My question always was, if breakthroughs are made though new research (like discovering tree of thought or many shot), why can't AI eventually do that research and give ideas to improve itself. AI is moving so fast, and we are still discovering new things, even about older models, so assuming even if GPT-6 is the limit, just having that model and using that model to research better ideas could be enough.

Another one is, we have no idea how multiple passes over the data influences big models. This has been done with extremely small models and results were pretty amazing. Imagine going though the data a million times. Or multimodality combined with big computing. And we have not even used tree of thought extensively for research.

It might be valuable to do few million tree of thoughts inputs for a given question, as even miniscule fractions of a % of improvements will drastically improve your model if the model costs 10 billion to train. Imagine after getting gpt-6, instead of training gpt-7, you spend 3 months of compute not to train the model, but to do extensive research using tree of thoughts, improving the algorithms and giving ideas or even doing things like lithography research, energy and thermodynamics that would improve efficiency of your computing cluster. But instead, everyone assumes that we can't use the AI to improve the AI itself.

To add to that, I think they don't want to just make the shield work, they want to make it economically. So I think we might be getting A LOT of reuse of first stage early on, then big focus on many tests of upper stage. While expensive, Starship Heavy will basically only have one configuration, so for some time, we might be getting like 10 to 1 productions of upper stages, with majority of flights not even carrying Starlink, just testing solutions. We are likely to see non contracted test flights around moon and so on as well, as SpaceX lists of requirements is likely much higher than that of NASA's HLS.

My thought was that reinforced learning caused it to not like answers with terrible people like Hitler, Stalin, Mao and so on, so it picked the worst person he could think of that was not in the common bag of "politically incorrect" names. People have heard of western politicians, and same for Asian so they are all automatically excluded but Modi is unlikely to appear in the west as a controversial figure, which is how he snuck though.

I do not believe there have been any changes. Maybe you got lucky before and now it's more of a normal behavior. You could try different prompts to try to get the behavior you want.

gpt-4o is pretty much same level as gpt-4 has been since the start. If you want major improvements, you need a prompt like "Have a conversation with me, and make your sentences shorter as if it was a discussion with a friend", or have "one shot" or "few shot" examples to more customize your experience for a given chat. A lot of those techniques will take up your free limits, so the more performance you want, the more tokens and messages you might need. I don't use it too much, but let me link you how I make it do better:

https://chatgpt.com/share/6d78ab00-b677-4e0f-8d0f-df73ae503629

This only has one prompt, but notice how I put emphasis on the item being used on mounts at the start and at the end of the prompt. When used well, "reminding" an important part can do wonders to get the answer you want.

On the other hand, here is an use of an extremely long and complex prompt to get better answer:

https://chatgpt.com/share/a740cb29-b20e-4d85-b2d4-fa97612a4a23

It works at the start, but as you can see, it gets worse with time, as the more tokens you use, AI generally gets worse. So sometimes it's good to just restart the conversation from scratch.

Then in here, we have an example of getting lucky and using AI for what it is good for, creative writing and assistance

https://chatgpt.com/share/18889e44-bc6f-4a6d-aed9-f131f30b2d8e

As you can see, this is an advice I possibly could get from a veteran DM who is interested in history, but it would be difficult to get so many options and for the DM to spend so much time explaining and giving options to you. And I think this was gpt-3.5 as at that point, as it was before gpt-4o released.

There are a lot of different techniques like few shot, many shot or tree of thought that require dozens or hundreds of messages, but depending on your use case, you might want to do it. But as you can see, I got some pretty good results even with a single prompt.

My bad, there are only 3. So there are no free ports currently, not IDA at least.

It is also worth noting that Ariane 6 uses 2 SRB, which add to the vibrations during launch, and the flight for Eumetsat was going to be a new version of Ariane 6 with 4 SRB, likely adding even more vibrations. It's not just speed and cost.

Can Starliner undock? The way I understand it, there is still a worry to undock it. It's not that it will fail, it's that it can fail.

Btw, Chinese as a country are sending the most cargo to LEO in the world, if you exclude SpaceX launches. They are willing to break shit, and risk their people, but not like it actually slows down the research, likely the reduced safety actually increases the speed. They just returned samples of moon rock from dark side of the moon too, which has never been done before. I'm just glad we have SpaceX.

Tofu architecture affects even the space program.

