Загрузка видео...

Не удалось загрузить видео

На главную

5-axis CNC machining – but different... Most 5-axis machines use serial kinematics: stack a rotary A-axis on top of a rotary B-axis, mount that on linear X/Y/Z stages. Each axis carries the weight of everything after it. Heavy and slow. 🤖 The Sprint Z3 uses parallel kinematics: three linear...

JacklouisP's profile picture

Jack 🤖

14,073 subscribers

121,259 просмотров • 11 месяцев назад •via X (Twitter)

ИскусствоОбразованиеНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

5-Axis CNC machining of turbocharger impellers sits in the same league as aerospace manufacturing, precision engineering pushed to physical limits. These machines run with positional accuracy in the 1-5 micron range, spindle speeds of 20,000-40,000 RPM, and continuous 5-axis interpolation to carve complex blade geometry without collision, vibration, or thermal drift during cutting of nickel alloys and titanium. A production-grade 5-axis system for impellers typically costs $300,000 to $3 million+ The real constraint is not motion, but stability, suppressing chatter, tool deflection, and microscopic surface defects that would quietly destroy efficiency at 100,000+ RPM operating cycles. Without this class of machining, modern turbocharged engines, high-efficiency industrial compressors, and compact propulsion systems simply would not exist.
0:10

5-Axis CNC machining of turbocharger impellers sits in the same league as aerospace manufacturing, precision engineering pushed to physical limits. These machines run with positional accuracy in the 1-5 micron range, spindle speeds of 20,000-40,000 RPM, and continuous 5-axis interpolation to carve complex blade geometry without collision, vibration, or thermal drift during cutting of nickel alloys and titanium. A production-grade 5-axis system for impellers typically costs $300,000 to $3 million+ The real constraint is not motion, but stability, suppressing chatter, tool deflection, and microscopic surface defects that would quietly destroy efficiency at 100,000+ RPM operating cycles. Without this class of machining, modern turbocharged engines, high-efficiency industrial compressors, and compact propulsion systems simply would not exist.

Ammanichanda

48,750 просмотров • 3 месяцев назад

A machine like this can cost $500,000 to well over $1 million to make parts that may be worth only $10-15 This is the real manufacturing story. This INDEX six-spindle automatic carries 6 motorised spindles, up to 12 CNC tool carriers, operates at 8,000+ rpm, and weighs 7.2 tonnes. Different operations happen simultaneously as the spindle drum indexes each workpiece from station to station. A real scale production example produced a precision component in 11 seconds, versus 38 seconds on a conventional single-spindle lathe. That's roughly 327 parts per hour before downtime. But the machine is only the hardware. Tool geometry, CNC programs, cutting parameters, spindle synchronisation, tooling, bar feeding, chip evacuation, coolant, inspection and collision checked simulation all have to be engineered around the exact part. That is what it takes to integrate this machine into a factory workflow. This accumulated capability is what kept Germany and Japan at the pinnacle of machine-tool manufacturing for decades almost unchallenged. The advantage wasn't just building the hardware, but knowing how to program, tool and integrate these machines for thousands of different manufacturing requirements globally. China has now built much of that ecosystem at extraordinary scale, machines, controls, tooling, software, automation and integration. Its huge domestic manufacturing base has accelerated that learning curve dramatically. Today, Chinese manufacturers can increasingly offer sophisticated CNC and multi-spindle systems at 30-40% lower total costs in most applications, putting serious and relentless price pressure on German and Japanese builders. China produced 37% of the world's machine tools in 2025, compared with 12% for Germany and 10% for Japan. These are the machines that make the machines and ultimately determine how much an economy can manufacture. China is the biggest player as of now and growing faster than anyone else in manufacturing high-end machining tools. Source, Daniel Jansson
0:16

A machine like this can cost $500,000 to well over $1 million to make parts that may be worth only $10-15 This is the real manufacturing story. This INDEX six-spindle automatic carries 6 motorised spindles, up to 12 CNC tool carriers, operates at 8,000+ rpm, and weighs 7.2 tonnes. Different operations happen simultaneously as the spindle drum indexes each workpiece from station to station. A real scale production example produced a precision component in 11 seconds, versus 38 seconds on a conventional single-spindle lathe. That's roughly 327 parts per hour before downtime. But the machine is only the hardware. Tool geometry, CNC programs, cutting parameters, spindle synchronisation, tooling, bar feeding, chip evacuation, coolant, inspection and collision checked simulation all have to be engineered around the exact part. That is what it takes to integrate this machine into a factory workflow. This accumulated capability is what kept Germany and Japan at the pinnacle of machine-tool manufacturing for decades almost unchallenged. The advantage wasn't just building the hardware, but knowing how to program, tool and integrate these machines for thousands of different manufacturing requirements globally. China has now built much of that ecosystem at extraordinary scale, machines, controls, tooling, software, automation and integration. Its huge domestic manufacturing base has accelerated that learning curve dramatically. Today, Chinese manufacturers can increasingly offer sophisticated CNC and multi-spindle systems at 30-40% lower total costs in most applications, putting serious and relentless price pressure on German and Japanese builders. China produced 37% of the world's machine tools in 2025, compared with 12% for Germany and 10% for Japan. These are the machines that make the machines and ultimately determine how much an economy can manufacture. China is the biggest player as of now and growing faster than anyone else in manufacturing high-end machining tools. Source, Daniel Jansson

Ammanichanda

38,359 просмотров • 1 месяц назад

AI has had exactly two scaling axes that worked so far, and the second one is starting to look finite too the first one was pretraining: with scaling parameters and data, we got world knowledge (i.e. ChatGPT had read enough to know things), but it started saturating a while ago the second one was RL, and people had been doing RL the whole time before that: RLHF is RL but it never scaled far because it was trying to control the exact output, which tokens come out, how the text reads, but you can only push that so far before you’re just polishing RLVR dropped that constraint: giving the model a task, then checking whether the final answer is right, and ignoring everything in between -- so the model does whatever it wants in the middle and only the endpoint gets graded, and that’s much closer to actual RL and it’s what bought us planning and reasoning (arguably, tool use sits around 2.5 on this list -- while useful, it's not a different kind of thing) so one axis gave knowledge, the other gave reasoning, and both of them are one model working alone the next axis is how many models you can get working on the same problem, which is a different kind of axis than the previous two we know that multi-agent RL has always been the harder problem: I spent years in that literature and the gap between single-agent and multi-agent is definitely not incremental -- it’s a whole different class of difficulty! which is also why the derivatives are steep at the start, nobody has picked the easy wins yet... and the thing that gates this multi-agent coordination is communication: models can only coordinate as well as they can exchange information, and right now they do that by writing sentences to each other imagine what could we possibly achieve if we properly open that third axis development by letting models to exchange information in their native "language" without loosing any computational data that they produce during inference
0:14

