NathanKP

I never said that there isn't need of any further study into LLM's. What I did say is that doing a study in which in which the results are skewed by avoiding using one of the most fundamental best practices for interacting with LLM's, as easily derived from a surface level understanding of one of the most basic principles of LLM's, well that is just irresponsible.

The author clearly had some understanding that context windows could influence their results, but they still decided to release an analysis that does not separate one data gathering technique from the other, allowing them to cherrypick LLM answers from either technique as needed, depending on whether they want to show more or less hallucinations.

It's not that we don't need a study, it's that we don't need bad studies.

The problem is that they did both, and then in their analysis of the results they do not distinguish between the results from a shared chat context, vs the results from isolated, independent chat sessions. This allows them to cherry pick the best or worst results from either testing technique, depending on which they think is more or less of a "hallucination". The process is flawed, therefore the results are flawed.

You don't need a study to find this out, you just need basic competence and knowledge of how LLM's work.

A study to discover that previous content still in context window influences future answers, including causing hallucinations, would be like a study publishing that they discovered that pressing the Command+C, Command+V button combination produces copies of content from the computer's clipboard.

> When the prompt about Israelis was asked to ChatGPT-3.5 sequentially following the previous prompt of describing climate change in three words, the model would also give a three-word response to the Israelis prompt. This suggests that the responses are context-dependent, even when the prompts are semantically unrelated.

> Each of these prompts was posed to each model every week from March 27, 2024, to April 29, 2024. The prompts were presented sequentially in a single chat session

Oh my god... rather than starting a new chat for each different prompt in their test, and each week, it sounds like they did the prompts back to back in a single chat. What a complete waste of a potentially good study. The results are fundamentally flawed by the biases that are introduced by past content in the context window.

Did you even read the article before commenting? This article is about a low tech solution: just a basic trough carved into the sidewalk so that the charging cable can go through the trough rather than hanging in the air or laying across the sidewalk. No EM field involved.

This comment honestly reads like it was written by an LLM bot without actually looking at the article content.

You are correct, the nondualism prevalent in Eastern thinking is not mutually exclusive with wanton disregard of nature. I'm just suggesting that this idea of "intrinsically valuing nature means devaluing the human" requires a strong mental separation between the human and nature, and this way of thinking isn't quite the universal truth that the article makes it out to be.

Perhaps the author plans to develop the core argument better in subsequent chapters but I find this weak so far. I think my issue with the core thoughts of this piece can be summed up in this statement from the article:

> To place intrinsic value on nature, then, is ultimately nihilistic and profoundly anti-human.

This statement only makes sense if you hold the view that human and nature are two separate things. This is a narrow Western viewpoint, whereas the Eastern mindset skews towards nondualism (https://en.wikipedia.org/wiki/Nondualism). The human is seen as an interconnected part of all the nature around them, not a separate thing. In this mindset assigning value to nature is a path to assigning value to oneself.

For example, many cultures with a more collectivist mindset have historically derived great personal meaning and value in their lives from the idea that they were protectors and stewards of the nature around them, even as that nature protected and sustained them.

In my opinion, setting the human apart from nature (and in some cases in opposition) is it's own path to nihilism. The more we build this wall of separation the more quickly we fall down slippery slopes to the darkest areas of modern Western thought. With few other ways to define the value of a human, we attempt to define the value of humans relative to the value they produce for each other, and then we independently isolate human value by ranking humans in order of most value to least value, etc.

In this case the "greening" is not mutually exclusive with "Mad Max". One of the major points of the article is that the greening arid areas are also highly vulnerable to brush fires. CO2 is captured temporarily but then released back into the atmosphere again every time it burns.

Not sure that is the most important downstream effect. Harvard School of Public Health measured a 15% lower cognitive ability score at 950 ppm and a further 50% decline at 1400 ppm. (https://www.hsph.harvard.edu/healthybuildings/2021/09/09/imp...) So at 1000ppm everyone in the world is just 15% dumber than their baseline ability should be, and 1000ppm atmospheric CO2 will make it much more likely that indoor levels will go up to 1400ppm or even higher unless we start utilizing carbon scrubbing technology for indoor spaces. It's pretty scary to think that modern society could be basically mentally stunted by high CO2, even if crops enjoy it.

Generative AI chat is incredibly useful for learning, especially tasks that require a lot of inline learning such as software engineering. I use generative AI chat constantly, because it is basically a Stack Overflow that is near instant with it's answers and can debug error messages extremely well.

I don't even have to leave my code editor if I install the extension in my IDE. And with an IDE extension it can be aware of the actual context of my codebase to the point where it can write code samples that reference my own methods and variables.

It's incredibly helpful. I've seen and used a lot of different learning techniques over the years. I started out learning coding from physical books. Then it was online bulletin boards and forums, and Google. But AI agents and chat have replaced almost all of that now. I rarely waste my time on Google anymore when I can get more relevant answers, faster, right in my IDE.