Maybe I misread or it was an old contract, sorry.

There is currently only one non Russian port left on the ISS, and that port needs to be free so cargo missions can deliver cargo to the ISS. I guess they could kick out cygnus, but both capsules are sometimes connected to ISS. The point is, they need that docking port, especially now that the Russians are not cooperating as well as before. If we actually deliver Crew Dragon to the ISS, there will be only one port left for cargo deliveries, which likely means that the amount of crew will have to be reduced to amount that a single capsule can carry back.

It's been a while now, so I might be wrong, but the LEO tank test is a specific test for Starship design required to get a milestone, not truly required as a test. But there are 2 hard points in the contract, that 2 finished versions of the HLS are supposed to perform a landing on the moon, before the crew landing. All other tests are either optional, or written somewhere else in the contract.

Compared to how most IT engineers explain it, yeah. It's not technical skill, it's just speaking skills.

I'm not sure, but every time people claimed a model got worse, when benchmarked, it did not have the drop in performance. Maybe because it incorporated contents of the benchmark into the dataset, but the benchmarks improved all over the board, which makes it unlikely there is a drop in performance.

The giveaway is that it was rly well explained.

I don't understand why people are saying gpt-4o is the apex of intelligence. AI video is improving so much with new models, and gpt-4 has been trained in 2022. Do people think that 3 years of improvements and massive increase in compute are going to do nothing?

The only way to know is to see Will Smith eating spaghetti.

There is a lot of emails going on in a company that big, and a lot of the templates will have a shit ton of code and considering how cancerous UI for Japanese sites is, possibly imbedded images. Presentations in big resolution can take a lot of space, same for video demos.

I hate Idiocracy memes as it's mostly people who have no idea how government works, and that hate "the other guy" but man, it is hard to not compare the fucking movie.

The goal is for the ticket to be a price of a house, so that you can sell all your possessions on earth and travel to Mars, and work there. with likely 30+ million cost of the trip, 100 passengers per trip, this is likely to be possible. Although it's most likely that vast majority of colonists in first few decades would be funded by governments, or that they would have massive incentives by governments that want to have presence. Mars is almost 100% going to have it's own government eventually, so countries will want their own presence there to level things out.

He might cash out some of it to pursue other projects. He was hoping for a while now that he can eventually leave Tesla to pursue other projects, but too many people are still worried Tesla might not do well after he leaves.

This is stock, not cash. Don't know the details of the case, but you can just dilute the stock and issue 56 billion in stock to Elon.

Ok, I'm a nobody, so I might get this wrong, but this is my general understanding of life on Mars.

There is not a good reason why there would be current life on Mars. What rovers are generally looking for are signs that there used to be life on the planet. There are general signatures on a planet that would indicate an active life, and those do not exist on Mars. If life on Mars exists, it's either deep underground or in secluded, rare places on the planet, and while not needed, human exploration would drastically improve the rate at which we would be able to search though Mars.

Another matter is that even if we have found life on Mars, it would be quite trivial to tell that it's Martian life, not just by dna tests, but also though comparing that life to life no earth. Also, the intentional terraforming life we would bring to Mars could be genetically modified to have a stamp in the DNA to show what lab it was made in, that way we would have 100% certainty what is the source.

Also, while there are other planets, the reasons why it's a preety good place to set up a colony, is also a reason why having active life is unlikely. Life needs a bit of stirring, a bit of danger. Floods, fires, earthquakes and weather, were essential for life creation on Earth, but are not great for humans right now. Besides some light winds, Mars does not have those, which makes it good for us, but bad for life. Way more likely places for finding active life is Titan and Europa, as both of them have active weather, seismic activity and liquid water, and those places should not be colonized, at least not by humans, for a very long time.

Lastly, while search for life is important on Mars, more important is securing 2nd planet for overall humanity survival long term, so that we could have a chance to find life on other planets and moons, if it exists.

But good and inquisitive questions. I hope that way more rovers get sent to Mars thanks to Starships as well.

Point is, both options suck, and it really does not seem that Starliner is busted yet, as long as they can keep their current thrusters just for orbiting away from ISS, they can actually lose more thrusters and still be able to deorbit. You need the most thrusters to dock to ISS, then to undock you need less, then to rendezvous with ISS you need even less, and you need the least to deorbit.