Every now and then I fall into this rabbit hole in Excel, but I have enough other crap to do that I can't devote enough time to it so I just manually do the work and move on with life.
One time I did completely alter a vehicle slip torque calculation tool in Excel because it wasn't set up for transverse AWD systems like the one I was working on, I was actually kind of proud of that one. But then that program (imagine if the performance wing of a major OEM got their hands on a baby off-road cute-ute) got killed and nobody ever used my fancy calculation tool again. They probably reinvented the wheel in their own silos.
One of the many things not understood in Redmond is muscle memory. Work computer just changed from Win10 to Win11. For so many things my mouse hand twitches to where a function used to be. Grrrrrrrrrrr indeed.
Reddit was great when thousands of socially retarded nerd adjacent college chicks would post pictures of their tits in exchange for a rapidly escalating number next to an “up” arrow, or maybe a flashy little icon next to their username. Now it’s just OF whores engagement farming, and we’re poorer for it.
While I'll admit to a playful slap on the ass or two, I've never choked, spit on, or tried to humiliate a woman with whom I was enjoying sexual congress. I wonder if early exposure to all that is making it seem normal to those under 30.
NOAA has demonstrated a model trained on both Reddit (all of it; not just weather subreddits) and past weather data is more accurate than one trained solely on past weather data.
I - and they - have no idea why this is the case. And we've seen similar outcomes where you, say, add the back catalog of WSJ to 'x' economic data.
There's an optimal mix of "text" and "context" that seems to be helpful for some use cases. But again, no one quite seems to really know why or how to tune the ratios.
theyre so old and dumb that if i walked into congress and showed them how to work a graphing calculator i could walk out with a few million in grant money
Looking at just a few articles about Trump's temporary suspension of government grants to NGOs and related activist groups in order to get a handle on things reveals what a breathtaking level of grift must exist.
We used to have a fairly robust Office of Technology Assessment... which Gingrich completely cut in 1995. STAA kinda/sorta filled the void, but it was years later and nowhere near the scope or capacity.
OTA work still had to get done though and you know who did it? Tech lobbyists.
shoutout to deepseek for wiping out a whole ass trillion dollars of valuation, maybe that will finally wake them up out of their "just add more gpus bro" development ideology (it wont)
man i just want some kind of ai thing that can tune cars or something similar
One of my college buddies - and probably the most intelligent person I have ever met - is the head of “Business Finance” at OpenAI; he led the October $6.6BN raise that valued the company at $157BN.
I spoke to him over the weekend about fundraising for MY business (it’s a little bit smaller than OpenAI - for now 😂), so I was still top of mind for him yesterday when the market got hammered. I didn’t want to nag him on what was probably a bad day, but I was curious…
He sent a pic of his kids with the caption “These two don’t seem to care that much about DeepSeek.”
He is Chinese(-American) btw, and I would assume he knows more about it than anyone in this comment section.
'He sent a pic of his kids with the caption “These two don’t seem to care that much about DeepSeek.”'
He might know a lot about nerd shit, but you know less about people. "Here's a oddly contextualized picture of my kids" is social media and text code for "My career is fucked and I'm suicidal."
You understand that when Tom Wolfe has your namesake regard himself as a "Master Of The Universe", it's because everyone else thinks he's a nonentity?
I don't know what his situation is, but when grown men start using pictures of their kids to trivialize something, it's because they are deeply upset about that something. Source: I follow 1,500 parents on Instagram.
ive heard of auto tune functions but i was hoping for something far more in depth, one that could tune any combination rapidly from scratch with a few user input parameters and limits
a hyper complex ai tuner could optimize for every condition and make modifying engines far easier (hopefully)
I built something like this in a past life. Requires a very specific platform. unfortunately didn't come to pass as a commercial product except ONE FSAE team still using it, partially.
This is difficult for me to do without identifying myself.
When I graduated in 2009 I was working for a startup with a new product in EFI where the controller did not ship with any software, but came with tools to build your own models using basic building blocks (like a programming language) and could write something as sophisticated as you could think of.