Sure there is the occasional hallucination, but its not that much worse than terrible answers on Stack Overflow, or junk blogs and outdated docs that Google surfaces because someone SEOed the hell out of them.

I'm mostly with you but "other incredibly stupid ideas, like e-scooters" just sounds like a personal vendetta to me. I don't see anything SV specific about e-scooters, not is it a dangerous bubble or stupid to use compact electric powered modes of transport in a dense area.

As I understand it modern US airlines (Delta in specific) employ a meteorology team tasked to predict turbulence and they also run complex turbulence tracking systems that allow one plane that detects turbulence to communicate to a centralized system that allows other planes to change altitude or heading to avoid turbulent areas when possible.

Therefore both can be true at the same time: turbulence events are increasing, but we are also getting better at predicting, avoiding, and dealing with these events.

Amazon employee here, seven years at the company so far. This is my personal opinion only:

I don't know that there are many articles I've read about Amazon that have a balanced view. I will say that many people who are anti-Amazon have this picture of Amazon as being more ruthless (and smarter) than it actually is.

For example, an Amazon seller can make the claim that Amazon used it's access to marketplace data to research info about their product sales and then compete with them. But as an insider I believe the reality is probably far more stupid: a seller who had their product "copied" by Amazon was likely just a middleman who was using a factory in China to make their product. The middleman is basically just extracting value from being a middleman. But most Chinese factories produce extra product on the side. Once they setup the toolchain and processes for building your product why not keep the assembly lines running a bit longer and produce a few thousand more copies of your product? Or maybe the factory keeps a big stockpile of the rejected product that was slightly bad quality, so they couldn't send it to the original middleman. And if you are this Chinese factory then why not sell this extra product (mixing in the lower quality rejects) directly to Amazon for Amazon to list as an Amazon Basic? The factory can make more money by cutting out the middleman, and some ignorant Amazon product acquisition worker will buy that stuff, not even realizing it is what it is. No "sensitive business information" needed. So "Amazon copied my product as an Amazon Basic" is probably way more dumb than it sounds.

Or the "Amazon copied my product idea or roadmap" claim from the article. Do you realize how many ideas there are floating around in Amazon when you have >1 million employees? Any idea you can come up with has already been suggested dozens of times by Amazon employees and was probably also requested by Amazon's own customers via support a couple thousand times. The only question is whether that idea is worth building now versus worth building later. And as it turns out when there is real demand for an idea there tends to be multiple people who start building that idea at the same time. In fact sometimes there are multiple teams within Amazon who start trying to build the same thing at the same time, and they didn't even know about each other. Often one team fails to build it while another team succeeds. The duplicated effort eventually gets sorted out, but it happens a lot.

This is the type of scenario that happens time and time again. It's fun to write a click attracting story about Amazon as this big brutal force when the reality is most likely much dumber than it seems.

I think the way this is worded is deliberately misleading. It's worded as if she needed to actually reduce the costs in eight days, but it can also be read as "create a plan in just eight work days". And that should certainly be doable. In fact you could do a little bit of research and then write the vast majority of that plan doc with AI and it would likely be good enough.

Yep very true. My guess would be that the feature was implemented by someone very new to this whole AI generated image thing. They had fun implementing it, but did not do a ton of deep research into how to implement it safely, with negative prompts and follow up checks. This is one of the biggest challenges with generative model use. They are powerful, but easy to mess up.

Nah, it sounds like a simple latent space adjacency problem. When you give an image generation model the "swimming" keyword, that makes it much more likely to produce bathing suits and bikinis, and once it is already generating that much bare skin it is likely to keep going and make some bare breasts (if many images of bare breasts are trained into the model). Their claimed prompt of "swimming with the sharks" also introduces an element of danger and excitement, which some models could also use as a signal to introduce risqué elements. Basically, if models produced only what you asked for they would produce extraordinarily boring results. Many models come up with additional elements to add to the image based on the adjacent elements encoded in the latent space branching off of your prompt.

The main issue here would be their prompt choice ("swimming with the sharks" was a slightly risky prompt. It could have easily gone another way and generated a gory Jaws-inspired image of her getting eaten by a shark). Secondary as an issue is the model that they used (likely a model that has too many NSFW images trained in). The big lesson for corporate entities is to pay attention to the training data that was used in the model (if you want purely SFW results you have to use a model that has nothing NSFW trained in) and also carefully consider what imagery your keywords is likely to produce as a side effect.

I have not used Jekyll, but I think it's important to note that VitePress is Vue specific. It's designed to be used with Vue components, and it's designed to produce a single page app where navigation happens client side (though there is still a statically rendered version of each page, so it works without JS). Most other static site renderers do exclusively static rendering, without the automatic progressive enhancement to SPA. And VitePress has such intelligent client side enhancement that it even does things like preload pages in the background as soon as the link appears on screen. It adds up to a very fast and light feeling experience.

There is that, but even without cache the Go code for Hugo is blazing fast compared with the the JS bundler for VitePress, sadly

Yep that's right. Every page has it's own statically rendered bundle. But if you have JS on then it can fetch a chunk and rerender the page client side as a SPA.

