clay_the_ripper

Please call me when Yann is right about literally anything.

No one has been more wrong and for longer about anything than Yann on LLMs.

in the first few years of any new technology only highly innovative companies (startups) use it.

people still were using punch cards when tape was invented, and still using tape when floppy’s were invented

and those were relatively small innovations not requiring substantial changes to the work. AI is another ballgame, this makes me more bullish on it than ever.

You know i used to think Ray Dalio was really smart, but literally every single article he writes is exactly the same

“this argument is laid out in my book” blah blah blah

This guy is so obsessed with being right about his sole theory that he has staked his entire identity on it.

When you only have a hammer, everything is a nail - Ray has been shopping the same theory without updating anything whatsoever about it for over 10 years.

this.

it’s genuinely shocking how much time is spent coming up with ideas like this, implementing them and making everyone report on it - for no outcome whatsoever

Doing hard things has always been, and always will be, hard.

Building a static HTML page was “hard” in the 90’s. It took actual skills.

Any piece that gets easier automatically opens up more hard avenues to tackle.

no one is willing to pay you for easy.

This view is now consensus, which is highly comforting as consensus is always wrong.

when every car around you is even bigger than this, you’ll appreciate not driving a minuscule car in america.

it is terrifying

Hiring illegal immigrants is risky and hurts businesses that choose not to cheat.

This article makes it out like neither the immigrants or the businesses that hire them have broken the law - when in fact they all have.

I’m no fan of ICE, the tactics are disgusting. but this article does not mention even once that both coming illegally and hiring illegal immigrants is a serious crime… in any other country you would deported for over-staying and no one would bat an eye. Why is the United States any different?

The flip side is: efficient capital allocation is better for our economy as a whole than trying to save certain jobs for emotional reasons.

Struggling businesses that can’t operate effectively and provide a poor return on capital should be shuttered - the only way to deal with the mountains of paperwork involved is to incentivize very smart people to work at very bad companies. smart people don’t like to work at failing companies. how to reconcile this and ensure efficient capital allocation? huge monetary compensation.

for an incredibly long article on a niche topic readers are unlikely to be familiar with, astonishingly the author never bothers to describe what ABA actually is or provide more than 1 vague example.

this is a cool idea. tried a few searches that were not confidence inspiring: missing prices, blurry, weird photos.

the searches in case you want to take a look: “dog collar” and “cat scratching post”

good luck with the product!

“the cave paintings of nudes are corrupting the neanderthal youth”

and yet - somehow we always survive.

If I am in a city (sf, ny) I am putting in my AirPods and booking it.

Why?

The only people that try to interact with you are either homeless, trying to sell you something, or trying to rob you. Or all 3.

It’s a shame as it really changes the experience of a city and makes everyone wary of everyone else.

If someone could actually enforce laws and clean up these cities, people might consider taking their AirPods out.

Until then, I’ll have my AirPods in, sadly.

LeCun has been saying basically this same thing since 2021-22.

It’s a lot easier to say what something is not than what it is.

His comments now are really sounding a lot more like sour grapes from someone who is ultimately behind in this race.

When was the last time he launched something truly new and innovative that isn’t just a copycat? Genuinely asking.

I used to pay a lot of attention to him but he’s like the Jim Cramer of AI. Always wrong.

Hey this is cool - our application helps real estate investors select locations. Our users would get a lot of value from this. Be happy to feature you!

This is totally unsurprising, but an interesting example of how people are expected to conform to “normal” behaviors. People that fall outside this narrow definition of what’s acceptable are still labeled “deviants” which is basically what the university is saying.

Also, the first amendment applies to the government restricting speech. A university is not the government so they are free to say that to be employed by the university you must not also do porn.

Bold moves by this guy, way to go in not being ashamed of who you are.

China's strategy has now changed: they used exports to build massive growth in their economy, but they are now using their exports as a kind of economic weapon that can be used in much the same way that Amazon crushed their competitors by artificially lowering prices to put others out of business...then raising the prices once they were the only game in town. China is doing something similar by making their currency artificially cheap to boost exports and by subsidizing production. Developed nations are taking note and building up their own industries again that were largely decimated by offshoring. Without protectionism, Chinese companies will threaten domestic businesses that are critical to the functioning of our economy. China is trying to save its ailing economy by dangling ever-cheaper goods to the world, but the world should not fall for this ploy.

Netlify for the front-end, google cloud for extended storage and Akamai for hosting. Akamai is great as they are less customizable than AWS (which is a good thing if you don’t need extra config) and much less expensive.

I can relate to this, especially on a side project where there is no one looking over your shoulder or holding you to a deadline, it’s very easy to spend too much time on the wrong things.

The stress you feel is the feeling of not having enough time to actually do the thing you’re wanting to do. But it’s important to recognize that the reason you don’t have enough time is because the thing you are working on, will not move the needle on the side project.

Deep down you know this, but it doesn’t matter because it’s hard to stop when you get stressed and emotionally deregulated. This is a very common trait in ADHD.