I was the first application development engineer for this platform and when they found a customer to sell the whole platform to, the 'customer' refused to entertain either hiring me or using any of my work as a starting point. So the FSAE team that served as the ground floor for all this development ended up being my test bed, in addition to personal projects for improving my models.
The fueling feedback is straightforward and there are many ways to do it. My preferred method is to use a crude guess and rely on a modified form of the newton raphson method called a delta controller to find the best value based on oxygen sensor feedback. Every decent aftermarket controller has their own method and most probably work just as well (not saying this based on experience) - but the thing to note is I was doing this back in 2010.
For transient fueling I actually referenced the megasquirt code and replicated it and modified it. Funnily most new squirts don't have the x-tau tuning algorithm anymore even though it is one of the best that simply works once you get it working on any single engine.
For ignition - you need feedback from knock sensors, or estimated wheel torque from passes, or ideally dyno feedback. Its simply a question of following the same process as what a tuner would - increase numbers until you get to a flat spot and back off a safe amount... if you hit knock first you can back off sooner without finding MBT.
In industry some companies like FCA have some custom tools to do the same with INCA and Matlab and Simulink, and others use off the shelf tools like AVL CREO or PUMA (I infer that this is what F1 teams use as well).
The key to the system I was working on was you only paid for the hardware and then the tools to build software were (are) free and basically limitless. By comparison you're looking at 100k in INCA, Matlab and Simulink and someone who knows how to use all of these well for 6-10 months before you have something useful. The AVL tools are also tens of thousands of dollars and require a lot of support and add on packages.
Since you are this curious, some of this early work is the only reason I'm in the country, because this startup wanted me to go to shows like PRI and SAE world congress and talk with people and give demos, including chip company executives, RIGHT IN the depths of the 2009/2010 recession. I was 2 days away from leaving since I could not find a job as a fresh grad in a market where people with mortgages to pay were willing to work for 70k. They hired me as a 1099 and gave me an employment letter that I could use for an OPT application. But hey, Pajeets are only good for shitting in the streets.
This is all fairly high level and straightforward... I'm not saying I reinvented the wheel in any fashion, but I was an important part of an American business that thrived due to that early work and did well for its founders. They couldn't afford to hire me full time which is how I ended up at Navistar with my faked credentials from knowing all the answers to the questions on tests handed to me through the window at the exam.
Sarc off/ I have offered data analysis and calibration help to anyone here (including our gracious host) that needs it for their weekend track toy or racecar or whatever else in the past. Just that I have a lot of competing priorities so I can't really guarantee a quick turnaround for time sensitive things.
Worse, in big corporations we are going to have hordes of idiot middle and upper managers that will want to replace lots of employees with AI bots because all they care about is this quarter's budget and P/L numbers. It will be worse than the outsourcing of jobs to "low cost labor countries" even if that is hard to believe.
You don't need AI to provide crappy customer service.
Eddie Bauer simply refuses to honor the lifetime warranty that came with my Ridgeline down parka when I got it. That's besides the fact that almost nothing they sell is made in America anymore.
I suppose that I shouldn't complain since the coat was effectively free to me, but I think the story says something about the decline of a great American brand.
After my father died in 1989, I inherited his Eddie Bauer parka, that he probably got back when the Detroit Lions played outdoors at Tiger Stadium (he had season's tickets that he shared with his best friend, Bill Coleman). I wore it for a while until the zipper's cotton fabric ripped apart. When I wore the deteriorating coat to a friend's house, it turned out that he had matching EB stadium coats for him and his wife, and he told me about their lifetime warranty. I took the coat to the local Eddie Bauer store, they looked up how much the coat cost when new and gave me $185 credit towards the purchase of a $225 Ridgeline. I asked my ex if it was okay to spend $40 on a $225 coat and she said sure. A month later the jacket went on sale for $185 and the EB store even refunded me the difference.
