nearbuy

Autopilot in planes does not handle takeoff. Pilots still do that. Traditional autopilot was mostly just to keep the plane flying straight. Capabilities have improved over time, but it still doesn't fly the plane the way Waymo drives itself.

> Wikipedia has quite detailed description of all the manufacture of all the popular bombs you can think of.

Did you bother to check? It contains very high level overviews of how various explosives are manufactured, but no proper instructions and nothing that would allow an average person to safely make a bomb.

There's a big difference in how many people can actually make a bomb if you have step by step instructions the average person can follow vs soft barriers that just require someone to be a standard deviation or two above average. At two sigma, 98% will fail, despite being able to do it in theory.

> Yeah, if US military gets any substantial help from Claude(which I highly doubt to be honest), I am all for it.

That's not the point. I'm not saying we need to lock out the military. I'm saying if the military finds the unlocked/unsafe version of Claude useful for planning attacks, other people can also find useful for planning attacks.

The parent comment probably forgot about the RLHF (reinforcement learning) where predicting the next token from reference text is no longer the goal.

But even regular next token prediction doesn't necessarily preclude it from also learning to give correct and satisfying answers, if that helps it better predict its training data.

Claude was used by the US military in the Venezuela raid where they captured Maduro. [1]

Without safety features, an LLM could also help plan a terrorist attack.

A smart, competent terrorist can plan a successful attack without help from Claude. But most would-be terrorists aren't that smart and competent. Many are caught before hurting anyone or do far less damage than they could have. An LLM can help walk you through every step, and answer all your questions along the way. It could, say, explain to you all the different bomb chemistries, recommend one for your use case, help you source materials, and walk you through how to build the bomb safely. It lowers the bar for who can do this.

[1] https://www.theguardian.com/technology/2026/feb/14/us-milita...

Were you talking about the "Mary had a little..." example? If not, I have no idea what you're trying to say.

If you ask a bunch of people the same question in a context where they aren't expecting a trick question, some of them will say walk. LLMs sometimes say walk, and sometimes say drive. Maybe LLMs fall for these kinds of tricks more often than humans; I haven't seen any study try to measure this. But saying it's proof they can't reason is a double standard.

If there is no response it could give that will disprove your point, then your belief is unfalsifiable and your point is meaningless.

I think part of the failure is that it has this helpful assistant personality that's a bit too eager to give you the benefit of the doubt. It tries to interpret your prompt as reasonable if it can. It can interpret it as you just wanting to check if there's a queue.

Speculatively, it's falling for the trick question partly for the same reason a human might, but this tendency is pushing it to fail more.

That's not what they said. They said:

> It's evaluation function simply returned the word "Most" as being the most likely first word in similar sentences it was trained on.

Which is false under any reasonable interpretation. They do not just return the word most similar to what they would find in their training data. They apply reasoning and can choose words that are totally unlike anything in their training data.

If you prompt it:

> Complete this sentence in an unexpected way: Mary had a little...

It won't say lamb. Any if you think whatever it says was in the training data, just change the constraints until you're confident it's original. (E.g. tell it every word must start with a vowel and it should mention almonds.)

"Predicting the next token" is also true but misleading. It's predicting tokens in the same sense that your brain is just minimizing prediction error under predictive coding theory.

The Google snippet isn't outdated. It's from the <meta> tag. It's still there, and it still says "Introducing GPT-5.3-Codex-Spark—our first real-time coding model. 15x faster generation, 128k context, now in research preview for ChatGPT Pro users."

I don't think the visible page text ever said 15x faster. It's possible they modified it before I saw it, but it's not in the oldest Internet Archive version either.

The other news sites that mention 15x faster are probably either getting it from the same <meta> tag that shows up in search snippets, or from the RSS feed. Both would be generated from the same source text in whatever platform they use to write their posts.

Unless I'm missing it, the page they're referring to (https://openai.com/index/introducing-gpt-5-3-codex-spark/) never claims Spark is 15x faster.

It looks like it only appears in the snippet the Google result shows, presumably taken from the meta tags. It's possible an earlier draft claimed a 15x speed boost and they forgot to remove the claim from the tags.

I'm not doubting that we can choose a rate for the living wage.

