Jake Miller (@makerjak) header
Jake Miller (@makerjak) profile picture

Building in healthcare, tech, & retail. Sculpture nerd. Helping scientists at CU Health. Fmr: Founder @ Toggle Health, Humana, Walgreens, Target, LaunchBlueKY

Louisville, KY
Joined November 2009
Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)Photo by Jake Miller (@makerjak)
This is great news. Kids deserve effective meds too and shouldn't have to wait if medically appropriate
At #ObesityWeek and in @NEJM STEP Young offers evidence that effective medical treatment can change the trajectory. This is something worth celebrating, #OW2026 conscienhealth.org/2026/11/1…
1
1
71
Jake Miller retweeted
So many times I have had this interaction with normies where they don't understand what the jackfriks I'm working on as I'm just making random things for fun and messing around with things And to them whatever I'm doing looks like a complete and utter waste of time, it has no function, no direct rewards, it's not making money, it's not successful But so many times those projects evolve or lead to things where there is a massive payout or success after But most people don't have the ability to see that far ahead, and they're risk averse so even if you tell them "I don't know maybe this becomes something, or maybe not, it doesn't really matter" that's simply too uncertain enough for them to accept. Which is of course why they love employment so much, it's stable and certain. The problem with normies is that they will quickly shut this creative play down in themselves and even worse in others. And that way whatever that thing is, it won't ever grow into anything because they killed it before it ever got a chance, proving their point Almost all my projects started by just messing around with no real goal, just playing for fun But it requires some kind of peacefulness because the moment you start putting a requirement of success on to something new, the thing instantly dies Creativity is really magical, and it messes with you if you try control it or expect things from it
A lot of the most successful people I know spent years of their life just bumming around, before they suddenly exploded into action and did something great. I don't think that's a coincidence. noahpinion.blog/p/bumbition
219
162
58
2,910
151,423
Jake Miller retweeted
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included). Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs. Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines. This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!! Enjoy: science.caseyhandmer.com/phy…
359
881
232
6,563
417,893
Jake Miller retweeted
My first column for Works in Progress! Science isn’t a series of lucky accidents. I’ve been investigating stories of scientific discoveries for a while and it’s surprising how much the popular retellings flatten the real stories; some are entirely made up. Alexander Fleming’s discovery of penicillin most likely didn’t come from an open window. His colleagues testified the window was seldom, if ever, opened. The sequence hasn’t been replicated as he described, because penicillin would have needed to be present on the plate before the bacteria grew, as it blocks bacteria from building their cell walls, rather than killing them once they’ve grown. Louis Pasteur’s discovery of the chicken cholera vaccine – the second-ever vaccine – is often said to be accidental too. But historians have found the key experiments were done deliberately by Émile Roux, his assistant at the time, while Pasteur was away for his daughter’s wedding. Wilhelm Röntgen’s accidental discovery of X-rays? The story was just a hypothesis by others later on. Röntgen had his notes burned after his death, so it’s unclear what really happened. He had also long been in a dispute with Philipp Lenard over credit for the discovery, so there are reasons to be suspicious. ** But what really happened? How did Fleming actually discover penicillin? Why burn or hide your notebooks? And most of all, why credit chance events instead of hard work? For the first time, I’ve decided to come up with some half-cooked theories. Read them here and tell me what you think: worksinprogress.co/issue/acc…
23
78
11
453
76,861
Good summary for the uninitiated. I heard only 18% of employers are currently covering, so most people's choice will come down to what is covered, under what conditions (eg do you have to have diabetes), and what you can afford
Replying to @henloitsjoyce
Obligatory, "discuss with your physician" as it depends on individual nuances and goals. Injectable semaglutide is a pure GLP-1 agonist, sold as Ozempic (for diabetics) and Wegovy (for weight loss). There are also oral pill versions, sold as Rybelsus (for diabetics) and Wegovy (for weight loss). There is also orforglipron, an oral small-molecule GLP-1 agonist, sold as Foundayo. Injectable terzepatide is a dual GLP-1 and GIP agonist. It offers more weight loss but also better glycemic control for diabetics because of the GIP arm. Injectable retatrutide is a triple agonist for GLP-1, GIP, and glucagon. Because of the triple agonism, it is generally the most effective. It is not yet approved in the United States yet, but should be soon. These are all strong therapeutics with their own side effects and risk profile. You should always start at the lowest dose and titrate up. If you are not obese or diabetic, i.e. have mild weight loss goals, then you should probably go with the 'mildest' version: the lowest dose of semaglutide (believe it's 2.5mg if injectable). FYI, the injectable versions come in "auto-injectors." You do not see the needle. It's merely a long tube that you press against the injection site. Anyone can use it.
2
171
queen of hearts is unmatched
Here is the output of a task-specific DNN AI model. Grok/ChatGPT/Claude is not designed for this.