A few years ago, when the zipper pull broke, I took it in to the local EB store and they were willing to give me a new coat, but it was made in China, wasn't as well made as the original Ridgelines, and it also the XL didn't fit properly even though I don't quite need an XL.
Now, they have a 60 day return policy and the warranty against defects is only good for a year. Never mind their conditions of sale 30 years ago.
Now that the Ridgeline is trashed and my backup fiberfill parka also consumed its zipper, I ended up buying a used Ridgeline on eBay for $50 plus shipping to get through Michigan's winter.
I used to think that way. Now I realize that the real reason we have to listen to this AI BS on-and-on is not because they didn't / don't read the ACF. No, it's because people who are "all in" on AI cannot afford to publicly backtrack. The higher up / more visible they are, the longer they will extolling the virtues of the AI.
Obviously Deepseek won’t amount to anything. They are missing all the usual 2025 requirements for hiring: “
“DeepSeek's hiring preferences target technical abilities rather than work experience, resulting in most new hires being either recent university graduates or developers whose AI careers are less established.[18][4] Likewise, the company recruits individuals without any computer science background to help its technology understand other topics and knowledge areas, including being able to generate poetry and perform well on the notoriously difficult Chinese college admissions exams”
One suspects that the "notoriously difficult Chinese college admissions exams" are about as reliable as anything else Chinese. Just how does that society keep its normal bullshit, lies, and deception out of its critical institutions? Edit: We certainly haven't kept all that stuff out of our critical institutions.
My friends who lived in the USSR tell me that the rot extended everywhere in that society.
Having grown up in a (formerly) Communist country, I can confirm that the purpose of "tough college admission exams" was not to implement high standards, but to leave enough "room" above the admittance line for the "party faithful" from the members of leadership who choose to "make the call."
I don’t doubt any of this. The valuations will be interesting long term for one particular reason- the Chinese are liars. You cannot trust anything they say about their economy, tech (or anything).
How are we trading this? AI will drive data, which will drive continued data center development.
Also- I am not planning to download any Chinese company owned app. The deep seek app is a privacy nightmare and I sure as hell will never use it.
Is it time for the ACF AI and have our own system? Made in America for Americans? I have no computer skills past being a user, and lack any interest in achieving these skills, so some of you would need to do the heavy lifting in the world of 1’s and 0’s.
With enough eyes on opensource AI code over time, the worst of the violations will be removed. (Let's not get into the Ken Thompson Hack right now.) It will be like Android code.
My harmonica components are made by a company in Germany. In an email to me, their production manager used the term "Invention-spirit", which I figured just had to be literally translated from German and it turns out there's a word "Erfindergeist", for inventiveness.
"I am Moses Mendelsohn of Borg. Prepare to be assimilated."
That's a very funny joke, but you have to live in two worlds to get it.
This is the most intellectual joke I've heard in years:
Two friends work for years on a time-machine and when it's finished they decide that one of them will use it to go back in history and kill Hitler before he came to power. The machine sends him back to 1910 but it malfunctions and puts him in Princeton, New Jersey instead of Vienna so he decides to kill Woodrow Wilson instead. When he returns to the present, he exits the machine and tells his friend, "The machine works but I was unable to kill Hitler," and his friend replies, "Who's Hitler?"
The market panic was about the cost- they claim they built and taught this model for only $5.6m. There is a healthy amount of skepticism about that, especially since slave laborers in China work for free.
If their code is revolutionary and can run more efficiently (requiring less processing power) that would be something. You’d think we would know that soon. This will affect demand for the highest performance chips and the infrastructure around them.
As good an explanation as any. Naked corporate espionage to torpedo a bubble in the US economy in the face of Trump's trade war with all the ripples that popping entails.
Bambu Labs, a Chinese company which has been very successful supplanting Prusa as the best bang for the buck prosumer 3D printer, recently made some changes to their firmware that makes it impossible to use 3rd party software to control the machines and reduces functionality if you don't print via the servers in China. For companies running print farms this is very inconvenient.
