Nate Schaffner (@thx1766) header
Nate Schaffner (@thx1766) profile picture

Rutgers CS 2017 - Professional Software Developer

Bridgeton, NJ
Joined January 2008
Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)Photo by Nate Schaffner (@thx1766)
Microsoft ~364
1
6
367
Nate Schaffner retweeted
In lieu of flowers, memorial contributions may be made to: Johns Hopkins Alzheimer’s Disease Research Center Fund    alzresearch.org/giving/ Alzheimer’s Foundation of America Caregiver Resources     alzfdn.org/donate/ Photo Credit: Tommy Garcia
6,674
79,672
17,357
345,154
11,998,499
Nate Schaffner retweeted
geoffrey hinton mass-quit google in 2023 and mass-told everyone he was scared of what he built. nobody asked the right question. not "is AI dangerous." the right question is "why does backpropagation work at all." hinton couldn't answer it. neither could lecun. neither could sutskever when he left openai. the three founders of deep learning built a $4.6 trillion industry on an algorithm none of them could mathematically prove should work. until last month. a physicist at ETH zurich published a 41-page proof that backpropagation is a special case of the second law of thermodynamics. not an analogy. a derivation. the math: thermodynamics: dS/dt ≄ 0. entropy of a closed system never decreases. backpropagation: dL/dt ≀ 0. loss of a trained network never increases. one is entropy going up. the other is loss going down. they are the same equation with a sign flip. the proof shows that gradient descent on a neural network is mathematically identical to a physical system falling toward thermal equilibrium. every weight update is a molecule finding its lowest energy state. this is why transformers scale. this is why chinchilla scaling laws look like thermodynamic equations. this is why nobody could prove convergence guarantees - they were using the wrong field. the answer was never in computer science. it was in statistical mechanics. now watch what this means for the frontier labs [Fig. 1] - openai trained GPT-4 at an estimated mass-energy cost of 50,000 MWh. anthropic trained claude 3.5 at approximately 31,000 MWh. google trained gemini ultra at over 60,000 MWh. the ETH proof predicts a theoretical minimum energy for each model's capability level. the equation: E_min = kT × N × ln(1/Δ) k = boltzmann constant. T = operating temperature. N = parameters. Δ = target loss. GPT-4's actual energy usage: 17x above the theoretical minimum. claude: 12x above minimum. gemini: 23x above minimum. anthropic is closest to the thermodynamic floor. not by accident. dario amodei has 3 physicists on the training infrastructure team. openai has zero. the companies that understand this equation will compress 3 years of scaling into 6 months. the companies that don't will spend $10 billion on training runs that a physicist could tell them are 20x less efficient than possible. hinton understood the gradient. he never understood the entropy. the network doesn't learn. it cools. save this
Readers added context they thought people might want to know
The specific 41-page proof by an ETH Zurich physicist published last month does not exist. A 2025 paper explores how backpropagation emerges in physical models of neural networks: arxiv.org/abs/2503.09980 Training energy claims and the exact equivalence described are not supported by it.
53
77
30
574
50,076
My TrueNAS appliance has KVM functionality with a Java client. Excited to reverse engineer something, I decompiled it to make one for iPad! github.com/thx1766/FreeNasKV

1
36
Nate Schaffner retweeted
Who needs VNC when there's browsers.
44
13
3
288
53,365
me: "hey codex, after u get hello world going, can u just tell the slack channel that hello world is working" (silent for 35 minutes) codex: "3,433 tests passed."
18
49
3
1,987
53,311
Nate Schaffner retweeted
I just witnessed the world’s most patient Apple Store employee try to explain to someone for 25 minutes that they don’t make ChatGPT and what software is
35
71
12
6,766
117,465
Nate Schaffner retweeted
This is great news again. Swift on the server rocks! đŸ€˜
Google: “[Swift] has matured into a viable systems and cloud language, pairing Rust-like data-race safety with predictable, reference-counted performance.” Announcing their Swift-based cloud SDK for backend services: cloud.google.com/blog/topics

3
16
164
17,444
Nate Schaffner retweeted
Google: “[Swift] has matured into a viable systems and cloud language, pairing Rust-like data-race safety with predictable, reference-counted performance.” Announcing their Swift-based cloud SDK for backend services: cloud.google.com/blog/topics