AI has had exactly two scaling axes that worked so far, and the second one is starting to look finite too the first one was pretraining: with scaling parameters and data, we got world knowledge (i.e. ChatGPT had read enough to know things), but it started saturating a while ago the second one was RL, and people had been doing RL the whole time before that: RLHF is RL but it never scaled far because it was trying to control the exact output, which tokens come out, how the text reads, but you can only push that so far before you’re just polishing RLVR dropped that constraint: giving the model a task, then checking whether the final answer is right, and ignoring everything in between -- so the model does whatever it wants in the middle and only the endpoint gets graded, and that’s much closer to actual RL and it’s what bought us planning and reasoning (arguably, tool use sits around 2.5 on this list -- while useful, it's not a different kind of thing) so one axis gave knowledge, the other gave reasoning, and both of them are one model working alone the next axis is how many models you can get working on the same problem, which is a different kind of axis than the previous two we know that multi-agent RL has always been the harder problem: I spent years in that literature and the gap between single-agent and multi-agent is definitely not incremental -- it’s a whole different class of difficulty! which is also why the derivatives are steep at the start, nobody has picked the easy wins yet... and the thing that gates this multi-agent coordination is communication: models can only coordinate as well as they can exchange information, and right now they do that by writing sentences to each other imagine what could we possibly achieve if we properly open that third axis development by letting models to exchange information in their native "language" without loosing any computational data that they produce during inference

Sasha Malysheva

15,053 просмотров • 1 месяц назад

Bernoulli's Principle It’s a key idea in fluid dynamics named after Daniel Bernoulli. In simple terms: For a fluid flowing smoothly, an increase in the fluid’s speed happens at the same time as a decrease in pressure or a decrease in the fluid’s potential energy. Where: - P = pressure in the fluid - ρ = fluid density - v = flow velocity - g = acceleration due to gravity - h = height above a reference point Bernoulli's Principle basically says that for a fluid that's flowing smoothly, when its speed goes up, its pressure goes down. Think of it like this: fast-moving air or water pushes less sideways than slow-moving air or water. The energy in the fluid stays the same overall, but it shifts between motion, pressure, and height. That's why an airplane wing creates lift. The air moves faster over the curved top of the wing than underneath, so the pressure on top is lower. The higher pressure below pushes the wing up.
0:10

Bernoulli's Principle It’s a key idea in fluid dynamics named after Daniel Bernoulli. In simple terms: For a fluid flowing smoothly, an increase in the fluid’s speed happens at the same time as a decrease in pressure or a decrease in the fluid’s potential energy. Where: - P = pressure in the fluid - ρ = fluid density - v = flow velocity - g = acceleration due to gravity - h = height above a reference point Bernoulli's Principle basically says that for a fluid that's flowing smoothly, when its speed goes up, its pressure goes down. Think of it like this: fast-moving air or water pushes less sideways than slow-moving air or water. The energy in the fluid stays the same overall, but it shifts between motion, pressure, and height. That's why an airplane wing creates lift. The air moves faster over the curved top of the wing than underneath, so the pressure on top is lower. The higher pressure below pushes the wing up.

Sam Mathverse

27,791 просмотров • 4 месяцев назад

EVERYONE'S PROMPTING FOR CINEMATIC PERFECTION. THE FUNNIEST STUFF COMES FROM LOCKING DOWN THE PHYSICS INSTEAD. Fixed camera axis. No slow-mo, no replays, no cutaways. One object that moves in one direction, never stops, never reverses, never breaks the geometry of the set. That's the whole trick behind making a gameshow challenge clip read like real broadcast footage instead of an AI fever dream. Here's a full prompt built that way. Copy it, swap the character and challenge, works for almost any physical-comedy bit: SETUP: Real TV camera coverage, daytime outdoor set, real-time only - no slow motion, no replays, no freeze frames. Reference images lock venue layout, character identity, starting positions, and target pose. POINTS A & B: Fixed and never reversed. The obstacle always travels A→B, makes contact, keeps pushing - it only stops the instant she hits the water. OBJECTIVE: Match a cutout shape before the obstacle arrives. She nails the pose. The cutout is just slightly too small. That's the entire joke - not a missed pose, a missed size. POSE LOCK: Orientation matters as much as the shape. Once she's in position, back stays toward A, face stays toward B -no last-second turn to "check" on the obstacle. That's the detail that keeps it feeling like a real stunt instead of a render guessing at anatomy. TIMELINE: Beat it out second by second - buzzer, pose, contact, push, fall, reaction. No dialogue on the impact beat. Just the splash, then a second of silence before she surfaces. CAMERA: Wide ENG shot to establish the full space and the one-way geometry. Medium for the pose. Wide again for contact, so the full body and the full cutout read in one frame. Cuts never imply the obstacle moved backward. ENDING: No shouting, no crying, no laughing, no line. Just a wet, unbothered stare into the lens. That's the punchline. A few things worth noticing about why it's built this way: Fixing A and B before anything else is what keeps a multi-cut sequence from ever "teleporting" the obstacle. Direction discipline does more work here than any camera instruction. The pose isn't a vibe (do a handstand) - it's a construction: three contact points, hips up, legs locked wide, back toward the danger. That specificity is what stops the model from defaulting to a normal handstand. The failure condition is spelled out as harshly as the success condition. She fails because of size, not form removes the model's favorite shortcut - quietly rewriting the ending into a clean pass-through. Comment PROMPT and I’ll send it to you.
0:15

EVERYONE'S PROMPTING FOR CINEMATIC PERFECTION. THE FUNNIEST STUFF COMES FROM LOCKING DOWN THE PHYSICS INSTEAD. Fixed camera axis. No slow-mo, no replays, no cutaways. One object that moves in one direction, never stops, never reverses, never breaks the geometry of the set. That's the whole trick behind making a gameshow challenge clip read like real broadcast footage instead of an AI fever dream. Here's a full prompt built that way. Copy it, swap the character and challenge, works for almost any physical-comedy bit: SETUP: Real TV camera coverage, daytime outdoor set, real-time only - no slow motion, no replays, no freeze frames. Reference images lock venue layout, character identity, starting positions, and target pose. POINTS A & B: Fixed and never reversed. The obstacle always travels A→B, makes contact, keeps pushing - it only stops the instant she hits the water. OBJECTIVE: Match a cutout shape before the obstacle arrives. She nails the pose. The cutout is just slightly too small. That's the entire joke - not a missed pose, a missed size. POSE LOCK: Orientation matters as much as the shape. Once she's in position, back stays toward A, face stays toward B -no last-second turn to "check" on the obstacle. That's the detail that keeps it feeling like a real stunt instead of a render guessing at anatomy. TIMELINE: Beat it out second by second - buzzer, pose, contact, push, fall, reaction. No dialogue on the impact beat. Just the splash, then a second of silence before she surfaces. CAMERA: Wide ENG shot to establish the full space and the one-way geometry. Medium for the pose. Wide again for contact, so the full body and the full cutout read in one frame. Cuts never imply the obstacle moved backward. ENDING: No shouting, no crying, no laughing, no line. Just a wet, unbothered stare into the lens. That's the punchline. A few things worth noticing about why it's built this way: Fixing A and B before anything else is what keeps a multi-cut sequence from ever "teleporting" the obstacle. Direction discipline does more work here than any camera instruction. The pose isn't a vibe (do a handstand) - it's a construction: three contact points, hips up, legs locked wide, back toward the danger. That specificity is what stops the model from defaulting to a normal handstand. The failure condition is spelled out as harshly as the success condition. She fails because of size, not form removes the model's favorite shortcut - quietly rewriting the ending into a clean pass-through. Comment PROMPT and I’ll send it to you.