I recently switched to LAN mode for printing, as I'd noticed their servers had been slowing down, and it's pretty clear that they really want people to use the cloud (i.e. Chinese servers). Whether that's because of industrial espionage or just because they want a closed environment I can't tell. Unlike wifi capable laser printers, I can't see the printer on my LAN. I have to use Bambu's software to control it and about 30% of the time have to reenter an 8 digit alphanumeric access code to connect, so there's probably some authentication and other communication still going on with their servers.
I'll tell you what: I'll buy a claim of having developed True AI from space aliens or far-future humans who'd look at us and say something like, "These savages are still using electricity."
KITT was not possible with early 1980s computer technology and as long as we continue to use silicon-based hardware and language-based programming, the 2080s won't get us any closer to such a consciousness.
It will have to be local, meaning individuals and businesses have their own personal AI, for security reasons, right? Firewalled to get the learning in but not send data out? I’m reaching about the limit of my computer science comprehension….
Maybe that is the future, cloud or desktops that we all lease/buy to host our personal AI.
That is possible, of course, but it’s also possible that a specific AI medical tool has a lower failure rate than a doctor. A medical AI tool that can statistically prove this could be a great thing for certain medical issues.
every day i get closer to making bomb threat to substack hq with the intent of them making their site less shit
Apparently no one who works on substack also uses it. I don't know how it continues with no improvement.
I forgot - QA means automating a happy path test written by the devs. No bugs! Product is great!
oh its worse than no improvement
they actively change it to make it less enjoyable and useful
i hate these people
It would be really funny to ask AI to summarize the ACF comments for a given post.
i like how we both had the same idea lol
print the comments pages to a pdf file, then send that to the AI bot.
Like a 7th grader fudging answers.
i need to shitpost with far greater efficiency and ai could help me do that
AI efficient shitposting would just be hard slurs right off the bat, no wasting time trying to pick on the context.
wait unitl you hear about tay.ai
AI efficient shitposting would just be hard slurs right off the bat, no wasting time trying to pick on the context.
Is this doublepost an indication that you yourself are A.I.tposting?
If so, 🫡
Every now and then I fall into this rabbit hole in Excel, but I have enough other crap to do that I can't devote enough time to it so I just manually do the work and move on with life.
One time I did completely alter a vehicle slip torque calculation tool in Excel because it wasn't set up for transverse AWD systems like the one I was working on, I was actually kind of proud of that one. But then that program (imagine if the performance wing of a major OEM got their hands on a baby off-road cute-ute) got killed and nobody ever used my fancy calculation tool again. They probably reinvented the wheel in their own silos.
Don't .... just don't ... get me started on Redmond changing things, again, grrrrrrrrrrrr
Just for change’s sake, not to be anything useful.
One of the many things not understood in Redmond is muscle memory. Work computer just changed from Win10 to Win11. For so many things my mouse hand twitches to where a function used to be. Grrrrrrrrrrr indeed.
see now that actually sounds cool as hell
It was. Shame we never got far enough in to build one and show it to people. That tends to convince them to approve programs!
'It's trained on reddit you morons. AI is gonna be a fat furry equivalent.'
Don't ask it about sex. DON'T.
Give us at least a hint why, or you know every single one of us will!
I haven't asked -- but if it's trained on Reddit, the answers will disgust you and every other decent man out there.
Reddit was great when thousands of socially retarded nerd adjacent college chicks would post pictures of their tits in exchange for a rapidly escalating number next to an “up” arrow, or maybe a flashy little icon next to their username. Now it’s just OF whores engagement farming, and we’re poorer for it.
write your congressman
demand homegrown amateur titty pics
I’ll be running on this platform in 2028.
Lenny Bruce was making fun of fake tits 60 years ago.
Imagine AI trained on Pornhub.
While I'll admit to a playful slap on the ass or two, I've never choked, spit on, or tried to humiliate a woman with whom I was enjoying sexual congress. I wonder if early exposure to all that is making it seem normal to those under 30.
yeah humiliating women is great fun
wait you mean sexually
Pro: you’re going to live to 300
Con: you’ll be a gay nazi catgirl
worse:
youll be a communist troon
Yes, but...
