mrguyorama

People are saying correct things (The bottleneck is downstream at phase 3 trials, labwork is slow) without seemingly recognizing what those things actually mean or imply about any LLM assistance.

The reason that Phase 3 trials are any sort of bottleneck instead of just checking boxes is that drugs that make it to that point are still extremely likely to fail and not be productized. The phase 3 trial is hard, expensive, and slow because it's the only strong filter we have! After phase 3 trials is the only point at which you become confident that this chemical doesn't do crazy unknown things to humans.

This is because the previous trials are not sufficient enough and high fidelity enough analogues of real human biology to be able to prevent that outcome!

Literal living human cells in a petri dish are not sufficient to tell you if a drug will work or not because the entire human system is extremely chaotic and interconnected. Even animals "99%" similar to humans are not similar enough to prevent unpredicted catastrophic outcomes in that full blown human trial.

We do not have enough knowledge anywhere about any of this for an LLM to be useful at all, because we have simply never collected the data necessary to be able to predict outcomes of novel interactions in the full scale system.

We haven't quantified the human system nearly enough.

You want to fix this? Don't go buy tokens, spend $2 trillion on grad bioscience students quantifying literally every protein and gene and structure and behavior and interaction in the human body. You literally have to brute force this problem.

You can't use "simulators" because there's no data for the sim to be evaluated and verified against in the first place. Our knowledge is too sparse.

LLMs cannot magic new information from limited data. Even magic "Superintelligent" AI wouldn't be able to fix fundamental deficiencies in what we know about the human body. There's this extremely silly and weird belief that it will be able to figure everything out "from first principles" and it's driving me up a wall.

Atom level simulations of anything in our reality are computationally infeasible at scales larger than "One cell", and you should expect that to not change since there's not expected to be a magic machine that can somehow undo computational complexity. Well, there is, in that our reality is that machine, but it's not very amenable to isolation for experimentation.

Fundamentally, there seems to be a serious fault among "AI" boosters of deeply believing the "Great man" theory of reality, that everything just needs a magically "smart" enough person to solve, as if someone clever enough could somehow trick physics itself. It's extremely ignorant of reality. There are problems in the current world that no amount of "intelligence" can solve, because the problem is not constrained by reasoning ability in the first place. A lot of bioscience fits that.

Simulating physics to the level of making meaningful predictions in chemistry is computationally infeasible. So no, we will not be atom-level simulating humans any time soon and LLMs cannot change that.

>The precise mechanism of action of dimethyl fumarate is not clear.

And here is the real problem. Instead of burning gigawatthours of power with GPUs, if Anthropic really wanted to help bioscience, they should just be throwing their $2 trillion at making it possible to live off of studying things like this.

We aren't in a place where a magic "Intelligence" could just look at the data and solve all our problems because we have never funded the collection of that data in the first place

We spent a lot of money on the Human Genome project and got large dividends from it but seemingly never reinvested! There's so many clear and straightforward questions we should be answering in biology that we just aren't, because nobody wants to pay for it.

LLMs aren't the right course here!

Would we have been allowed to dedicate $2 trillion dollars to mathematicians doing this?

I always find this funny.

Max Planck himself got to watch physics radically change multiple times, through avenues that didn't involve the old guard dying.

As long as they don't get fired or jailed for even murder, cops don't "need" to do shit at all. If you survive, you might get free money for your experience.

The term among lawyers for this is "you can beat the wrap but not the ride"

You can be convicted of resisting arrest, with no underlying crime.

>Taxi medallion laws were created to artificially screw with the market

My area had no taxi medallion laws or system.

That's part of the problem. The excuses were always bullshit made up from generalizations of local problems. Uber wasn't needed to "Bust the taxi monopoly" because in most places there was no taxi monopoly, at least, not until Uber showed up and made one.

The excuse was never meant to be actually valid though. It was always just a simple "gotcha" that could be employed as a thought terminating cliche by people who had already made up their mind on which outcome they wanted.

You ever play that really old Zachtronics game where you have to design circuits at the silicon level? (KOHCTPYKTOP: Engineer of the People)

Give me an inverter and single logic gate and I can by myself recreate computing. Give me a silicon wafer and dopants and I no longer know what I am doing. The structure is 3D and also somewhat abstract and I fail to be able to maintain the mental state to understand it.

I don't understand my mental barrier. I get the abstract theory and rough physical concepts, but somehow I can't pull features out of a micrograph like Ken Shirriff can.

The fun thing about society and nations is that rolling over and giving up is literally the only actual failure state.

Congress is capitulating it's duties at least partially because anyone who tries to block Trump as a republican gets voted out. Oh sure, they stirred this shit up for 50 years and just reaping what they sew, but the train is free-running now and they can't really stop it.

The Voters want a Trump King. A government cannot ignore it's own people forever.