Nexlow

29,192 просмотров • 1 месяц назад

EARTH HAS A COSMIC CLOCK. Earth's climate doesn't stay exactly the same forever. Over thousands to hundreds of thousands of years, Earth's orbit and orientation in space slowly change. These changes are called the Milankovitch cycles. There are three major cycles: ECCENTRICITY Earth's orbital shape changes from more circular to more elliptical over roughly 100,000 years. OBLIQUITY Earth's axial tilt varies between about 22.1° and 24.5° over roughly 41,000 years. PRECESSION Earth's rotational axis slowly wobbles like a spinning top over roughly 26,000 years. Together, these cycles change where and when sunlight is distributed across Earth, particularly at high latitudes. Over geological timescales, these orbital changes are an important factor in Earth's long-term climate variations and the timing of many glacial and interglacial periods. Think about the scale: Earth's orientation changes over tens of thousands of years... while a human lifetime is only a tiny moment on that clock. The planet isn't just traveling around the Sun. Its entire orbital geometry is slowly changing too. Which cycle surprised you most: eccentricity, obliquity, or precession?
0:10

EARTH HAS A COSMIC CLOCK. Earth's climate doesn't stay exactly the same forever. Over thousands to hundreds of thousands of years, Earth's orbit and orientation in space slowly change. These changes are called the Milankovitch cycles. There are three major cycles: ECCENTRICITY Earth's orbital shape changes from more circular to more elliptical over roughly 100,000 years. OBLIQUITY Earth's axial tilt varies between about 22.1° and 24.5° over roughly 41,000 years. PRECESSION Earth's rotational axis slowly wobbles like a spinning top over roughly 26,000 years. Together, these cycles change where and when sunlight is distributed across Earth, particularly at high latitudes. Over geological timescales, these orbital changes are an important factor in Earth's long-term climate variations and the timing of many glacial and interglacial periods. Think about the scale: Earth's orientation changes over tens of thousands of years... while a human lifetime is only a tiny moment on that clock. The planet isn't just traveling around the Sun. Its entire orbital geometry is slowly changing too. Which cycle surprised you most: eccentricity, obliquity, or precession?

cosmic pulse

50,341 просмотров • 13 дней назад

Unsure the level of concern at the moment (1.09 FIP + 27.5 K-BB%) but I think it's worth nothing that Sanchez's FB velo is down: First 3 G (2026): • 94.4 MPH (-2.1) • 94.3 MPH (-1.5) • 94.6 MPH (-0.6) *(L: '26, R: '25)* Last year: • Sanchez was a bit more efficient w/ his hip + shoulder separation • His body was more linear, helping that center of mass move correctly. • Better lower half allowing him to keep that pelvis a bit more linear -- which is likely result of the ribcage is stacked over his pelvis + sitting in that load longer on foot strike (??) This year: • Energy is leaking from that back knee into rotation • Upper + lower working w/ each other more and less of working against one another (arm doing more than it should) • His pelvis is noticeably tipping down through rotation (not a direct cue but more of a result of less lower half activation) Angles/WBC fatigue could also be at play here as well -- but assuming they're not could be some things to clean up
0:11

Unsure the level of concern at the moment (1.09 FIP + 27.5 K-BB%) but I think it's worth nothing that Sanchez's FB velo is down: First 3 G (2026): • 94.4 MPH (-2.1) • 94.3 MPH (-1.5) • 94.6 MPH (-0.6) *(L: '26, R: '25)* Last year: • Sanchez was a bit more efficient w/ his hip + shoulder separation • His body was more linear, helping that center of mass move correctly. • Better lower half allowing him to keep that pelvis a bit more linear -- which is likely result of the ribcage is stacked over his pelvis + sitting in that load longer on foot strike (??) This year: • Energy is leaking from that back knee into rotation • Upper + lower working w/ each other more and less of working against one another (arm doing more than it should) • His pelvis is noticeably tipping down through rotation (not a direct cue but more of a result of less lower half activation) Angles/WBC fatigue could also be at play here as well -- but assuming they're not could be some things to clean up

Joseph Cammisa

43,317 просмотров • 6 месяцев назад

A GIRL WALKED OFF THE TOP OF A WATERSLIDE, THREW TWO FLIPS DOWN THE FACE OF IT, AND LANDED IN THE POOL 20,000 likes None of it is generated. That is exactly why it is worth eleven seconds, because your feed has spent a year training you to assume the opposite run it as a test. here is what is in the frame that no model reliably puts there yet: → the water. spray leaves her body at the right angle and volume for her speed, and the sheet on the slide deforms under her weight instead of flowing past her → the rotation conserves. she tucks and the spin accelerates, she opens and it slows. angular momentum is a constraint, and generators approximate it rather than obey it → the crowd reacts late. people on the platform turn after she has already gone, because they are reacting, not choreographed → the camera operator loses her. the frame lags the subject and catches up. that is a mistake, and mistakes are the expensive thing to fake → and the landing is ugly. real impacts are. generated ones almost always resolve too cleanly none of that is a checklist you can memorise, and pretending otherwise is how people get caught. every one of those tells has a shelf life. three of them were already unreliable a year ago what does not expire is where you look. continuity, physics, and mistakes - the three places a model has to simulate a system instead of reproducing a surface. surfaces are solved. systems are not, yet and the reason this gets more useful every month is not fakes. it is that real footage is starting to get accused. the cost of being wrong now runs in both directions the fastest way to calibrate your own eye is to make one yourself. image-to-video from a still, one line about the motion - Picsart runs it from a phone. you start spotting the tells about ten minutes after you have made your own this one is real. the fact that you had to check is the actual story
0:11

A GIRL WALKED OFF THE TOP OF A WATERSLIDE, THREW TWO FLIPS DOWN THE FACE OF IT, AND LANDED IN THE POOL 20,000 likes None of it is generated. That is exactly why it is worth eleven seconds, because your feed has spent a year training you to assume the opposite run it as a test. here is what is in the frame that no model reliably puts there yet: → the water. spray leaves her body at the right angle and volume for her speed, and the sheet on the slide deforms under her weight instead of flowing past her → the rotation conserves. she tucks and the spin accelerates, she opens and it slows. angular momentum is a constraint, and generators approximate it rather than obey it → the crowd reacts late. people on the platform turn after she has already gone, because they are reacting, not choreographed → the camera operator loses her. the frame lags the subject and catches up. that is a mistake, and mistakes are the expensive thing to fake → and the landing is ugly. real impacts are. generated ones almost always resolve too cleanly none of that is a checklist you can memorise, and pretending otherwise is how people get caught. every one of those tells has a shelf life. three of them were already unreliable a year ago what does not expire is where you look. continuity, physics, and mistakes - the three places a model has to simulate a system instead of reproducing a surface. surfaces are solved. systems are not, yet and the reason this gets more useful every month is not fakes. it is that real footage is starting to get accused. the cost of being wrong now runs in both directions the fastest way to calibrate your own eye is to make one yourself. image-to-video from a still, one line about the motion - Picsart runs it from a phone. you start spotting the tells about ten minutes after you have made your own this one is real. the fact that you had to check is the actual story