NOAA has demonstrated a model trained on both Reddit (all of it; not just weather subreddits) and past weather data is more accurate than one trained solely on past weather data.
I - and they - have no idea why this is the case. And we've seen similar outcomes where you, say, add the back catalog of WSJ to 'x' economic data.
There's an optimal mix of "text" and "context" that seems to be helpful for some use cases. But again, no one quite seems to really know why or how to tune the ratios.
This succinct take on AI should be widely heeded.
it should but everyone that needs to hear it isnt listening
Can you imagine the elderly and room temperature IQ members of Congress enacting legislation around this?
theyre so old and dumb that if i walked into congress and showed them how to work a graphing calculator i could walk out with a few million in grant money
Looking at just a few articles about Trump's temporary suspension of government grants to NGOs and related activist groups in order to get a handle on things reveals what a breathtaking level of grift must exist.
i mean its the government
not sure you can have one without insane levels of corruption
But... We can hope, and we should try.
That you failed to mention he's simply setting the pins up for him to do the biggest grift of his life is saddening .
-Nate
We used to have a fairly robust Office of Technology Assessment... which Gingrich completely cut in 1995. STAA kinda/sorta filled the void, but it was years later and nowhere near the scope or capacity.
OTA work still had to get done though and you know who did it? Tech lobbyists.
Does this feel eerily similar to now?
shoutout to deepseek for wiping out a whole ass trillion dollars of valuation, maybe that will finally wake them up out of their "just add more gpus bro" development ideology (it wont)
man i just want some kind of ai thing that can tune cars or something similar
is that too much to ask
One of my college buddies - and probably the most intelligent person I have ever met - is the head of “Business Finance” at OpenAI; he led the October $6.6BN raise that valued the company at $157BN.
I spoke to him over the weekend about fundraising for MY business (it’s a little bit smaller than OpenAI - for now 😂), so I was still top of mind for him yesterday when the market got hammered. I didn’t want to nag him on what was probably a bad day, but I was curious…
He sent a pic of his kids with the caption “These two don’t seem to care that much about DeepSeek.”
He is Chinese(-American) btw, and I would assume he knows more about it than anyone in this comment section.
'He sent a pic of his kids with the caption “These two don’t seem to care that much about DeepSeek.”'
He might know a lot about nerd shit, but you know less about people. "Here's a oddly contextualized picture of my kids" is social media and text code for "My career is fucked and I'm suicidal."
He’s not a nerd, he’s a financier.
And his career isn’t “fucked” - he’s worth 8 figures liquid and 9 on paper.
He could walk away today if he wanted to.
'He’s not a nerd, he’s a financier.'
He's not a bird, he's a fowl!
You understand that when Tom Wolfe has your namesake regard himself as a "Master Of The Universe", it's because everyone else thinks he's a nonentity?
I don't know what his situation is, but when grown men start using pictures of their kids to trivialize something, it's because they are deeply upset about that something. Source: I follow 1,500 parents on Instagram.
The “nerd” part would be being a tech weenie, right?
It's like when parents suddenly start posting pics of their kids talking about "what really matters". Sure sign a divorce is imminent.
Even Megasquirt has had "auto-tune" for many, many years. And it isn't all that bad.
ive heard of auto tune functions but i was hoping for something far more in depth, one that could tune any combination rapidly from scratch with a few user input parameters and limits
a hyper complex ai tuner could optimize for every condition and make modifying engines far easier (hopefully)
I built something like this in a past life. Requires a very specific platform. unfortunately didn't come to pass as a commercial product except ONE FSAE team still using it, partially.
man im so glad you commented becuase i had a feeling you did something to that effect lol
how well did it work for what it was intended to do?
Jeez, this is awesome. Would you be so kind as to describe in greater depth? Software, hardware, etc?
"Tuning for Harambe" article?
oh hell yeah
why am i only now realizing how interesting a boom article would be
This is difficult for me to do without identifying myself.
