
Founder https://teapawt.org/t.co/Hf9jhBTjw0 | Software Engineer | Quantitative Finance | AI & Data Science | @everpesa_app | Aviation & Car Stuff @SafariDrift
Vector Space
Joined July 2017
- Tweets1.7K
- Following2K
- Followers522
- Likes14.1K
Excited to share my journey on the MIT OpenCourseWare podcast! @MITOCW was instrumental for me to learn data science & programming,which inspired me to launch Everpesa Technologies, a fintech platform for sustainable investments & core-banking in sub-Saharan Africa @everpesa_app
Meet Jerry Vance Anguzu, who turned the Covid-19 lockdown in Uganda into a golden opportunity.
Through MIT OpenCourseWare, Jerry dove into data science and programming. Today, he's the proud founder of @everpesa_app, a platform offering sustainable investments and financial literacy in sub-Saharan Africa. Learn more about Jerry's journey by listening to our podcast: bit.ly/32eXIXR
πππ«π«π² πππ§ππ retweeted
MIT has an entire course on the stuff you should know at your first engineering job
shell, git, debugging, profiling, shipping code
the 2026 version even has agentic coding. all free.
link: missing.csail.mit.edu/
πππ«π«π² πππ§ππ retweeted
One of my favorite results of all time is Sinkhorn's - dating to just 1963. Sinkhorn balancing is an iterative procedure that turns a non-negative matrix into a doubly stochastic matrix by iteratively normalizing its rows, then columns. It's amazing that it works at all - a small miracle that makes many things possible:
It solves optimal transport problems, it's at the core of the PageRank algorithm, it's used in SLAM, in multi-omics alignment, supply chain routing, market design, and more
Itβs such a neat statistical idea when you learn about it. Suddenly it makes so much sense.
Why do we divide by n-1 instead of n when we want to estimate the variance? This is Bessel's correction term, and it made no sense to me back in university when I first learned it.
But I finally understood why!
Got 2 minutes to spare? Here's the explanation!!
1
92
πππ«π«π² πππ§ππ retweeted
Glivenko's theorem: If Ο is a propositional formula, then Ο is a classical tautology if and only if Β¬ Β¬ Ο is an intuitionistic tautology.
The whole Megaspeed affair reminded me of something I saw in Shenzhen, China, last year: NVIDIA H100 chips being openly sold in the market.
This was at the height of US export restrictions, yet the chips were seemingly readily available if you knew where to look. Quite a surreal sight.
Itβll be interesting to hear what $NVDA CEO has to say on the earnings call.
2
110
Interesting story!
A researcher studying flatworms for decades may have discovered some gene variants key to the next blockbuster drug.
This is our story by @bborrell on how @VariantBio is finding new therapies in places everyone else has ignored. corememory.com/p/variant-bioβ¦
1
60
If youβre even remotely interested in Finance, @matt_levine βs Money Stuff - Bloomberg Newsletter is frankly the best think youβll ever come across! π―
PS: Not Investment Advice!
2
124
πππ«π«π² πππ§ππ retweeted
A lot of people in finance tend to get stuck in the middle stages of their careers, particularly making jumps from Sr Associate to VP or from VP to Principal
That is generally around the time when just being a hard worker is no longer enough to climb the ladder. Analyst years are spent just putting in raw hours into the job, following instructions from VPs and MDs, and just putting your head down to learn the technical skillset
At most banks and buyside shops, the transition from Analyst -> Associate -> Senior Associate happens quite naturally. Unless you are explicitly bad at the job, you will probably make it there
However, at the VP and Principal level, the way you navigate office politics and build relationships starts to matter way more than you might think. The job goes from simply being a technical Excel and PowerPoint role to running both internal and external processes.
In banking, you might be juggling between promising deliverables to the client, figuring out what the deliverable should look like, and then communicating both of those effectively to your junior team so they can work towards it
At the same time, you are managing expectations upwards, where the MD or director's goal is to execute on the transaction as fast as possible to maximize value for the client and also so they can move on to chasing the next transaction.
On the buyside, you might be handling external party calls with lawyers and consultants, working with the associate to refine model assumptions and IC memos, and coordinating with the MD to figure out how to best present the deal in front of the committee
Various people around the transaction each have their own incentives, and it is up to the mid-level person to juggle between these and do the execution work required to get to a finish line.
My couple cents of advice for people going through this situation
1. Figure out what your seniors dont want to spend time on, and focus on those tasks as your main priority. The more annoying tasks you can take off their hands, the better rapport you will build up with the decision makers.
2. Build expertise in areas that your group tends to lack. This could be as simple as learning an industry vertical extremely well or knowing how to navigate the financing process for a deal. If you become the go-to person for a specific niche skill set, you essentially become irreplacable to your team.
3. Communication is everything. This is the stage where you are responsible for communicating both down and up the hierarchy. Learn how to communicate in a precise but effective way with both seniors and juniors. Relay the bad news early, make sure workstreams are not getting lost by checking in as often as possible, and figure out what each person's incentive is.