Valentin

600,828 просмотров • 14 дней назад

There has been a lot of hand wringing on the appropriate valuation of SpaceX. Some large institutions believe SpaceX can only be valued at half what the market seems to be willing to pay for it. Others are claiming it has 15X appreciation ahead of it. Almost all of this difference of opinion comes down to how comfortable you are modeling beyond 2030 and what valuation method you use. 2030 valuation using a traditional Gordan DCF produces a very different result than a 2040 EV/EBITDA Multiple. Both have pros and cons. Most analysts don’t really discuss this and lead with a headline number. We are very comfortable modeling out to 2040, as large portions of what SpaceX is proposing is real world infrastructure, which provides modelable physics constraints to anchor against. The analysis we released today explores this in-depth, its open to the public all the way through IPO. I highly encourage you check it out prior to then. We’ve run 5,000 monte carlo runs across 500 variables (real number, even though it sounds fake) and three valuation methods. This video is of a 3D cloud chart showing every simulation outcome expected in valuation output across two of the most impactful variables to the model when using an EV/EBITDA multiple from 2026 to 2040. The horizontal axis is the steepness of the orbital data center demand S-curve. The vertical axis is the rate at which chip compute efficiency becomes cheaper. Each of the 5,000 dots is one simulated future; green dots are the ones where SpaceX's 2040 value clears the $1.77T IPO line, over time. Under EV/EBITDA valuation through 2040, 96% of our simulated futures clear the expected IPO price once the bell rings Friday. We aren’t publishing this publicly to tell investors what the stock is worth, we’re publishing this to help investors understand the world of outcomes, what the fundamentals suggest through 2040, and what frankly most analysis simply won’t share. SpaceX is a generational company working on long term infrastructure harnessing a domain no one has been able to tap in so far: space. It deserves doing the work as an investor. because this in not financial advice. The cleanest way to hold SpaceX is a bond stapled to a call option (AI-Compute); Starlink is the bond, the near term SatCom annuity that funds the next flywheel. Understand the world of outcomes and take your position accordingly. Comparables and P/E won't take you far enough.
0:14

There has been a lot of hand wringing on the appropriate valuation of SpaceX. Some large institutions believe SpaceX can only be valued at half what the market seems to be willing to pay for it. Others are claiming it has 15X appreciation ahead of it. Almost all of this difference of opinion comes down to how comfortable you are modeling beyond 2030 and what valuation method you use. 2030 valuation using a traditional Gordan DCF produces a very different result than a 2040 EV/EBITDA Multiple. Both have pros and cons. Most analysts don’t really discuss this and lead with a headline number. We are very comfortable modeling out to 2040, as large portions of what SpaceX is proposing is real world infrastructure, which provides modelable physics constraints to anchor against. The analysis we released today explores this in-depth, its open to the public all the way through IPO. I highly encourage you check it out prior to then. We’ve run 5,000 monte carlo runs across 500 variables (real number, even though it sounds fake) and three valuation methods. This video is of a 3D cloud chart showing every simulation outcome expected in valuation output across two of the most impactful variables to the model when using an EV/EBITDA multiple from 2026 to 2040. The horizontal axis is the steepness of the orbital data center demand S-curve. The vertical axis is the rate at which chip compute efficiency becomes cheaper. Each of the 5,000 dots is one simulated future; green dots are the ones where SpaceX's 2040 value clears the $1.77T IPO line, over time. Under EV/EBITDA valuation through 2040, 96% of our simulated futures clear the expected IPO price once the bell rings Friday. We aren’t publishing this publicly to tell investors what the stock is worth, we’re publishing this to help investors understand the world of outcomes, what the fundamentals suggest through 2040, and what frankly most analysis simply won’t share. SpaceX is a generational company working on long term infrastructure harnessing a domain no one has been able to tap in so far: space. It deserves doing the work as an investor. because this in not financial advice. The cleanest way to hold SpaceX is a bond stapled to a call option (AI-Compute); Starlink is the bond, the near term SatCom annuity that funds the next flywheel. Understand the world of outcomes and take your position accordingly. Comparables and P/E won't take you far enough.

Aaron Burnett

1,521,934 просмотров • 4 месяцев назад

The IRGC are now hiding their remaining precious launchers dressed up as civilian trucks and inside highway tunnels—a sort of Hamas-inspired strategy. Getting the last 10% was always going to be the hardest part. But how effective are they anyway? In the four rounds of “True Promise” attacks, they have fired around 1,500 ballistic missiles at Israel and continue to fire 5 to 10 per day. This is well over half their stock, even accounting for rebuilding. With this, they have managed to kill 48 Israelis and cause a few billion in property damage. If they have 500 missiles and 150 launchers left—it is probably less—they will be able to keep “resisting” for another month or so and kill a few more Israelis at random, but then it’s all over. Their grand “Axis of Resistance” will no longer be a strategic threat to Israel, just an irritant to be occasionally swatted at. In truth, this has been the case for a while.
0:10

The IRGC are now hiding their remaining precious launchers dressed up as civilian trucks and inside highway tunnels—a sort of Hamas-inspired strategy. Getting the last 10% was always going to be the hardest part. But how effective are they anyway? In the four rounds of “True Promise” attacks, they have fired around 1,500 ballistic missiles at Israel and continue to fire 5 to 10 per day. This is well over half their stock, even accounting for rebuilding. With this, they have managed to kill 48 Israelis and cause a few billion in property damage. If they have 500 missiles and 150 launchers left—it is probably less—they will be able to keep “resisting” for another month or so and kill a few more Israelis at random, but then it’s all over. Their grand “Axis of Resistance” will no longer be a strategic threat to Israel, just an irritant to be occasionally swatted at. In truth, this has been the case for a while.

Saul Sadka

152,697 просмотров • 6 месяцев назад

🇮🇷🇷🇺🇺🇦 Russia's knockoff of Iran's own drone is apparently so good that Tehran came shopping. The Geran started as a copy of Iran's Shahed, and Western security officials say Tehran asked Moscow this year for the advanced versions to use in its war with Israel and the U.S. Ukraine is finding out why the hard way. Kyiv is intercepting only about 60% of the new jet-powered Geran-4 and Geran-5, compared with the 90 to 95% it routinely manages against older models. The Geran-5 flies at up to 600 km/h with a 90 kg warhead and a 1,000 km range, which puts it closer to a cruise missile than a drone. Newer versions also carry thermal cameras and AI-powered targeting. Russia mass-produces them and fixes problems on the fly instead of going through years of testing, and they still cost a fraction of the cruise missiles Moscow used to rely on. Zelensky says 67 Ukrainian companies are working on ways to shoot them down. For most of this war, Ukraine was the one setting the pace on drones, and now it's the one playing catch-up. Source: FT / Writer: Julie
0:15

🇮🇷🇷🇺🇺🇦 Russia's knockoff of Iran's own drone is apparently so good that Tehran came shopping. The Geran started as a copy of Iran's Shahed, and Western security officials say Tehran asked Moscow this year for the advanced versions to use in its war with Israel and the U.S. Ukraine is finding out why the hard way. Kyiv is intercepting only about 60% of the new jet-powered Geran-4 and Geran-5, compared with the 90 to 95% it routinely manages against older models. The Geran-5 flies at up to 600 km/h with a 90 kg warhead and a 1,000 km range, which puts it closer to a cruise missile than a drone. Newer versions also carry thermal cameras and AI-powered targeting. Russia mass-produces them and fixes problems on the fly instead of going through years of testing, and they still cost a fraction of the cruise missiles Moscow used to rely on. Zelensky says 67 Ukrainian companies are working on ways to shoot them down. For most of this war, Ukraine was the one setting the pace on drones, and now it's the one playing catch-up. Source: FT / Writer: Julie