When I graduated in 2009 I was working for a startup with a new product in EFI where the controller did not ship with any software, but came with tools to build your own models using basic building blocks (like a programming language) and could write something as sophisticated as you could think of.
I was the first application development engineer for this platform and when they found a customer to sell the whole platform to, the 'customer' refused to entertain either hiring me or using any of my work as a starting point. So the FSAE team that served as the ground floor for all this development ended up being my test bed, in addition to personal projects for improving my models.
The fueling feedback is straightforward and there are many ways to do it. My preferred method is to use a crude guess and rely on a modified form of the newton raphson method called a delta controller to find the best value based on oxygen sensor feedback. Every decent aftermarket controller has their own method and most probably work just as well (not saying this based on experience) - but the thing to note is I was doing this back in 2010.
For transient fueling I actually referenced the megasquirt code and replicated it and modified it. Funnily most new squirts don't have the x-tau tuning algorithm anymore even though it is one of the best that simply works once you get it working on any single engine.
For ignition - you need feedback from knock sensors, or estimated wheel torque from passes, or ideally dyno feedback. Its simply a question of following the same process as what a tuner would - increase numbers until you get to a flat spot and back off a safe amount... if you hit knock first you can back off sooner without finding MBT.
In industry some companies like FCA have some custom tools to do the same with INCA and Matlab and Simulink, and others use off the shelf tools like AVL CREO or PUMA (I infer that this is what F1 teams use as well).
https://www.avl.com/en/testing-solutions/all-testing-products-and-software/test-system-automation
The key to the system I was working on was you only paid for the hardware and then the tools to build software were (are) free and basically limitless. By comparison you're looking at 100k in INCA, Matlab and Simulink and someone who knows how to use all of these well for 6-10 months before you have something useful. The AVL tools are also tens of thousands of dollars and require a lot of support and add on packages.
Since you are this curious, some of this early work is the only reason I'm in the country, because this startup wanted me to go to shows like PRI and SAE world congress and talk with people and give demos, including chip company executives, RIGHT IN the depths of the 2009/2010 recession. I was 2 days away from leaving since I could not find a job as a fresh grad in a market where people with mortgages to pay were willing to work for 70k. They hired me as a 1099 and gave me an employment letter that I could use for an OPT application. But hey, Pajeets are only good for shitting in the streets.
This is all fairly high level and straightforward... I'm not saying I reinvented the wheel in any fashion, but I was an important part of an American business that thrived due to that early work and did well for its founders. They couldn't afford to hire me full time which is how I ended up at Navistar with my faked credentials from knowing all the answers to the questions on tests handed to me through the window at the exam.
Sarc off/ I have offered data analysis and calibration help to anyone here (including our gracious host) that needs it for their weekend track toy or racecar or whatever else in the past. Just that I have a lot of competing priorities so I can't really guarantee a quick turnaround for time sensitive things.
Thank you for contributing such an immensely thorough and edifying response.
incredibly fascinating - thank you very much for elaborating on this
Thanks for sharing that. Do you have any experience with MoTeC and/or Bosch Motorsport stuff?
Yes, I won't say there's <magic> in engine tuning but it's a very specific thing, so many variables .
-Nate
Well done. Great take.
*Applauds*
Jack always has the best reads on AI. Of course, I had no clue what Deepseek was until I did the usual research that Jacks articles invite.
Worse, in big corporations we are going to have hordes of idiot middle and upper managers that will want to replace lots of employees with AI bots because all they care about is this quarter's budget and P/L numbers. It will be worse than the outsourcing of jobs to "low cost labor countries" even if that is hard to believe.
that AI has been trained on years of bad customer service recordings and scripts
You don't need AI to provide crappy customer service.
Eddie Bauer simply refuses to honor the lifetime warranty that came with my Ridgeline down parka when I got it. That's besides the fact that almost nothing they sell is made in America anymore.
I suppose that I shouldn't complain since the coat was effectively free to me, but I think the story says something about the decline of a great American brand.