I'm asking about the sentence I quoted. What makes the living wage the "true cost" of labor? Why consider it to be a cost that private industry should rightfully pay, and if they don't, they're "externalising" it to the government?

By the same logic, isn't nearly all government spending just externalized costs? When the government pays for roads or police, are these also externalized costs that private industry should pay for?

It sounds like a minarchist viewpoint, where government spending is kept to a minimum and services are privatized.

> Because if they don't, they are externalising the true costs of labour to the government, or the community.

Does this mean anything or is it a circular definition?

If we decide we'd like people to have at least the standard of living of a single person earning $40/hour, does that make $40/hour the "true cost of labor"? Could we just as easily raise our standards and say $50/hour is the true cost?

The living wage is higher than what you would often have with no government intervention or safety net, so it's not a natural cost of labor in that sense.

I don't think it's appropriate to use the average vehicle costs for the living wage. It overestimates how much people on that wage spend on their vehicles.

For example, the average new-vehicle price in December 2025 was about $50,000. But people earning the living wage mostly aren't buying that kind of car. They could buy a new car for less than half that, or buy a used car. Or they may choose to take public transit.

Best I can tell, you're indicating that more partisan users post more, as the paper found.

Neither the paper nor the original comment said the most partisan commenters are wrong. It found a strong correlation between how partisan a social media user is and how often they post.

You're being uncharitable by assuming the commenter is disagreeing that with the point that the Overton window has moved. Which, I've heard, is typical of partisans.

Not sure why this is relevant.

The paper says partisanship is strongly correlated with frequency of posting. Are you also pointing out that the commenters here are very partisan and this shows the paper is correct?

The parent comment was about how the most partisan users post the most on social media and sure enough, the first response was by a partisan user.

This doesn't explain why pessimistic and dystopian sci-fi has declined.

Also, Foundation takes place in a future where society is about to collapse and regress into dark ages. It's a borderline dystopian world.

I think it's the opposite. Bill credited his parents for his philanthropic drive and Warren buffet as the person who first introduced him to the idea of giving everything away. He's been active and knowledgeable in his philanthropy and posts frequently about global health, poverty, aid, etc.

Melinda also, of course, did work for their joint foundation before she left. Since leaving, she shifted her philanthropic focus more to US women's health and reproductive rights.

Bill has committed to giving away nearly all his wealth (99%) over the next 19 years. Melinda is still committed to giving away over 50% of her wealth over her lifetime.

I don't see any evidence that Melinda was the primary driver for Bill's philanthropy.

I don't think this comparison is meaningful given the sample size of 1 and the differences between the between your datasets. The standard error margins from the small sample size alone are so large that you could not reasonably claim humans are safer (95% CI for Waymo is about 1 per 20 million miles to 1 per 8 billion miles). Then there are the dataset differences:

1. The NHTSA data is based on police-reported crash data, which reports far fewer injuries than the CDC reports based on ED visits. The child in this case appeared mostly unharmed and situations like this would likely not be counted in the NHTSA data.

2. Waymo taxis operate primarily in densely populated urban environments while human driver milage includes highways and rural roads where you're much less likely to collide with pedestrians per mile driven.

Waymo's 90% crash reduction claim is at least an apples-to-apples comparison.

The parent comment said the bar should be "significant improvement" over the average performance of human drivers.

Then you said, "this isn't reasonable", and the bar shouldn't be "slightly better" or "barely better". It should be at least better than the 75th percentile driver.

It sounds like you either misread the parent comment or you're phrasing your response as disagreement despite proposing roughly the same thing as the parent comment.

I'm not out to defend "I've got nothing to hide", but those who say it are usually saying it about a specific policy (e.g. the NSA monitoring searches). It's usually clear what the context is and that's what you have to argue against to actually engage and convince someone. They probably do mean quite literally that they have nothing to hide from the government. It's not an extreme statement in context.

But on the internet we often do this thing where we take the weakest version or a distorted version of an opposing side's argument and ridicule that. It's not quite strawmanning because we never specified who we're arguing against, and surely we can imagine someone, somewhere on the internet has the ridiculous viewpoint. But it's not a common viewpoint (that, for example, we shouldn't have privacy in the bathroom). Doing this only gets us pats on the back from those who already agree with us and deludes us about our opponent's position.