If the voters reward corrupt and horrific behavior, and punish law and order behavior, what the hell can you do about that?

You mean sports betting enters the era it was in 100 years ago, right?

https://en.wikipedia.org/wiki/Black_Sox_Scandal

There's a reason why we didn't do things we are currently doing, and it had nothing to do with "technology". We've been here before.

The industry makes nearly all it's money from problem gamblers.

Know that these companies internally have models to predict how profitable you will be as a customer, so that they can best nudge those customers who are likely to be profitable.

This model is identical to if you trained a model explicitly to find problem gamblers. They are purposely building systems to find and exploit problem gamblers.

Because that is what the industry does. It finds and milks addicts. It is impossible for a for profit company to exist in this space and not somehow optimize towards finding and exploiting addicts. It's just more profitable. The people who drop a few hundred on a few hands of blackjack once a year cannot sustain an industry.

Extracting all possible wealth from 20% of the significant fraction of the population that plays has enormous negative consequences for all of society.

The only reason you don't ban gambling outright is that it's just too easy and straightforward to run an underground book, so it's a trivial market for organized crime to own. And that isn't good.

So you should have a state run institution that exists entirely to just barely serve the literal addicts, does absolutely nothing with the "income" (preferably just pays it all out, zero sum), and is rather unattractive to people who haven't gambled before. You make betting as boring as possible.

Those push notifications are tuned to trigger someone who might become a problem gambler. They are actively trying to get you addicted.

This thing has actually been around for 20 years now, right?

"Future weapons" on the Discovery Channel showed it off back in 2007ish.

It's been in development hell that long because it just sucks in a million ways. Water cannons and other tools just do it's job better, and for less contract dollars.

And that's ignoring anything around microwaving your own citizens being fucked.

Short form video inherently has fundamental differences to what it will support and optimize towards. Just the selection effect of who prefers to watch shorts over long form content produces a different optimum that IMO is far worse.

Take a population, and then remove anyone who is willing to invest more than 60 seconds on a idea, and then sample the "preferences" of that population.

What did you expect to end up with?

Creators who I really like still end up producing trash on shorts because that's what gets traction, because the problem is what is actually liked by someone who wants to just mindlessly scroll 30 seconds at a time is really bad

What could possibly go wrong with a (I think) VC funded news aggregator that becomes your authority on what "Right" vs "Left" thought is and how to interpret a bunch of news without actually reading any of it?

Where could that possibly end up causing problems?

You understand then that there is absolutely no possible fix for that right?

If your grandma will follow current instructions, she will follow these new instructions including a 24 hour wait.

Your grandma will follow instructions to walk down to walgreens and buy a brand new device if they tell her. The solution has nothing to do with the device and everything to do with you needing to be an adult and figure out how to get your grandma to not follow directions of scammers.

This isn't a tech problem and you cannot solve it with tech.

DOGE isn't a source so don't cite them. They just made shit up.

The American public is stupid and wrong and their perception is not useful here. They will believe the government wastes 50% of every dollar because they have been told that, even though it's just stupidly wrong.

The GAO is actually the only information you provided. They are trustworthy and happy to be critical of every administration, and they don't get enough love for their important work.

I don't know if you realize this, but 3-7% literal fraud is actually a pretty good number for a very complicated system. 15% "waste" where it's hard to even define what "waste" is is also a good number.

Visa and Mastercard accept far higher for a far simpler fraud profile.

Also you said "This guy thinks waste in government is only 15%" and then the only data you provided has to wave many hands to get up to 13-20%. What the hell man?

That's generation

How do you explain increases in the cost of distribution?

Um, we did take starlink away from Musk.

That's the point of StarShield. That's the point of government in general. We have the FCC and rules around space launches.

If the current US admin wanted Ukraine to be able to use something like starlink despite Elon's personal beliefs, then it would.

Elon is taking the blame because he is being allowed to take the blame. It's a distraction.

Trump wants Ukraine to stop hitting refineries because apparently he has only just now realized that blowing up refineries in Russia affects gas prices, which was the entire reason that Biden prevented Ukraine from doing such things. Which implies the hilarious caveat that Trump at some times has been more beneficial to Ukraine simply by not understanding the global economy and second order effects.

But Trump desperately wants to massage gas prices for a few weeks, which is why he does the stupid things like suddenly leaning on Ukraine as hard as possible and buying a pitiful amount of Russian diesel.

Allowing Starlink to limit Ukrainian utilization has absolutely nothing to do with any the bullshit "But ITAR" claims of weird people who don't seem to realize that wasn't a problem previously, somehow. It's about putting more pressure on Ukraine to not blow up refineries.

If there was an ITAR issue, then it wouldn't have taken months of coordination to purge the Russian Starlink terminals.

