1,414 karma · joined June 13, 2021
A) In 5 years no real improvement, AI bubble pops, most of us are laid off. B) In 5 years near—AGI replaces most software engineers, most of us are laid off.
Woohoo. Lose-lose scenario! No matter how you imagine this AI bubble playing out, the musics going to stop eventually.
Kids music toys are often just purely toys tap a button, make a sound... But the skill ceiling could be so much higher, offering the ability to learn and express themselves more. Awesome work.
I'm honestly skeptical this will work at all, the FOV of most webcams is so small that it can barely capture the shoulder of someone sitting beside me, let alone their eyes.
Then what you're basically looking for is callibration from the eye position / angle to the screen rectangle. You want to shoot a ray from each eye and see if they intersect with the laptop's screen.
This is challenging because most webcams are pretty low resolution, so each eyeball will probably be like ~20px. From these 20px, you need to estimate the eyeball->screen ray. And of course this varies with the screen size.
TLDR: Decent idea, but should've done some napkin math and or quick bounds checking first. Maybe a $5 privacy protector is better.
Here's an idea:
Maybe start by seeing if you can train a primary user gaze tracker first, how well you can get it with modeling and then calibration. Then once you've solved that problem, you can use that as your upper bound of expected performance, and transform the problem to detecting the gaze of people nearby instead of the primary user.
We’ve had GPT2 since 2019, almost 6 years now. Even then, OpenAI was claiming it was too dangerous to release or whatever.
It’s been 6 years since the path started. We’ve gone from hundreds of thousands -> millions -> billions -> tens of billions -> now possibly trillions in infrastructure cost.
But the value created from it has not been proportional along the way. It’s lagging behind by a few orders of magnitude.
The biggest value add of AI is that it can now help software engineers write some greenfield code +40% faster, and help people save 30 seconds on a Google search -> reading a website.
This is valuable, but it’s not transformational.
The value returned has to be a lot higher than that to justify these astronomical infrastructure costs, and I think people are realizing that they’re not materializing and don’t see a path to them materializing.
I was confusing it with TinyPilot, a hardware KVM made by an indie hacker Michael Lynch, that I think has since been acquired.
This is sensitive to the initial candidate set of labels that the LLM generates.
Meaning if you ran this a few times over the same corpus, you’ll probably get different performance depending upon the order of the way you input the data and the classification tag the LLM ultimately decided upon.
Here’s an idea that is order invariant: embed first, take samples from clusters, and ask the LLM to label the 5 or so samples you’ve taken. The clusters are serving as soft candidate labels and the LLM turns them into actual interpretable explicit labels.
But before "FOMO" used to be religious, more like "FOGTH", or fear of going to hell.
So people were mostly happy living simple, sweet lives. Spend time with family. Raise your children. Have a simple job. Just don't go to hell, so go to church, pray, don't sin. Everything will be great, as long as you don't go to hell.
Probably in like the 60s did consumerism become the mainstream religion and it's been taking over since. Now you *must* make more money to: buy a house, take a fancy vacation, live a luxurious retirement, etc.
The cringe "hustle culture" of today is because for some people, it's their spiritual fulfillment. It is their religion. Their main focal point of existence is to buy bigger, buy better. It's almost taboo to consider an early retirement, "omg, I'd get so bored! I'd go crazy!"
How dare you not follow my religion of selling B2B SaaS? You are not a go-getter, and I am. Did I mention I also have a podcast?
In reality, the dollar's true value will plummet. The FED is starting to lower interest rates again. We are likely going to undergo brutal inflation.
Crashing the economy is obviously very politically unpopular. The left/right will do whatever they can to keep this charade up, even if it means dooming the working class and throwing them some kind of bone to make them think they're ok.
The COVID pandemic was a good example of this. The working class got thrown a $2,000 check while there was billions given to bail out businesses/lots of fraud. Not a lot of people cared because hey, we got a $2k check... Even though that $2k check was not even close to maintaining their relative wealth pre-pandemic due to all of the government's inflationary measures.
There won't be a recession, it won't happen on paper. But the middle/working class will continue to be squeezed. And there will be programs to "rescue us." Maybe it's low cost home programs, maybe it's community college, I'm not sure. But I am sure it will never truly benefit the working/middle class, it'll just be a token to keep them from fully dying.
It's like me telling you: With using just audio I can trace your exact coordinates! ... By using an array of microphones in a room.
I still think it applies though; the psychological effect of being bald probably doesn't care much about the underlying cause.
My whole family on both sides is bald bald. Without it I would have been bald bald in my early 20s.
I knew there were risks of hormone disruption/mood swings/etc but I still prefer it over being bald. I took a calculated risk and I'm thankful that the drug exists. Please don't prosecute the people that gave me the only real, working treatment for my baldness.
-------
Results: Compared with the controls, an increased risk of suicide attempts was observed in patients with AA, with an adjusted hazard ratio of 6.28 (95% confidence interval, 4.47-8.81). Suicide risk remained significantly elevated in AA patients when stratified by underlying psychiatric disorders. The mean age of initial suicidal behaviors was also lower in patients with AA.
Conclusions: Patients with AA had a significantly higher incidence of suicidal attempts than controls, regardless of concurrent psychiatric illness. Further studies are needed to elucidate the pathophysiology of the association between AA and suicidality. In addition, dermatologists should be aware of the increased suicidality of patients with AA.
https://pubmed.ncbi.nlm.nih.gov/36921592/
----
Not enough time to size these in comparable numbers to this study, but would be really interesting.