After my father died in 1989, I inherited his Eddie Bauer parka, that he probably got back when the Detroit Lions played outdoors at Tiger Stadium (he had season's tickets that he shared with his best friend, Bill Coleman). I wore it for a while until the zipper's cotton fabric ripped apart. When I wore the deteriorating coat to a friend's house, it turned out that he had matching EB stadium coats for him and his wife, and he told me about their lifetime warranty. I took the coat to the local Eddie Bauer store, they looked up how much the coat cost when new and gave me $185 credit towards the purchase of a $225 Ridgeline. I asked my ex if it was okay to spend $40 on a $225 coat and she said sure. A month later the jacket went on sale for $185 and the EB store even refunded me the difference.
A few years ago, when the zipper pull broke, I took it in to the local EB store and they were willing to give me a new coat, but it was made in China, wasn't as well made as the original Ridgelines, and it also the XL didn't fit properly even though I don't quite need an XL.
Now, they have a 60 day return policy and the warranty against defects is only good for a year. Never mind their conditions of sale 30 years ago.
Now that the Ridgeline is trashed and my backup fiberfill parka also consumed its zipper, I ended up buying a used Ridgeline on eBay for $50 plus shipping to get through Michigan's winter.
Worse will be, as Jack points out, someone will use it to design a bridge that subsequently collapses with deadly results.
I feel this.
*looks at his calendar for tomorrow and sees a two hour mandatory Teams meeting for Generative AI Fundamentals*
Tell them you won't listen to their opinion until they read Paul Ford's "What is Code" and Stephen Wolfram's blog post on LLMs.
* https://www.bloomberg.com/graphics/2015-paul-ford-what-is-code/
* https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
More like 20 months
I used to think that way. Now I realize that the real reason we have to listen to this AI BS on-and-on is not because they didn't / don't read the ACF. No, it's because people who are "all in" on AI cannot afford to publicly backtrack. The higher up / more visible they are, the longer they will extolling the virtues of the AI.
Excellent take.
Obviously Deepseek won’t amount to anything. They are missing all the usual 2025 requirements for hiring: “
“DeepSeek's hiring preferences target technical abilities rather than work experience, resulting in most new hires being either recent university graduates or developers whose AI careers are less established.[18][4] Likewise, the company recruits individuals without any computer science background to help its technology understand other topics and knowledge areas, including being able to generate poetry and perform well on the notoriously difficult Chinese college admissions exams”
One suspects that the "notoriously difficult Chinese college admissions exams" are about as reliable as anything else Chinese. Just how does that society keep its normal bullshit, lies, and deception out of its critical institutions? Edit: We certainly haven't kept all that stuff out of our critical institutions.
My friends who lived in the USSR tell me that the rot extended everywhere in that society.
Having grown up in a (formerly) Communist country, I can confirm that the purpose of "tough college admission exams" was not to implement high standards, but to leave enough "room" above the admittance line for the "party faithful" from the members of leadership who choose to "make the call."
Ah, legacy admissions.
I don’t doubt any of this. The valuations will be interesting long term for one particular reason- the Chinese are liars. You cannot trust anything they say about their economy, tech (or anything).
How are we trading this? AI will drive data, which will drive continued data center development.
Also- I am not planning to download any Chinese company owned app. The deep seek app is a privacy nightmare and I sure as hell will never use it.
Is it time for the ACF AI and have our own system? Made in America for Americans? I have no computer skills past being a user, and lack any interest in achieving these skills, so some of you would need to do the heavy lifting in the world of 1’s and 0’s.
Community AI will be a thing.
With enough eyes on opensource AI code over time, the worst of the violations will be removed. (Let's not get into the Ken Thompson Hack right now.) It will be like Android code.
My harmonica components are made by a company in Germany. In an email to me, their production manager used the term "Invention-spirit", which I figured just had to be literally translated from German and it turns out there's a word "Erfindergeist", for inventiveness.
We are the Borg. You will be assimilated.