Mario Nawfal

21,238 просмотров • 1 день назад

$640,000 of humanoid robots died in 6 seconds because nobody ever shipped the code for running away. 9 men. Wooden handles. 40 machines that kept walking into the swing. That's the story of this clip. Not the violence. The gait. The column keeps walking because walking is all that stack does. Here's what's actually inside one of those bodies. The legs. 12 of the 43 joints live below the waist. Each knee runs a harmonic-drive or planetary actuator — a $600 to $2,000 part, sealed, non-serviceable in the field. One clean hit on a knee housing ends the unit. Not the software. The gearbox. The head. On most platforms that shell holds a depth camera and a LiDAR puck - around $250 for a RealSense, $500 to $700 for the LiDAR. Take the head off and the body doesn't die. It keeps balancing on IMU and joint encoders alone. That's why decapitated units in the clip stay upright for another 2 steps. The controller. Balance runs at 500 to 1,000 Hz. Perception runs at 30 frames a second. Those are different worlds. The balance loop is fast enough to catch a shove; the perception loop is slow, and it was trained on floors, boxes, doors, and stairs. A man sprinting in from 4 meters with a wooden handle isn't in the dataset. There's no class for it. Fall recovery exists. Every serious platform has it - G1 stands itself up, Atlas rolls and rises. Threat response exists on nothing that ships. Nobody sells it. Nobody's asked for it. Now the money. 40 units at $16,000 is $640,000 in hardware. 6 seconds of swinging takes out 60% of it. Actuators, shells, sensor stacks. The batteries - 9,000 mAh, 2 to 4 hours of walk time - are the part you don't want cracked open on a wet street. And the law is a blank page. In the US, smashing one is criminal mischief: property damage, valued at replacement cost. Same statute as a mailbox. No jurisdiction on earth has a separate line for it. The 4 known Spot attacks since 2019 all closed as vandalism. So the brief for the next generation writes itself. Not weapons, not defense. Cheaper knees, ruggedized shells, and a perception model that has finally seen a person running at it. 40 units, 43 joints each, 1,720 things to break. They didn't fail to fight back. Nobody shipped that feature.
0:14

$640,000 of humanoid robots died in 6 seconds because nobody ever shipped the code for running away. 9 men. Wooden handles. 40 machines that kept walking into the swing. That's the story of this clip. Not the violence. The gait. The column keeps walking because walking is all that stack does. Here's what's actually inside one of those bodies. The legs. 12 of the 43 joints live below the waist. Each knee runs a harmonic-drive or planetary actuator — a $600 to $2,000 part, sealed, non-serviceable in the field. One clean hit on a knee housing ends the unit. Not the software. The gearbox. The head. On most platforms that shell holds a depth camera and a LiDAR puck - around $250 for a RealSense, $500 to $700 for the LiDAR. Take the head off and the body doesn't die. It keeps balancing on IMU and joint encoders alone. That's why decapitated units in the clip stay upright for another 2 steps. The controller. Balance runs at 500 to 1,000 Hz. Perception runs at 30 frames a second. Those are different worlds. The balance loop is fast enough to catch a shove; the perception loop is slow, and it was trained on floors, boxes, doors, and stairs. A man sprinting in from 4 meters with a wooden handle isn't in the dataset. There's no class for it. Fall recovery exists. Every serious platform has it - G1 stands itself up, Atlas rolls and rises. Threat response exists on nothing that ships. Nobody sells it. Nobody's asked for it. Now the money. 40 units at $16,000 is $640,000 in hardware. 6 seconds of swinging takes out 60% of it. Actuators, shells, sensor stacks. The batteries - 9,000 mAh, 2 to 4 hours of walk time - are the part you don't want cracked open on a wet street. And the law is a blank page. In the US, smashing one is criminal mischief: property damage, valued at replacement cost. Same statute as a mailbox. No jurisdiction on earth has a separate line for it. The 4 known Spot attacks since 2019 all closed as vandalism. So the brief for the next generation writes itself. Not weapons, not defense. Cheaper knees, ruggedized shells, and a perception model that has finally seen a person running at it. 40 units, 43 joints each, 1,720 things to break. They didn't fail to fight back. Nobody shipped that feature.

HodlReaper

131,075 просмотров • 1 месяц назад

"Lift lighter as you get older. Protect your joints." The standard advice. The wrong advice. What heavy lifting actually does for an ageing body: - Builds tendon stiffness, which protects joints rather than wearing them - Maintains the high threshold motor units that are first to disappear with age - Stimulates bone density, the thing osteoporosis attacks - Develops the strength reserve that keeps you independent at 80 What high reps do for an ageing body: - Generate more total fatigue per session - Take longer to recover from (a problem that compounds with age) - Do nothing for tendon stiffness - Do nothing for fast twitch preservation - Make you tired without making you strong Recovery capacity declines with age. The worst thing you can do is pick the more fatiguing option and call it gentle. The actual protocol: - Heavy enough to challenge the muscle (4-6 reps, near failure) - Machines for most of the work (less spinal load, less balance demand, more stimulus to the muscle, less risk) - Skip the 1RM testing, skip the maxing out on barbell squats and deadlifts - That isn't where the magic was anyway You don't need to max out a barbell to lift heavy. You need to challenge the muscle with a load it respects. The leg press at 200kg does this. The hack squat does this. The chest-supported row does this. None of them are ego lifts. All of them are heavy. Lifting lighter to "protect" your joints is how you arrive at 75 unable to stand up from a chair. Lifting heavy on machines is how you arrive at 75 carrying your own shopping.
0:13

"Lift lighter as you get older. Protect your joints." The standard advice. The wrong advice. What heavy lifting actually does for an ageing body: - Builds tendon stiffness, which protects joints rather than wearing them - Maintains the high threshold motor units that are first to disappear with age - Stimulates bone density, the thing osteoporosis attacks - Develops the strength reserve that keeps you independent at 80 What high reps do for an ageing body: - Generate more total fatigue per session - Take longer to recover from (a problem that compounds with age) - Do nothing for tendon stiffness - Do nothing for fast twitch preservation - Make you tired without making you strong Recovery capacity declines with age. The worst thing you can do is pick the more fatiguing option and call it gentle. The actual protocol: - Heavy enough to challenge the muscle (4-6 reps, near failure) - Machines for most of the work (less spinal load, less balance demand, more stimulus to the muscle, less risk) - Skip the 1RM testing, skip the maxing out on barbell squats and deadlifts - That isn't where the magic was anyway You don't need to max out a barbell to lift heavy. You need to challenge the muscle with a load it respects. The leg press at 200kg does this. The hack squat does this. The chest-supported row does this. None of them are ego lifts. All of them are heavy. Lifting lighter to "protect" your joints is how you arrive at 75 unable to stand up from a chair. Lifting heavy on machines is how you arrive at 75 carrying your own shopping.

