December.
🗓️ | December 28.
presale —> theoutsideworldgh.com
follow @oasisgatherings for more info!
VĪ retweeted
Just reverse engineered this patent 😂 Should I 3D print it?
fun patents i found recently #2
an 1896 patent for a hat that lifts, rotates and puts itself back after u bow. a little clockwork mechanism does the greeting for u.
we have been trying to automate social interaction for a very long time
patents.google.com/patent/US…
ALT Original patent drawing for J. C. Boyle’s Saluting Device, U.S. Patent 556,248, issued March 10, 1896. A cutaway of a bowler hat shows a mechanism above the wearer’s head. A smaller side view shows the spring-driven gears and lifting linkage that raise and rotate the hat.
VĪ retweeted
True story: In 2021, I was there for an interview for a role that wasn't advertised to the public. How did I get there? Because my dad worked there from his early 30s until his retirement. He applied for a role in '92 and was employed in '96. Legit recruitment process. Thread...
Greetings Your Excellency @JDMahama
Please I have one question for you.
Please will the average Ghanaian ever get a chance to work at @thebankofghana without any form of connection?
Like pure meritocracy route?
Thank you.
Concerned citizen.
VĪ retweeted
This is true, and I saw it happen in my own career.
At Shell I started on the technical side, then moved into management, and later into an executive role. Each move changed what I was being paid for.
On the technical side, you're paid for what you know. You understand how things work and you can fix them when they break. If your work is good, that's usually enough.
In management, you still need the technical knowledge, but a lot of your time goes into reviewing other people's work and explaining why something should be done a certain way. You spend more of your day getting people to agree with you.
At the executive level, most of the work is done by others, and your job is to get people to say yes. That could be a client, a board, a partner or your own team. So yes, it turns into a sales job.
The technical years still matter, though. They're a big part of why people trust you when you're the one asking them to back a plan.
The important thing is to know the level you’re at so you can play the game well.
Each level rewards something different, and a lot of people struggle because they keep playing the game of the level they just left.
If you're early in your career, get very good at the technical work, but also start learning how to explain your ideas to people outside your field.
VĪ retweeted
Our likeness to Christ also matures in the midst of difficulties, such as when work seems fruitless, when our strength wanes or when we must start anew. Eventually endurance leaves its mark. Over the years, a gaze can become more merciful, hands more willing to serve and words more capable of comforting. May trials always find us united to Jesus, so that they may help us to grow in His love. #ApostolicJourney #Uruguay
VĪ retweeted
FDM parts of any material are an inappropriate (and i would argue unsafe) choice for any loaded application where they’re critical (supporting humans, a part of a critical safety mechanism, etc.)
this goes beyond anisotropy (which is a challenge, but solvable). you could in theory use non-linear FEA to model the behavior of the part, but it takes a herculean effort and still doesn’t provide the whole point of FEA in the first place for safety critical parts — the predicted failure mode at a specific loading
this isn’t because these parts are hard to model, instead, it’s because FDM is inherently flawed. there are no guaranteed mechanical properties. vendors will list toy values on their websites, but they do not reflect the “real” values and are misleading. real means as built
every car, plane and multi-story commercial building utilizes an important property of materials that manufacturers have gifted to engineers over the centuries, and that’s reliable mechanical properties
aluminum, steel and most metals have a yield or ultimate value by which engineers design to, enabling us to predict when and how parts will fail. FDM does not, and it isn’t the material’s fault, it’s the process’s fault
when i analyze a part, it produces values by which i can meter against mechanical properties of the material. the main problem with FDM is that it does not. the process itself invalidates the underlying mechanical properties
the produced part does not have bulk material properties of the constituent material due to the geometry of the “lay up” of the filament that produces the part. this means an accurate analysis would not be on the “as modeled part”, it would have to be on the “as manufactured” model of the part, or the “sliced” model of the part
and even if you did this: it would still be for moot because the shear resistance of each layer produced is dependent on the orientation and settings of the print and the environment in which it was printed. the properties can vary wildly between prints
i won’t even get into cold flow which is exacerbated by this lamination issue. i still think FDM is great in the limited cases of onesie-twosie fixtures and tooling
here’s a solution: sheet metal is cheap. use a sheet metal substrate to build up the load bearing structure of your parts. they become safe with proper analysis and can also get rid of the temperature independent cold flow problems that hamstring most FDM parts
VĪ retweeted
Geologists and mining professionals on here, I need your help 🙏🏽
I’m working on a final-year project involving remote sensing and GIS mapping of hydrothermal alteration zones for gold exploration.
Does anyone have access to geological maps, hydrothermal alteration datasets, or know someone who could help?
Any location in Ghana works. Non-confidential data only.
A repost could connect me to the right person.🙏🏽
VĪ retweeted
Paula Scher, inside Pentagram’s Work : Identity, Typography and Where AI... youtu.be/UsXYxezOUMU?si=WXG6… via @YouTube
VĪ retweeted
Anduril is an “all in” type of company.
Submarines are one of the most complex systems humans build. Anduril is committing $3.7 billion to help America build more of them.
I’m even more proud now that Anduril is a Flow customer.
The money is going into Arsenal-2, a new shipyard at Sparrows Point in Maryland. It’ll build components and large assemblies for Virginia-class submarines. The Navy has signed a contract worth up to $2.9B, paid out as Anduril actually produces.
Arsenal-2 isn’t expected to open until 2030, but Anduril isn’t waiting. They’re setting up a facility in Orange County, California, to work out their manufacturing methods and train the first team now. The process will have been tested long before the shipyard exists.
The machine that builds the machine is 100x harder, as Elon says. Getting the process right is most of the work. A part for a submarine has to meet its requirements, fit into a boat being assembled at Electric Boat or Newport News, pass inspection, and show up on time.
Anduril is the opposite to the cost plus model. They’re putting their own money to make this work, ahead of schedule. This is the new era for engineering.
VĪ retweeted
Hey Siri, update my CV. 😆
I’ve joined the Governing Council of BlueCrest University College!
New room, new responsibility, and another opportunity to put my experience in strategic communication, positioning and reputation to work.
Let’s get to work! 🥂