Our government is holding Ukraine's arms behind it's back against a literal invader because the American public does not care about Ukraine's basic sovereignty nearly as much as they care about an extra 25 cents per gallon of diesel. That's it. Oligarchs or not, that's the core issue causing most problems in America. People don't pay attention, don't give a fuck, don't learn, don't think, and then complain that everything is going to shit.

If you don't want to do these things, you have to elect different people. For as much money as there is in our elections, they are still decided by vote, not price. A lot of work each election goes into convincing people to stay home, entirely because voting works

I have heard from random lawyers that if the facts are in your favor, get a bench trial, if you are actually guilty, you can maybe sway a jury.

We don't have juries because they are smarter or more fair or more just on average than a judge. We have juries entirely due to the concept of jury nullification. It's a formalization of the concept the founders sometimes expressed where "Yes I did it, but that doesn't mean I should go to jail" is an intended outcome. It's explicitly that Justice isn't the same as "follow the letter of the law".

They put a lot of effort into building a very secure safe, thick steel walls, can survive a fire forever, too heavy to move.

Then they locked it with the key you find on a chew toy for a baby.

No, the control scheme is just bad. Tank controls just don't work for shooters, especially with how they were developing at the time.

Similarly, old shooting games before doom sometimes had no dedicated control for strafing. They didn't seem to realize how important it was. But it's bad.

If you boot up Goldeneye and Perfect Dark in an emulator, you can set them up with the control scheme that uses the C buttons as the second "stick" and remap things to give yourself near real dual stick controls like in modern shooters, and it's night and day. The games stop feeling clunky. The autoaim becomes actively harmful.

I did this once to my brother in multiplayer and there was absolutely nothing he could do to combat my fluidity of movement. It was trivial to just circle strafe him.

Having the two axis of a control stick do radically different things (turn vs move) is terrible. It's like playing a racing game where pushing up on the stick is your accelerator. It's just hard to use that input in a controlled way while also trying to deal with the other axis.

I think a lot of people feel like social media scratches the social itch they are feeling.

I think that is simply an illusion though. I have high social needs, and no amount of social media ever fed them in a healthy way.

Just because you are hungry, doesn't mean you should immediately reach for the bag of chips. It's important to reflect how often you do so, and whether that is getting you where you want to be in life. It's hard but really really important.

A lot of people say they are very lonely despite spending all their time on social media. They then change nothing.

It doesn't work

The problem, as it is for a shitload of women, is that the delta between "Minimum amount of calories needed to not be somewhat malnourished" and "calories you eat now without losing weight" is sometimes zero. Short women working desk jobs are fucked

I was with a girl that barely ate, and couldn't get her weight down for about a decade because of that. Eventually she got really excited about a local dance studio and suddenly her weight dropped like a rock. She had to exercise more, but that was miserable until she found exercise that was effective and actually clicked, then it was trivial

Meanwhile it was trivial for me to eat less, because my budget should be around 1800 calories a day, and I was eating 3000. It's easy to cut out an entire bag of doritos. It is less easy to literally starve yourself.

People seem to forget as well that when you cut calories, your body will scream bloody murder until it re-adjusts. My body was convinced I was starving while I was 50 pounds overweight and eating 2000 calories because I had previously eaten more. Your body did not evolve in an environment where that was ever the wrong thing to feel! "Please eat more or we die" is one of the most powerful biological signals.

That these medications cut that signal is entirely why they work! If people just "couldn't stop eating" then not having hunger or appetite signals wouldn't have worked either.

That you believe those people with all the money are "successful" is actually the entire problem.

Yeah, I'm not stupid rich, but the evidence is that I'm far better put together than Elon Musk emotionally, and even happier in regular day to day life. Which is hilarious because I just had my entire life upended.

But I've never felt a compulsion to pretend to be good at a video game to impress strangers on the internet because I have responded to difficulties and personal failings by bettering myself. I embrace recognizing my faults, fixing them, and knowing that I am human.

Do you want a child like me or a child like musk?

What do you run an open model on when Nvidia stops selling you compute?

The problem is that a lot of people don't want things to be deterministic because they want the risk of getting one over on you. They choose a worse system because essentially gambling gives them a thrill.

They work in GPS contested environments by reverting to expensive INS and TERCOM systems.

The US had expensive systems because INS and accuracy is expensive without GPS.

GPS will be denied in any future war environment. That's why we stopped work on things like the Ground Launched Small Diameter Bomb. It was useless in Ukraine due to GPS denial.

Why do you think the US and it's empire pretending to have the moral high ground is "unique"?

Great Britain at the height of her power was insistent that they were "Civilizing" people they oppressed, and Rome did much the same.

Physics. The incoming gets to play way easier games than your interceptor.

Harder to shoot an incoming bullet than it is to fire that bullet in the first place.