Sama Hoole

35,515 просмотров • 4 месяцев назад

Every price tag on every product in Europe is about to be repriced. Elon Musk just confirmed the timeline. Most people scrolled past it. Musk: “We’ve got the Tesla Semi coming up, so the Tesla heavy truck. And that’ll be going to Europe, hopefully next year.” Most people are watching the Robotaxi rollout. The quiet money is watching the Semi. Because the Robotaxi is the regulatory battering ram. The Semi is the economic payload. The global supply chain has one bottleneck no amount of capital has ever solved. The human driver. They get tired. They make mistakes. They sleep. Musk is about to remove all three from the equation simultaneously. Pair a fully electric heavy truck with Full Self-Driving software and the cost structure of moving goods across an entire continent does not improve. It collapses. No fatigue. No sleep schedules. No fuel volatility. No human error. 24 hours. Every day. FSD is already statistically outperforming human drivers on safety metrics. The moment regulators accept that AI navigates a city safer than a human, applying that same software to commercial freight stops being a debate. It becomes a legal obligation. When the trucks move autonomously, the cost of everything on the shelf moves with them. Add humanoid robotics to the warehouse and the marginal cost of moving a product from point A to point B approaches zero. This is not a logistics story. It is a price-of-everything story. Physical transportation is just another data problem waiting to be solved by compute. And once that problem is solved, the inflation that has quietly taxed every human being alive for a century gets a knife in its throat.
0:14

Every price tag on every product in Europe is about to be repriced. Elon Musk just confirmed the timeline. Most people scrolled past it. Musk: “We’ve got the Tesla Semi coming up, so the Tesla heavy truck. And that’ll be going to Europe, hopefully next year.” Most people are watching the Robotaxi rollout. The quiet money is watching the Semi. Because the Robotaxi is the regulatory battering ram. The Semi is the economic payload. The global supply chain has one bottleneck no amount of capital has ever solved. The human driver. They get tired. They make mistakes. They sleep. Musk is about to remove all three from the equation simultaneously. Pair a fully electric heavy truck with Full Self-Driving software and the cost structure of moving goods across an entire continent does not improve. It collapses. No fatigue. No sleep schedules. No fuel volatility. No human error. 24 hours. Every day. FSD is already statistically outperforming human drivers on safety metrics. The moment regulators accept that AI navigates a city safer than a human, applying that same software to commercial freight stops being a debate. It becomes a legal obligation. When the trucks move autonomously, the cost of everything on the shelf moves with them. Add humanoid robotics to the warehouse and the marginal cost of moving a product from point A to point B approaches zero. This is not a logistics story. It is a price-of-everything story. Physical transportation is just another data problem waiting to be solved by compute. And once that problem is solved, the inflation that has quietly taxed every human being alive for a century gets a knife in its throat.

Dustin

84,506 просмотров • 7 месяцев назад

Somewhere around sixty you get handed a new set of instructions. Lift lighter. Keep the reps high. Do not tax yourself too much. Put the saved effort into cardio. It is the exact reverse of what an ageing body needs, and the people handing it out have the mechanism sitting right in front of them. Recovery gets worse with age. Nobody argues with that. The older body clears fatigue more slowly, repairs more slowly, and tolerates far less accumulated work before progress stops entirely. Every GP, every physio, every trainer will nod along to that sentence. Then watch what gets prescribed on the back of it. High reps. Long burning sets. Circuits. Three sessions of cardio stacked on top. A protocol whose main product is fatigue, given to the person with the least capacity left to absorb any. They identified a recovery problem and prescribed more recovery cost. The answer runs the other way and it is not complicated. If your recovery budget has shrunk, you spend it on whatever returns the most growth per unit of fatigue, and that is a heavy set of five. Four to six reps, a handful of lifts, three minutes between sets, done inside the hour. Nearly every rep is a growth rep. Almost nothing goes on the burning, the sweating and the gasping, which build nothing at all and then bill you for four days. Twenty-five reps taken to failure is a fortnight of fatigue for a fraction of the stimulus. That is not the cautious option for a sixty-five-year-old. It is the most reckless thing on the timetable. Now the part that actually matters. Ageing is not one process. It is a list. Muscle wastes. Bone thins. Tendon softens. The fast fibres that catch you when the pavement arrives early vanish first while the slow ones sit there in perfect health. Motor units drop out. The nervous system stops asking for full effort because nothing has demanded full effort in fifteen years. Read that list back and tell me what heavy resistance training does. It builds muscle. It loads bone, which is the only language bone speaks. It stiffens tendon. It recruits the fast fibres, because that is what heavy means physiologically and there is no other route in. It forces the nervous system to ask for everything again. Every item on the list of what ageing takes is on the list of what a heavy set gives back. Nothing else on earth does that. Not a walk, not a class, not a pill, not twenty minutes on a machine with the paper open. You were told to go gently because somebody quietly decided you were finishing. Go heavy, because you are not.
0:19

Somewhere around sixty you get handed a new set of instructions. Lift lighter. Keep the reps high. Do not tax yourself too much. Put the saved effort into cardio. It is the exact reverse of what an ageing body needs, and the people handing it out have the mechanism sitting right in front of them. Recovery gets worse with age. Nobody argues with that. The older body clears fatigue more slowly, repairs more slowly, and tolerates far less accumulated work before progress stops entirely. Every GP, every physio, every trainer will nod along to that sentence. Then watch what gets prescribed on the back of it. High reps. Long burning sets. Circuits. Three sessions of cardio stacked on top. A protocol whose main product is fatigue, given to the person with the least capacity left to absorb any. They identified a recovery problem and prescribed more recovery cost. The answer runs the other way and it is not complicated. If your recovery budget has shrunk, you spend it on whatever returns the most growth per unit of fatigue, and that is a heavy set of five. Four to six reps, a handful of lifts, three minutes between sets, done inside the hour. Nearly every rep is a growth rep. Almost nothing goes on the burning, the sweating and the gasping, which build nothing at all and then bill you for four days. Twenty-five reps taken to failure is a fortnight of fatigue for a fraction of the stimulus. That is not the cautious option for a sixty-five-year-old. It is the most reckless thing on the timetable. Now the part that actually matters. Ageing is not one process. It is a list. Muscle wastes. Bone thins. Tendon softens. The fast fibres that catch you when the pavement arrives early vanish first while the slow ones sit there in perfect health. Motor units drop out. The nervous system stops asking for full effort because nothing has demanded full effort in fifteen years. Read that list back and tell me what heavy resistance training does. It builds muscle. It loads bone, which is the only language bone speaks. It stiffens tendon. It recruits the fast fibres, because that is what heavy means physiologically and there is no other route in. It forces the nervous system to ask for everything again. Every item on the list of what ageing takes is on the list of what a heavy set gives back. Nothing else on earth does that. Not a walk, not a class, not a pill, not twenty minutes on a machine with the paper open. You were told to go gently because somebody quietly decided you were finishing. Go heavy, because you are not.