2
7
970
So cool
The creator of Sci-hub has finally added semantic search. Side effect: guest queries are public, so you can see what the world is curious about
53
Jake Miller retweeted
Biotech CEOs are usually so poor that they can only have regulatory Affairs.
22
62
7
729
33,964
Jake Miller retweeted
Peer review, 2026 style:
We (partially) verified these results! Using NVIDIA B200/H200s, we reproduced the reported patterns for three models.* We also extended the DeepSeek-V4-Flash test to every weight position across the 42 layers examined and still detected the Blackwell pattern. See the results here: github.com/Institute-for-Pro… This is a super clever approach by @willdepue, and matches reporting that top Chinese AI labs are using Blackwells. * The repo didn’t contain all the inputs needed to test most of the other models
1
3
41
4,484
Jake Miller retweeted
One of the more interesting bits of technical independent journalism (with shades of security research) I’ve ever read, and it has obvious policy implications if accurate.
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.
24
45
1
743
44,482
Sounds legit. Will be interesting to see it externally validated
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.
46
Pretty cool. Current stereotactic devices work by taking pre-surgery MRIs, turning them to 3D images, & using the equivalent of an Xbox Kinect to register the image to the actual skull & instruments (which have metal balls on the end to make them visible to the detector)
Bring a normal robot near an MRI and this happens That’s why the best way to see the brain and the most precise way to operate in it have never worked together We just raised $8M @4DSurgical to change that
2
1
6
492
As soon as you cut into the skull, though, the brain moves. This brain shift makes the surgery much less accurate. Would be cool if this robot could facilitate real-time images so that the planned path can be more accurately mapped to the shifted brain
1
1
32
And I wonder how they will be doing cutting / cauterizing in this scenario. I'm guessing that you can't use an elecrocautery knife / Bovie in an MRI ;)
27
Good use of the 20 image limit
Since apparently we can post up to 20 photos/videos now here are a bunch of mine
44
Jake Miller retweeted
Not quite sure why I like this as much as I do. But I love it. iansilber.com/
1
1
4
469
Jake Miller retweeted
Getting it wrong for many decades. Since the 1950's, pacemaker leads have been placed in the right ventricle. A new randomized trial challenging that dogma found marked superiority of outcomes for pacing from the left bundle branch. nejm.org/doi/full/10.1056/NE…
18
84
6
346
50,656
This is going to be the dominant argument in support of AI adoption going forward
Zach Reitano on why we shouldn't prioritize jobs over lives: "About 40,000 people a year die from car accidents. That is almost the same exact number as the people who die from breast cancer every single year. Driverless cars have proven to be 10, 20 times safer than a human driver. So we basically for certain types of cancer have a cure, and the conversation is not, how do we have Operation Warp Speed for this cure for cancer. It is more about job displacement. It's not that I want to diminish job displacement. It's that we cannot prioritize jobs over lives. We were talking about a cure for cancer and someone said, 'what about the people currently treating cancer patients without the cure? I don't even know that they would be able to.' It was the stupidest idea ever."
2
80
Jake Miller retweeted
I was surprised at how much Medicare/Medicaid/Social Security are now propping up local economies and personal incomes If you look through the data, Medicare/Medicaid have become the biggest sources of income for particularly non-metropolitan areas. Some stats in the data: - Government transfers are now 17.6% of all personal income in the US, up from about 8% in 1970. Medicare and Medicaid are 46% of all government transfer dollars! - 53% of counties get more than 25%+ of their income from government transfers (this covers about 72M people). - Kentucky seems to be the home of the Colonel and government transfer programs. Twelve of the top fifteen counties receiving government transfers are there. Owsley County gets 63% of personal income from transfers, and Medicaid alone is 32% of all income in the county. Those government transfers via Medicaid/Medicare get transmuted alchemically into jobs. Phoebe Putney Health System has 6000+ workers and is the largest employer in southwest Georgia. Dougherty County, where it's based, gets 34% of its personal income from government transfers. In New York we have a program that pays friends and family as caregivers. In 2023 it grew by 57,000 jobs and became 38% of all job growth in the state (they’ve made changes to the program now that have flattened it). We have become a country of healthcare jobs with a propping of government dollars. Is that fine? I think this grand bargain worked for a while. We were still seeing economic productivity, government spending hadn’t yet gotten too crazy, etc. But a few things are happening now that suggest the bargain is falling apart.
157
528
168
2,984
384,163
Jake Miller retweeted
Once you start, you should not stop GLP-1
Stopping GLP-1 treatment linked to twice the risk of heart attack, stroke or death compared with continuing, in people with type 2 diabetes and no cardiovascular disease, according to a new study. Find out more in @LancetRH_Europe: spkl.io/60137wUtS
87
75
12
638
151,388