4. Be intellectually honest. Voice your opinion more, and don't just be a process monkey. Make sure you have a coherent view on the transaction when discussing with your seniors or IC. If someone asks what you think of a decision, you should have clear answers behind why. And most importantly, be willing to take accountability for those decisions. People want reliability and decision-making capability, not just someone who can summarize information back to them
5. Take relationship building a lot more seriously. Around this level is when both internal and external relationships matter a lot for getting promoted to the next level. You need people who will vouch for you behind closed doors. Just as importantly, you need outside relationships so you can eventually bring in clients and deals for your firm as a director or MD. Reputation matters a lot for the job, and there is never a time too early to start building it
Jensen drove supply chain brilliantly at Nvidia, but many of those deals were a handshake with Jensen, not a written contract.
Without Jensen, at least some of those critical deals will be lost.
Dylan Patel says Jensen personally talked tiny suppliers into letting Nvidia buy up their supply, and with him gone Google is already begging them for capacity
"Jensen had complete foresight and understood the demand required and went to all of these executives, even of tiny companies where they only generate a few hundred million dollars of revenue but they have an important subcomponent."
"He convinced them to let him buy the supply and build it all out..."
"And now with Jensen gone, these companies are looking around the market and it's like, well, Google's begging for more capacity from the whole supply chain."
"And without Jensen, can they maintain technology advantages given they're only a year ahead of their biggest competitors? It's not like the gap is massive."
πππ«π«π² πππ§ππ retweeted
We started shipping the Nvidia H100 in 2022.
A year ago it rented for $1.96/hour.
Today: $3.10.
+58%, on a chip that was supposed to be obsolete by now.
So much for GPU depreciation.
πππ«π«π² πππ§ππ retweeted
ICYMI: The "Mozart of Math"
Read the full story: ucla.in/4t3gB8s
πππ«π«π² πππ§ππ retweeted
About a decade ago, a guy named Jim Belosic started a hobby project to merge a Honda Accord with a Tesla. The result was something called Teslonda. It was very fast. Possibly too fast.
As he was building, Jim needed access to some decent metal shaping tools. He eventually decided to buy a $750,000 laser cutter and then he figured there were more people like him that also wanted access to good tools. So he started renting his laser cutter to others.
All of this led to the formation of a company called SendCutSend. It's now making about 45 million parts per year for 300,000 customers. In many ways, it has become the US's answer to China when it comes to cutting, shaping and finishing metal parts.
This is the deepest look at the SendCutSend factory and Jim's backstory that has ever been done. We went to his childhood home, met mom and dad (they're lovely) and drank Old Fashioneds at a saloon because that is the type of committment I am willing to put into being a video show host.
Come watch "China, Meet Jim" - youtube.com/watch?v=sC4JvyhBβ¦
πππ«π«π² πππ§ππ retweeted
The FBI is warning the public about FaceStealer, a highly effective social engineering scheme used by North Korean cyber actors to steal cryptocurrency and target the software supply chain. FaceStealer uses imposter versions of video conference platforms, secretly recorded web camera video, and computer intrusions to target individuals working in the cryptocurrency/Web3 industry and open-source software maintainers.
Over the last 18 months, the scheme has resulted in the theft of more than $400 million dollars. Today, the FBI announced the disruption of cloud-based infrastructure used in the scheme, identifying and seizing over 900 webcam videos of victims worldwide. For more on the @FBILosAngeles -led investigation, read: justice.gov/opa/pr/justice-dβ¦
Anyone who owns cryptocurrency, works in the cryptocurrency/Web3 industry, or maintains open-source software repositories is a potential target. The FBI recommends anyone potentially impacted visit fbi.gov/facestealer for additional information and follow the recommendations provided in the FLASH released today: ic3.gov/CSA/2026/261112.pdf
πππ«π«π² πππ§ππ retweeted
I have a major result to share.
Iβve found conclusive forensic evidence that the top 4 Chinese AI labsβDeepSeek, Moonshot, MiniMax and Ant Groupβhave built their newest models on NVIDIAβs 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 patternβa sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isnβt based on any insider information or leaking. Itβs 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 modelsβincluding DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3βwere initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIAβs flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds Iβd recovered, a Blackwell B300 reproduced DeepSeek-V4-Flashβs untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3βs to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flashβs, and changing the seed by one broke the match. (DeepSeekβs 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. Itβs 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/checkpoβ¦.
There is much more Iβve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Binance has completed coverage of assets impacted in the security incident and we have published a new Proof of Reserves snapshot. Once the zero-knowledge proof is generated, users can independently verify that their assets are backed at a minimum of 1:1 - binance.com/en/proof-of-reseβ¦.
To our users and community: thank you for your patience and understanding.
πππ«π«π² πππ§ππ retweeted
Self-hosted Claude Managed Agents is now in private beta.
You can host the agent harness, sandboxes, and session history, in your own cloud or on-premises, and choose where requests to Claude are sent.
πππ«π«π² πππ§ππ retweeted
The graveyards are filled with indispensable newsletters. bloomberg.com/opinion/newsleβ¦






