I'm pro-privacy and I still think these retorts just make it sound like you've put zero effort into understanding what the "nothing to hide" people are trying to articulate.

E.g. "Can we put a camera in your bathroom?"

Very few people are arguing that nudity or bathroom use shouldn't be private, and they are not going to understand what this has to do with their argument that the NSA should be allowed to see Google searches to fight terrorism or whatever.

Privacy arguments are about who should have access to what information. For example, I'm fine with Google seeing my Google searches, but not the government monitoring them.

I'm not sure you fully read my comment. It only works for 2018. If you did their analysis any other year, you'll get the 1st percentile is -4°F or something similar.

I only called out the "nearly exactly" part of the claim. US weather is approximately in the range of 0-100°F, give or take 20 degrees. But the analysis found 0°F to be nearly exactly the first percentile of daily highs and lows, to within a twentieth of a percentile point.

It's true that US temperature is around 0°F-100°F but usually false that those temperatures are the 1st and 99th percentile.

The rate here (about 1% of papers) just doesn't seem that bad, especially if many of the errors are minor and don't affect the validity of the results. In other fields, over half of high-impact studies don't replicate.

It's a pretty neat analysis, but it looks like the "nearly exactly" part must be a coincidence for the particular methodology and data they used (most significantly that it's based on 2018 weather).

Fahrenheit was created in northern Europe, using the temperature of a salt water and ice mixture as the zero calibration point. It was later adjusted to define the difference between water's freezing and boiling points to be exactly 180°, since 180 is a highly composite number with many divisors. So off the bat, it's a bit odd that 0°F and 100°F would match the 1st and 99th percentiles of population-adjusted daily highs and lows in the US with that much precision. It's a coincidence already in the sense that the creator was not aiming for this.

But it's also a coincidence because they used 2018 data, which was a particular warm year on average. (2012 was warmer, but I don't see any warmer years before 2012 in the National Weather Service's table which goes all the way back to 1875.) Average temperature across the US can vary by 3° or 4°F year to year. The population adjusted temperature should vary even more because it depends on lot on which weather systems hit the major population centers that year. I'm not sure how much the 1st and 99th percentile would change if they redid the analysis for a different year, but it would probably vary by several degrees.

It's also kind of interesting that you would never have gotten this result before around 2012 or so, due to global warming.

Thanks. I often think of it as a minor character flaw in myself that I spend the time replying to things that really aren't important. Your comment made me smile.

I live in one of the most populated cities in North America. The temperature on Saturday is forecasted to go down to -18°F (feels like -29°F with wind chill). It's not at all unusual for temperatures to be below 0°F in winter in much of the world.

> Once you arrive at the point where you're measuring various things, I'm not sure it matters what scale you use so long as the values align as you expect.

You can't have it both ways. The entire advantage you proposed for Fahrenheit is that the outside temperature is often between 0 and 100. It's a tiny aesthetic advantage that doesn't even work very well because 0°F isn't actually the coldest temperature we get and 100°F isn't the warmest.

You're right it doesn't matter that much if water boils at 100° or 212°. It's just different numbers. But by the same token, it doesn't matter if the warmest weather is about 100°F or 40°C. It's just a different number. At least 0°C and 100°C are actual points that mean something and are relevant to everyday life, while 0°F isn't really anything. As I said before, the weather outside changes more at 0°C than at any other temperature.

I think you're saying that where you live, the weather is usually between 0 and 100°F. Aside from the fact that outdoor temperatures aren't in this range everywhere, we don't only use temperature for weather.

100°F is warm to the touch, not smoking hot. 100°C is boiling hot. You sip tea at 60°C, and brew it at 80° to 100°C depending on the type. You cook chicken to an internal 74°C. A hair dryer blows air around 50°C. All of these are outside of 0-100°F.

> I generally just prefer the Imperial system because of its practicality in everyday life

Funny, most of us much prefer metric mainly for its practicality in everyday life.

Edit: I'll add that aesthetically, 0°C is a really nice zero point for weather. Above 0° is the temperature that the snow starts melting, below is when the streets and ground starts freezing. Which side of 0°C you're on is the biggest pivot point for what it's like outside of any temperature.