Sama Hoole

16,526 просмотров • 2 месяцев назад

🚨 MERCEDES JUST PUT A MOTOR ONLY 8 CM THICK INTO A CAR THAT CAN HIT 62 MPH IN 2.1 SECONDS. Instead of conventional radial flux motors, Mercedes is betting big on axial flux technology. In these motors, the electromagnetic force flows parallel to the axle, allowing two magnetic rotors to sandwich a central stator in a flat, disc-like layout. The result is dramatically smaller and more powerful. The front motor in the new all-electric Mercedes-AMG GT 4-door Coupe is just 9 cm wide. The rear motors are even thinner at roughly 8 cm each. Despite their tiny size, they help launch the heavy performance car from 0-62 mph in just 2.1 seconds, with a top speed of up to 186 mph. Why this matters: • Axial flux motors are significantly more power-dense and can be up to 50% lighter than traditional designs • Their extreme thinness frees up packaging space in the vehicle for better weight distribution, aerodynamics, or interior room • Mercedes acquired YASA in 2021 and has spent years developing the complex manufacturing processes needed to build them at scale • The technology is debuting in a high-performance AMG model, showing Mercedes is serious about using it in its most demanding cars The deeper implication: While most of the EV conversation focuses on batteries and software, the electric motor itself is undergoing a quiet revolution. Axial flux designs have long been seen as theoretically superior but extremely difficult to manufacture at scale. By solving the production challenges and putting these motors into a real high-performance car, Mercedes is pushing the entire industry forward. The next generation of electric performance cars may not just have bigger batteries they may have fundamentally better motors. We’re watching the physical hardware of EVs evolve as dramatically as the software has. How important do you think motor technology (rather than just battery size) will be for the future of electric performance cars? Follow for more frontier automotive engineering and electric vehicle technology.
0:10

🚨 MERCEDES JUST PUT A MOTOR ONLY 8 CM THICK INTO A CAR THAT CAN HIT 62 MPH IN 2.1 SECONDS. Instead of conventional radial flux motors, Mercedes is betting big on axial flux technology. In these motors, the electromagnetic force flows parallel to the axle, allowing two magnetic rotors to sandwich a central stator in a flat, disc-like layout. The result is dramatically smaller and more powerful. The front motor in the new all-electric Mercedes-AMG GT 4-door Coupe is just 9 cm wide. The rear motors are even thinner at roughly 8 cm each. Despite their tiny size, they help launch the heavy performance car from 0-62 mph in just 2.1 seconds, with a top speed of up to 186 mph. Why this matters: • Axial flux motors are significantly more power-dense and can be up to 50% lighter than traditional designs • Their extreme thinness frees up packaging space in the vehicle for better weight distribution, aerodynamics, or interior room • Mercedes acquired YASA in 2021 and has spent years developing the complex manufacturing processes needed to build them at scale • The technology is debuting in a high-performance AMG model, showing Mercedes is serious about using it in its most demanding cars The deeper implication: While most of the EV conversation focuses on batteries and software, the electric motor itself is undergoing a quiet revolution. Axial flux designs have long been seen as theoretically superior but extremely difficult to manufacture at scale. By solving the production challenges and putting these motors into a real high-performance car, Mercedes is pushing the entire industry forward. The next generation of electric performance cars may not just have bigger batteries they may have fundamentally better motors. We’re watching the physical hardware of EVs evolve as dramatically as the software has. How important do you think motor technology (rather than just battery size) will be for the future of electric performance cars? Follow for more frontier automotive engineering and electric vehicle technology.

TheNewPhysics

399,616 просмотров • 3 месяцев назад

When you train in the 10-12 rep range, most of your reps have no direct effect on growth. When you train in the 4-6 range, virtually all of your reps are growth reps. Both ranges can build muscle. The mechanism doesn't care about the rep count. It cares about how close you get to true failure on the reps where the high-threshold motor units are recruited and every available fibre is firing. Those are the stimulating reps. Everything else is filler. The catch with 10-12 is twofold. First, only the last 4-5 reps in a 12-rep set are actually stimulating. The first seven are buffer. They generate fatigue, lactic acid, and joint wear that the muscle has to push through before any growth signal arrives. Effort, yes. Stimulus, no. Second, and this is where the high-rep crowd quietly come undone: the long set produces so much afferent feedback (burning, gasping, the legs giving a small philosophical speech) that almost nobody actually takes the set to true failure. They stop two, three, sometimes four reps short, mistake the discomfort for the limit, and call it a hard set. The stimulating reps they were chasing never showed up. A set of 6 doesn't allow that confusion. Failure is mechanical. The weight either moves or it doesn't. No interpretive dance required. You'll grow on 10-12. You'll grow more on 4-6, with less joint wear, less recovery debt, and considerably less guesswork. One range tolerates your mistakes. The other doesn't have room for them.
0:18

When you train in the 10-12 rep range, most of your reps have no direct effect on growth. When you train in the 4-6 range, virtually all of your reps are growth reps. Both ranges can build muscle. The mechanism doesn't care about the rep count. It cares about how close you get to true failure on the reps where the high-threshold motor units are recruited and every available fibre is firing. Those are the stimulating reps. Everything else is filler. The catch with 10-12 is twofold. First, only the last 4-5 reps in a 12-rep set are actually stimulating. The first seven are buffer. They generate fatigue, lactic acid, and joint wear that the muscle has to push through before any growth signal arrives. Effort, yes. Stimulus, no. Second, and this is where the high-rep crowd quietly come undone: the long set produces so much afferent feedback (burning, gasping, the legs giving a small philosophical speech) that almost nobody actually takes the set to true failure. They stop two, three, sometimes four reps short, mistake the discomfort for the limit, and call it a hard set. The stimulating reps they were chasing never showed up. A set of 6 doesn't allow that confusion. Failure is mechanical. The weight either moves or it doesn't. No interpretive dance required. You'll grow on 10-12. You'll grow more on 4-6, with less joint wear, less recovery debt, and considerably less guesswork. One range tolerates your mistakes. The other doesn't have room for them.

Sama Hoole

63,051 просмотров • 4 месяцев назад

A good technical LLM interview question: Your RAG chatbot is working as expected locally. You deploy it behind a load balancer with 3 replicas. Users report that it forgets what they just asked, and answers get worse with each restart. Why did this happen? (answer below) A local setup has one process that owns everything. - The vector index is a variable in memory. - Conversation history is a Python list. - The documents are on local disk. You never treat any of them as infrastructure, because restarting rebuilds all three in seconds and there is only ever one copy. The setup does not carry over to production directly. The vector index might disappear on restart, so the app re-embeds everything on boot and serves empty results until it finishes. Conversation history may belong to one replica, so a follow-up routed elsewhere has no memory of the previous turn. Documents could be on whichever container ingested them, so the three replicas hold three different corpora. None of this is evident with one user and one process. So the actual work in shipping RAG is not just the retrieval logic, but also storing the vector index, the conversation history, and the documents outside the app, where every replica reads and writes the same copy. Which comes down to three requirements: > The vector store needs persistence and has to be reachable from every replica. pgvector inside Postgres keeps embeddings next to the rest of the data instead of adding another system to operate. > Conversation state has to be checkpointed outside the app. LangGraph writes its state to Postgres, so any replica can pick up a thread mid-conversation. > Docs need shared object storage, so ingestion happens once instead of once per replica. If you get those three right, the retrieval logic you wrote in the notebook works unchanged. To learn how all of it is wired together, Akamai's GitHub has a working reference implementation. - rag-langgraph-k8s-quickstart is an airline policy Q&A assistant built with FastAPI, LangChain, and LangGraph. Terraform provisions the LKE cluster, a Postgres instance with pgvector for embeddings, a second Postgres for LangGraph checkpointing, and an object storage bucket for the policy documents, in one apply. - akamai-workshop-ai-inference covers the next step, running the model yourself instead of calling an API, with prefill and decode, KV cache tradeoffs, and continuous batching under real concurrency. Both are available on Akamai's new Developer Hub, alongside their tutorials and code samples. It also links to Edge Case, their Discord, where four developer advocates architect and deploy a production app live every other Wednesday. If you create a new Akamai Cloud account, you can also get $300 in credits for joining. Join here: That said, this post assumes the retrieval logic was right to begin with, and that is doing a lot of work. Most RAG systems fail earlier, at the point where a chunk gets treated as a self-contained unit of meaning. I wrote about the two skills that fix that gap, and why the chunk is usually the wrong thing to embed. Read it below. Thanks to Akamai Cloud for partnering today!
0:14