"I am Moses Mendelsohn of Borg. Prepare to be assimilated."
That's a very funny joke, but you have to live in two worlds to get it.
This is the most intellectual joke I've heard in years:
Two friends work for years on a time-machine and when it's finished they decide that one of them will use it to go back in history and kill Hitler before he came to power. The machine sends him back to 1910 but it malfunctions and puts him in Princeton, New Jersey instead of Vienna so he decides to kill Woodrow Wilson instead. When he returns to the present, he exits the machine and tells his friend, "The machine works but I was unable to kill Hitler," and his friend replies, "Who's Hitler?"
'It will be like Android code.'
That's what I'm afraid of...
Still, it is open source, no? The sausage making will eventually be decoded.
The market panic was about the cost- they claim they built and taught this model for only $5.6m. There is a healthy amount of skepticism about that, especially since slave laborers in China work for free.
If their code is revolutionary and can run more efficiently (requiring less processing power) that would be something. You’d think we would know that soon. This will affect demand for the highest performance chips and the infrastructure around them.
As good an explanation as any. Naked corporate espionage to torpedo a bubble in the US economy in the face of Trump's trade war with all the ripples that popping entails.
yeah, the timing is sus
And the chaser to make it open source. Genius.
Bambu Labs, a Chinese company which has been very successful supplanting Prusa as the best bang for the buck prosumer 3D printer, recently made some changes to their firmware that makes it impossible to use 3rd party software to control the machines and reduces functionality if you don't print via the servers in China. For companies running print farms this is very inconvenient.
I recently switched to LAN mode for printing, as I'd noticed their servers had been slowing down, and it's pretty clear that they really want people to use the cloud (i.e. Chinese servers). Whether that's because of industrial espionage or just because they want a closed environment I can't tell. Unlike wifi capable laser printers, I can't see the printer on my LAN. I have to use Bambu's software to control it and about 30% of the time have to reenter an 8 digit alphanumeric access code to connect, so there's probably some authentication and other communication still going on with their servers.
Well Ronnie, that information just took the P1 off my wishlist hahah
That is a clever way to engage in industrial espionage if you require all 3D prints to go through your computers.
Imagine the designs you could steal -- with "camera ready" artwork just given to you.
Nah, they would never do that ...... er ....
I'll tell you what: I'll buy a claim of having developed True AI from space aliens or far-future humans who'd look at us and say something like, "These savages are still using electricity."
"Dialysis? What is this, the Dark Ages?"
It’s open sourced but from what I gather what it’s been trained on is unknown.
I'd rather china have my data than Facebook
Same thing, as Zuck would sell out anyone or anything, he has no scruples. None…,
I think you mean, "Same thing, as the entire social media sector would sell out anyone or anything, it has no scruples. None."
ACF AI could be the one without the gates and the fences.
Then I'd use it rather than actively avoiding it like I do other AI products.
Great stuff, thank you.
Although I‘m not sure about the tulip analogy - those were at least tangible.
Thank you
True AI would be, in short, an artificial person.
KITT was not possible with early 1980s computer technology and as long as we continue to use silicon-based hardware and language-based programming, the 2080s won't get us any closer to such a consciousness.
wait... you're saying KITT wasn't real?
It will have to be local, meaning individuals and businesses have their own personal AI, for security reasons, right? Firewalled to get the learning in but not send data out? I’m reaching about the limit of my computer science comprehension….
Maybe that is the future, cloud or desktops that we all lease/buy to host our personal AI.
The lizards will find 90% more than acceptable, particularly for medicine.
The systemic harm will be more insidious than the last bubble, subprime.
Except for their own needs, of course. Then they'll hire doctors.
wasnt there a story about a robot or program that started killing all humans becuase that was the best way to prevent them from feeling pain
Why do we need AI for that, it’s already happening with real doctors
real
That is possible, of course, but it’s also possible that a specific AI medical tool has a lower failure rate than a doctor. A medical AI tool that can statistically prove this could be a great thing for certain medical issues.