32
275
68
1,621
349,531
Nate Schaffner retweeted
Aviation solved AI slop in 1986. Karpathy is telling people to ask their LLM to write in ASD-STE100, a controlled language built for aircraft maintenance manuals. I went down the rabbit hole on where this spec comes from, and the backstory makes the hack even better. Back in the late 1970s, European airlines were handing English manuals to mechanics around the world who mostly didn't speak English as a first language. One ambiguous sentence in a fuel line procedure can get people killed. So the industry built a version of English where ambiguity is banned at the dictionary level. The spec allows roughly 900 approved words. Each word gets one meaning and one part of speech. "Close" can only be a verb, because "close the door" and "the close door" can't be allowed to coexist in a repair manual. "Follow" only means to come after. If you want obey, you write obey. Procedural sentences max out at 20 words. One instruction per sentence. Active voice only. Around 1,200 common words are explicitly banned, each with a mandated replacement. You don't "commence pumping." You start the pump. English has about 170,000 words in current use. Aviation decided mechanics get 900, and planes became safer to maintain because of it. Now flip it to LLMs. Models hedge and synonym-cycle because the training data rewards sounding fluent. STE is 40 years of accumulated rules for stripping exactly that out, written by people whose readers would die if a sentence could be read two ways. It might be the strongest anti-slop prompt in existence, and it's a free PDF.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
71
307
32
3,015
328,919
Nate Schaffner retweeted
ESTOY MUY EN CONTRA DE LA PIRATERÍA. Y NO HAY QUE DIFUNDIR MÉTODOS O SISTEMAS QUE LA FACILITEN. Porque por ahĂ­ cae alguno y te dice "bajate qbittorrent e instalale los plugins". DespuĂ©s viene otro y te tira este link: github.com/qbittorrent/searc
 Y seguro cae oootro gil y te dice "seguĂ­ este tutorial en YouTube": youtu.be/nksLKqotTys?si=NPpX
 Y finalmente no falta el DELINCUENTE que te afirma "es la mejor, la herramienta de bĂșsqueda de qbittorrent una vez configurada con los pasos anteriores es el sistema de bĂșsqueda y descarga mĂĄs eficiente de todo internet en 2026". Por eso como decĂ­a, no descarguen contenido pirateado, no perjudiquen a las corporaciones que tanto bien nos hacen, nadie deberĂ­a apoyar la piraterĂ­a!!!1uno
357
1,682
79
14,944
812,647
I'm sorry 7z can do WHAT
149
836
80
14,561
836,335
Nate Schaffner retweeted
your friend will be obsessed with vibe coding and youll ask them what theyve vibe coded and itll just be this
292
484
96
12,092
1,142,277
Nate Schaffner retweeted
Me and my son have been sneaking out in the middle of the night in our Ewok and Chewbacca costumes just to mess with our neighbor's trail cams.
2,015
13,291
1,236
106,377
2,539,228
"This Gear Solved A Problem Most Engineers Couldn't Handle" At first, engineers thought gears only needed one thing. Teeth that mesh. But when factories started running bigger machines at higher torque and speed, something strange happened. The gears sounded like metal hammers smashing together. Vibration exploded. Bearings wore out faster. Entire systems became unstable. Then came the herringbone gear. Instead of straight teeth slamming together all at once, its angled double-helical design transfers force gradually across the surface. Smoother contact. Less shock. Almost silent operation. But here's the genius part most people miss. Normal helical gears create massive sideways axial thrust that pushes shafts apart under load. Herringbone gears cancel that force completely by using mirrored tooth geometry. No extra thrust bearings. Less wear. Higher load capacity. That's why you'll find them inside marine propulsion systems, heavy industrial gearboxes, and oil-field machinery where failure is not an option. The crazy part? Engineers still avoid them because manufacturing these gears is brutally expensive. Which brings us back to the question... if regular gears already work, why build the impossible one?
68
356
17
2,660
290,378
Nate Schaffner retweeted
The mass of the proton has almost nothing to do with the mass of the things inside it. A proton is made of three quarks. You can look up the mass of those quarks. Two up quarks and one down quark, added together, account for about one percent of the proton's mass. The other ninety nine percent is not matter in any ordinary sense. It is the energy of the gluon field binding the quarks together, churning so violently that Einstein's equation converts it into what we measure as weight. This means that when you step on a scale, almost none of the number is the stuff you are made of. It is the binding energy of fields holding that stuff in place. If you could somehow switch off the strong force for a second, your quarks would still be there, but you would weigh about a kilogram. The rest of you is not substance. It is tension. The calculation that confirms this was not finished until 2008. A team using some of the largest supercomputers then available simulated the interior of a proton from first principles and got the mass right to within a few percent. Before that, we knew the rule worked but could not derive the number. Ninety nine percent of the mass of ordinary matter, the mass of every star and planet and body, was unexplained from the underlying theory until well into this century. You are mostly the sound of something holding itself together.
63
122
18
755
40,838
Nate Schaffner retweeted
đŸłïžTime to face obvious facts. iOS Dev courses are no longer viable income for me. I've been disrupted. I'm looking to join a team working on a cool iOS project (full-time or part-time contract). Remote. RT's are appreciated! DMs open.
92
152
46
804
239,539
Nate Schaffner retweeted
The problem with those “build me my idea for equity” type of people is that they don’t truly believe in their idea, because if they did, they would go to a bank to get a loan to fund their idea or figure it out how to build themselves. They don’t do that because deep down they know their idea will most likely fail, so they don’t want to lose money/time, they prefer to find a developer to lose his time(which is money btw) instead.
i keep getting requests in DMs on linkedin where they want to build an app in exchange of equity and would ask me to sign an idea before even telling me the idea 😭
1
3
471