A good technical LLM interview question: Your RAG chatbot is working as expected locally. You deploy it behind a load balancer with 3 replicas. Users report that it forgets what they just asked, and answers get worse with each restart. Why did this happen? (answer below) A local setup has one process that owns everything. - The vector index is a variable in memory. - Conversation history is a Python list. - The documents are on local disk. You never treat any of them as infrastructure, because restarting rebuilds all three in seconds and there is only ever one copy. The setup does not carry over to production directly. The vector index might disappear on restart, so the app re-embeds everything on boot and serves empty results until it finishes. Conversation history may belong to one replica, so a follow-up routed elsewhere has no memory of the previous turn. Documents could be on whichever container ingested them, so the three replicas hold three different corpora. None of this is evident with one user and one process. So the actual work in shipping RAG is not just the retrieval logic, but also storing the vector index, the conversation history, and the documents outside the app, where every replica reads and writes the same copy. Which comes down to three requirements: > The vector store needs persistence and has to be reachable from every replica. pgvector inside Postgres keeps embeddings next to the rest of the data instead of adding another system to operate. > Conversation state has to be checkpointed outside the app. LangGraph writes its state to Postgres, so any replica can pick up a thread mid-conversation. > Docs need shared object storage, so ingestion happens once instead of once per replica. If you get those three right, the retrieval logic you wrote in the notebook works unchanged. To learn how all of it is wired together, Akamai's GitHub has a working reference implementation. - rag-langgraph-k8s-quickstart is an airline policy Q&A assistant built with FastAPI, LangChain, and LangGraph. Terraform provisions the LKE cluster, a Postgres instance with pgvector for embeddings, a second Postgres for LangGraph checkpointing, and an object storage bucket for the policy documents, in one apply. - akamai-workshop-ai-inference covers the next step, running the model yourself instead of calling an API, with prefill and decode, KV cache tradeoffs, and continuous batching under real concurrency. Both are available on Akamai's new Developer Hub, alongside their tutorials and code samples. It also links to Edge Case, their Discord, where four developer advocates architect and deploy a production app live every other Wednesday. If you create a new Akamai Cloud account, you can also get $300 in credits for joining. Join here: That said, this post assumes the retrieval logic was right to begin with, and that is doing a lot of work. Most RAG systems fail earlier, at the point where a chunk gets treated as a self-contained unit of meaning. I wrote about the two skills that fix that gap, and why the chunk is usually the wrong thing to embed. Read it below. Thanks to Akamai Cloud for partnering today!

Akshay 🚀

31,971 просмотров • 1 месяц назад

A Letter to Our Community: The Road Ahead for Robotics To our Community and Partners, As we step into 2026, our mission at Axis is clearer than ever: Constructing the definitive End-to-End Scaling Layer for Robotics. Our goal is to accelerate the transfer of diverse human intelligence into Robotics General Intelligence (RGI). By owning the critical path of intelligence creation, we are turning the physical limitations of robotics into a scalable, software-driven future. Here is our strategic outlook and roadmap for the year ahead. The Core Thesis: Simulation is the Only Way Out The path to RGI is currently blocked by Data Scarcity, Generalization Fragility, and Hardware Fragmentation. At Axis, we believe Simulation is the only way out. Our Simulation Data Platform and Data Augmentation Engine transform raw data into "Synthetic Gold". Backed by academic milestones like Roboverse, Skill Blending, and GraspVLA, we have proven that pure simulation can achieve the generalization required for the real world. We don’t just collect data; we architect it. The Engine: Why Crypto? We believe RGI should come from all, not a few. Crypto is not just a feature; it is the primitive that powers our entire ecosystem flywheel: - Incentive Mechanism: Democratizing contribution and rewarding the trainers and developers. - Assetization: Turning proprietary data and refined models into liquid, ownable assets. - Verifiable Workflow: We are opening the "Black Box" of AI. By bringing total transparency to the Task Generation → Data Collection → Model Training pipeline, we ensure every byte of intelligence is verifiable, traceable, and secure. 2026 Strategic Deliverables This year, we are committed to delivering three foundational pillars: - The World's Largest Training Dataset for Robots: A robot training set—diverse, high-quality interaction data at an unprecedented scale. - A Robotics Foundation Model: A universal robotic brain trained on our pure simulation and synthetic data, capable of robust cross-embodiment transfer and open-world adaptability. - Evolvable Robot Hardware: Robots deployed with Axis models that autonomously evolve through continuous interaction, turning every deployment into a self-improving node within our RGI network. The Ultimate Vision We are building more than models; we are architecting the Distributed Machine Economy. A future where every dataset, model, and robotic embodiment is a verifiable asset in a global, autonomous network. Thank you for building the future of intelligence with us✌️📷
0:13

A Letter to Our Community: The Road Ahead for Robotics To our Community and Partners, As we step into 2026, our mission at Axis is clearer than ever: Constructing the definitive End-to-End Scaling Layer for Robotics. Our goal is to accelerate the transfer of diverse human intelligence into Robotics General Intelligence (RGI). By owning the critical path of intelligence creation, we are turning the physical limitations of robotics into a scalable, software-driven future. Here is our strategic outlook and roadmap for the year ahead. The Core Thesis: Simulation is the Only Way Out The path to RGI is currently blocked by Data Scarcity, Generalization Fragility, and Hardware Fragmentation. At Axis, we believe Simulation is the only way out. Our Simulation Data Platform and Data Augmentation Engine transform raw data into "Synthetic Gold". Backed by academic milestones like Roboverse, Skill Blending, and GraspVLA, we have proven that pure simulation can achieve the generalization required for the real world. We don’t just collect data; we architect it. The Engine: Why Crypto? We believe RGI should come from all, not a few. Crypto is not just a feature; it is the primitive that powers our entire ecosystem flywheel: - Incentive Mechanism: Democratizing contribution and rewarding the trainers and developers. - Assetization: Turning proprietary data and refined models into liquid, ownable assets. - Verifiable Workflow: We are opening the "Black Box" of AI. By bringing total transparency to the Task Generation → Data Collection → Model Training pipeline, we ensure every byte of intelligence is verifiable, traceable, and secure. 2026 Strategic Deliverables This year, we are committed to delivering three foundational pillars: - The World's Largest Training Dataset for Robots: A robot training set—diverse, high-quality interaction data at an unprecedented scale. - A Robotics Foundation Model: A universal robotic brain trained on our pure simulation and synthetic data, capable of robust cross-embodiment transfer and open-world adaptability. - Evolvable Robot Hardware: Robots deployed with Axis models that autonomously evolve through continuous interaction, turning every deployment into a self-improving node within our RGI network. The Ultimate Vision We are building more than models; we are architecting the Distributed Machine Economy. A future where every dataset, model, and robotic embodiment is a verifiable asset in a global, autonomous network. Thank you for building the future of intelligence with us✌️📷

Axis Robotics

28,096 просмотров • 9 месяцев назад