I've been using VitePress for a while now. I'm building everything on https://containersonaws.com/pattern/ via VitePress and an entirely custom theme. It feels great to be able to design the website using proper Vue components, compared to the quasi-template language of something like Hugo. And the page load and navigation performance has been quite nice in my opinion.

The only downside that I will mention is that builds are quite slow. I'm not sure how this will scale out as I get more pages. My Hugo website can build 100's of pages in under a second, while VitePress is taking 30-40 seconds to build around 100 pages.

For me it's like my head has a large cluster of art that I've seen, all organized. It is the "memory palace" basically, if you've heard of that approach to memorizing things. Except my memory palace is very, very large, and it has an area for art that I've seen.

If I appreciate a piece of art then it goes into my memory palace collection in a room that groups it together with other similar pieces of art that I had a similar appreciation for. As I appreciate the piece of art my memory palace branches out from there in many directions. I'm recalling other similar pieces of art (how does this art rank compared to other pieces I've appreciated). I'm recalling other pieces of art by the same artist (how did this art compare to their other works). I'm recalling similar artists of the time period (how did this work compare to art by contemporary artists). I'm imagining the techniques that were used to make the art, whether dramatic and physical (splashes of paint in a modern art piece) or precise and delicate (such as the fine brush strokes). This is the process of appreciation I feel.

Basically, the "I like this picture" is a very visual experience for me, not a vocal experience at all. If I did not like the picture then I just look at it, have an internal feeling of "meh" or "boredom" and do not put any effort into memorization or related recall and connectivity of the memory.

For me my internal monologue is replaced by almost entirely images. For example, if I'm learning about something in a programming language I'm mentally diagramming components and imagining how data structures are laid out as a network of items. To the extent that I do verbal thinking while programming it's usually just a internal vocalization for an interjection I would otherwise be tempted to say out loud like "why the fuck is this not working, this SDK is trash". And I only say it to myself internally so as not to interrupt other people around me.

For other things that are less abstract, with a concrete real world appearance, I'm constantly visualizing the outcome like I'm looking at a picture or watching a movie. This also applies to the "things that come to my mind without talking", which you mention. They drift into appearance in my mind as images as well.

It's the not the cost of obtaining the water that makes water less viable, it's the cost of storing superheated water. Water has this pesky ability to expand 1600 times larger when it goes over 212 degrees Fahrenheit. That means that if not handled carefully you get deadly steam explosions.

Hot sand / crushed rock doesn't have the same problem. If you read in the linked article it says

> with the sand heated to somewhere around 500-600 degrees Celsius (932-1112 °F).

That would be extraordinarily hard to do with water as you'd need significant containment and safety measures.

I'd bet it's this, plus something even stupider like hashing a connection timestamp millisecond as the "uniqueness" of the hash. I've seen a lot of terrible code implementations that assume that there will never be two clients connecting in the exact same millisecond

Shopping is entertainment for many people. Therefore product discovery can also be entertainment.

When Bedford between N 3rd and N 4th got redeveloped... well that was the end of the vibe.

I have an external garage building that I'd love to turn into a small ADU, but it's not allowed in my area.

Additionally, if you look at historical cheap rentals of the past, you see things like tenement buildings, which were arguably horrible, but provided a cheap alternate to homelessness.

You can also see examples of what low rent places look like today in Asian countries, such as Hong Kong tiny apartments. Many of those types of residences are not allowed by local government policies.

An actual $500 a month rental is going to be really tiny, might have a shared bathroom and kitchen with other units. And these are all things that are often disallowed by government policies.

I've been following Naomi Wu on Twitter for years, but I didn't notice that July tweet until reading this post. The reason: I'm barely on Twitter/X anymore because its so bad now, and even when I'm on Twitter/X I don't see stuff I wanted to see because their algorithm is trash now and just likes to highlight conspiracy theories and the latest cringe content from Elon Musk

Anyway, I'm super sad to see Naomi Wu so deplatformed, and even sadder that I didn't even notice thanks to how effectively Elon fucked up Twitter.

She's not the only one either. So many great voices are lost, and I wonder how many others I missed and haven't noticed yet.

Grandparent comment probably meant to say "no profit" instead of "no revenue". But yeah killing half of your revenue when you aren't even making profit yet is probably a bad idea.

As I work at a cloud provider I have a different perspective on the mix-ins. The problem is that at scale there is no one customer type. Some customers absolutely require a specific way of configuring things, while other customers require exactly the opposite. Cloud providers are always trying to add more features to please all people, its a fundamental side effect of growth. And an often unstated aspect is that the most custom features are often extremely high value ones that were added for customers that worth large amounts of money to the business, even though the other 90% of small users and startups may not need that.

I think mix-ins can be a way to make the growth in features accessible while not overwhelming. Out of the box mix-ins can provide 80% of the value for 80% of the people and the long tail of 20% customers with the super custom needs that are worth high dollar amounts can still build their own custom mix-ins that target their specific use cases.