When I start to feel that stress, I typically will try to force myself to step away from whatever I am doing and take a five minute break. After I take a break if I still want to work on whatever it is, I force myself to put an end time on it, and also to relate the task back to a larger goal that will move the needle on my project. If I can’t do either of those things, then I stop the task. Hope it can help, it’s not easy.

See if you can get yourself obsessed with working on things that will move the needle like a new feature instead of rewriting old code.

This is something I would use - not to steal but so I could listen to certain podcasts for sleep. The host intros and ads for this certain podcast are terribly distracting from what is otherwise incredible content.

My experience as a bootstrapped saas founder:

-at the early stages, usage models didn’t work for us because it attracted customers who were uncommitted to using the product

-the advantage of a subscription is that it filters out people who just want to use something occasionally (which is fine once you’re huge, but at the beginning you want people who are committed)

-I found it impossible to make enough money on the usage model. It’s maybe great for users but if you go out of business your users won’t be able to use it anyway

-we do automated real estate reports. When we charged per report people ran a lot less reports and tended to be stingy about running them. We wanted to encourage users to use it so an unlimited subscription made users happier, even if it meant their per report cost was higher. Also, we got tired of people computing the cost of a report mentally. Since the margins per report are good, an unlimited subscription works much better.

-once you reach a certain size having free and lower usage tiers can be a good thing but first you need a bedrock of recurring revenue that you can count on. Unless you have boatloads of VC money to burn.

-this will vary greatly depending on the specifics of your product. For example, developer tools that are easy to test out, but require a credit card to go to production make a lot of sense, but that was not us.

-if you can sign up very large customers that you know will use it a lot, usage models can make sense. But this requires a really really good product that already has PMF and requires that you can sell to mid market or enterprise customers and the sales cycles can be long. You can easily go out of business waiting to sign those customers.

Tinfoil hat time. The recent gpt2 chatbot that everyone thought was a new open ai product - could it be?

“ You start with the gpt2.c pure CPU implementation, and see how fast you can make it by the end of the course on GPU, with kernels only and no dependencies.”

Remarkably similar nomenclature. I give it 1% chance this is related. I did play with that chatbot and it was smarter than gpt4 whatever it was.

It does seem like this is a possible method to test if an LLM has your data in it.

People have found other ways to do that of course, but this is pretty clever.

This is an interesting analysis but it ignores the fact that although a cheap drone can destroy an $11MM tank, 2, 000 cheap drones do not give the possessor of the drones the same power as having a tank.

Even if Iran or whoever can produce $800 rockets that cost $50k to destroy, we are able to outspend them by such a gigantic margin that it still won’t matter. They could never bankrupt us this way. They will go bankrupt first.

We spend double Irans entire GDP just on defense. Even if they switched to 100% wartime production, we could outspend them forever. Who cares if a tank costs $11MM? We spend nearly $1T per year on defense. Good luck beating that with cheap rockets.

War doesn’t really work this way - we spend $1T on defense but the economic advantage to the US by being the de facto world power is so far in excess of that the cost would be worth it at 10x the price. We could “lose” a million dollars per missile and it still wouldn’t matter.

I think this fundamentally misunderstands how to use LLMs. Out of the box, an LLM is not an application - it’s only the building blocks of one. An application could be built that answered this question with 100% accuracy - but it would not solely rely on what’s in the training data. The training data makes it “intelligent” but is not useful for accurate recall in this way. Trying to fix this problem is not really the point - this shortcoming is well known and we have already found great solutions to it.

Well…anybody got an invite? Would like to give it a whirl.

I have never been happier to get rid of a car than my model Y lease. For anyone considering a model y wish I had known:

-the turning radius is terrible, very annoying to drive and maneuver in parking lots

-the visibility without the cameras is poor, esp in the rear

-autopilot is frankly dangerous. I’m confident with it but handing it to my mother/wife is frightening

-the steering is very sensitive, esp at high speed. Even on comfort mode, it’s twitchy

-the ride quality is absolutely garbage. Such a hard ride; and yes, it makes it handle like it’s on rails. But how often do you want that vs potholes to not feel like your back is on the road?

-availability of chargers. The charging time is bad enough, but then you have to wait to even get on a charger more than you’d like. Not practical at all.

-if you have a partner of the impatient variety, they are not going to be happy when road trips take an extra hour or 2 for no reason.

-customer service is bad, the app is bad and the entire ownership experience is terrible if you’re used to dealing with any other manufacturer.

-no physical controls is super annoying

-phone key is buggy

-lots of software bugs and build quality issues.

As a resident, I don’t support this at all.

Waymos are annoying to have on the road - they go extremely slowly (with good reason) and make weird traffic moves.

Do we want our roads getting clogged up with super slow taxis that steal jobs from ordinary people and give them to the tech industry?

They already have it pretty well defined on the ads side. If you run an ad and it’s about “issues of national importance” or social issues of any kind (non-profit, education, politics, public health etc) then the ads require additional review.

Actually it kinda seems like they are just going to blanket ban everything political, regardless. They have said publicly they do not want to be the arbiter of truth - so this is them trying to remove themselves entirely from that role.