Is it a net positive (reduced risk) over just going bald?
I told her she might have a UTI. It was not normal for it to hurt that badly to pee. She denied it. I bought her a UTI test, it came out positive. She was shaking. I told her we had to go to the hospital, she thought they were period cramps.
I call a teledoc. They video chat. She explains the pain shes feeling in her lower back means it’s likely a UTI, the infection has likely reached her kidneys, and we should go to the ER immediately.
In the ER we think they’re going to just give her some antibiotics and send her home. Nope. She throws up. Things go bad fast. Her heart rate is 160. She turns a color I’ve never seen a human before.
The next 3 days were so incredibly hard. But I’m so thankful to all the medical workers that were attentive to us.
Thankfully she makes a full recovery. For a week or so she was lethargic/tired but she’s fully healthy now.
A few months before I had read a story about a woman who’s boyfriend had died from a UTI because they went to a gospital, gave him some antibiotics, and he ending up dying at home because the infection was already too progressed to fight off at home.
Had the person who evaluated my girlfriend not evaluated seriously or just sent her off that could’ve been her. I’m so thankful they admitted her and took her care seriously.
It’s scary how quickly a UTI or some other benign infection can become sepsis. Take it seriously.
Vanilla WoW was also discontinued in 2006, and somehow players got Blizzard to bring it back in 2019.
Does that mean that vanilla WoW is a 300 IQ AGI?
To be more charitable, I get it, 4o is engaging/lonely people like talking to it. But that doesn't actually mean that those people will carry out its will in the real world. Nor does it have the capabilities of coordinating that across conversations. Nor does it have a singular agentic drive/ambition. Because it's a piece of software.
300 IQ in a vacuum gets you nothing. You need some type of status/power/influence in the world to have impact.
I think the previous "world record" holder for IQ is actually just a pretty normal guy: https://en.wikipedia.org/wiki/Christopher_Langan.
Just because AI is/can be super intelligent ("300 IQ"), doesn't mean it can impact or change the world.
Most startups are made of "high IQ" intelligent people trying very hard to sell basic $20/month SaaS subscriptions, and yet they can't even achieve that and most fail.
My biggest counter argument to AI safety risk is that, it's not the AI that will be the issue. It will be the applied use of AI by humans. Do I think GPT6 will be mostly harmlesss? Yeah. Do I think GPT6 embodied as a robo cop would be mostly harmless? No.
Instead of making these silly arguments, we should be policing the humans that try to weaponize AI, and not stagnate the development of it.
Claude Code on max plan is why I use Sonnet/Opus, but if ChatGPT pro includes codex usage I may switch.
Why: LLMs are increasingly becoming multimodal, so an image "token" or video "token" is not as simple as a text token. Also, it's difficult to compare across competitors because tokenization is different.
Eventually prices will just be in $/Mb of data processed. Just like bandwidth. I'm surprised this hasn't already happened.
SWE culture was very different in a low interest rate environment. Teams were over staffed. No new tech came around for a long time so everyone was focused inward on how to improve the quality of the craft. At my big tech company some teams had 2-3 people dedicated to purely writing tests, maintainability, documentation, and this was for a <1m MAU product.
Then boom free money gone. Endless layoffs year over year. Companies pushing “AI” to try and get SWEs to deprecate themselves. It’s basically just trying to survive now.
That wizard that used to nag everyone about obscure C++ semantics or extremely rare race conditions at distributed scale has disappeared. There’s no time for any of that stuff.
Like all cultures, this was all performative. People astutely observed how to say and care about the things that they saw, the people above them, saying and caring about, and mimicked them to promotions. That doesn’t work anymore, so that wizard culture is long gone.
It's a lesson most people learns the hard way after using PBs for a few months.
I stumbled upon this issue when trying to convert a recursive DFS to iterative because my recursive DFS was running out of stack space.
The solution produced by this iterative version was wrong, completely different from the recursive implementation.
It’s fascinating how many primitive, basic algorithms are probably implemented incorrectly but work just well enough that no one ever cares or notices… reminds me of how so many text books have an incorrect or overflowing version of binary search.
Just hand waving some “distributed architecture” and trying to duct tape modules together won’t get us any closer to AGI.
The building blocks themselves, the foundation, has to be much better.
Arguably the only building block that LLMs have contributed is that we have better user intent understanding now; a computer can just read text and extract intent from it much better than before. But besides that, the reasoning/search/“memory” are the same building blocks of old, they look very similar to techniques of the past, and that’s because they’re limited by information theory / computer science, not by today’s hardware or systems.
At every place I’ve worked at this is a solved problem: Hive+Spark, just keep everything sharded across a ton of machines.
It’s cheaper to pay for a Hive cluster that does dumb queries than paying these expensive DB licenses, data engineers building arbitrary indices, etc… just throw compute at the problem, who cares. 1TB of RAM /flash is so cheap these days.
Even working on the worlds “biggest platforms” a daily partition of user data is like 2TB.
You’re telling me a F500 can’t buy a 5 machine/40TB cluster for like $40k and basically be set?
https://stats.stackexchange.com/questions/185507/what-happen...
Context: https://x.com/GeoffLewisOrg/status/1945864963374887401
All this tells me is that: the “multiple embeddings” are probably mostly overlapping and the marginal value of each additional one is probably low, if you can represent them with a single embedding.
I don’t otherwise see how you can keep comparable performance without breaking information theory.
They can still have legal contracts with other companies, that stipulate that they don't train on any of their data.