vagab0nd

Think about it. He's stealing from the US military. The politicians are stealing from you. Who's laughing now?

I thought we've long passed the Turing test, until I tried to implement a chat bot.

It's not even close.

It's easy to "pass the Turing test" for 5 minutes. It's extremely hard if you try to hold a longer, continuous conversation. Anything longer than 10 minutes the user will immediately know it's not human. Some problems you'll encounter:

- The bot needs to handle all situations, especially the nonsensical ones. This is when the user types "EEEEEEEEEEEEE...", or curse words, repeatedly.

- Who would've thought that it's extremely hard to decide when to stop talking?

- No matter how well you build the "persona" for the bot, they'll eventually converge to the same one, which is that of the llm itself.

- You'll notice that the bot is ignoring something obvious (e.g. it's not remembering past convo), and then give it some instructions to help with that. And then that'll be THE ONLY THING it does.

This is great from a technical and artistic perspective. But for me personally, the visual style ruined a great game. I love detective/deduction games. I'm listing some of my all-time favorites in this genre. I'd love to finish Obra Dinn, but god it just makes my eyes hurt so much.

The Case of the Golden Idol

Chants of Sennaar

Her Story

IMMORTALITY

The Painscreek Killings

The Roottrees are Dead

Type Help

I recently learned a trick to improve an LLM's thinking (maybe it's well know?):

Requesting { "output": "x" } consistently fails, despite detailed instructions.

Changing to requesting { "output": "x", "reasoning": "y" } produces the desired outcome.

The "productivity gain" in the world of bits has been pretty much exponential for 100 years. But as they spill over into the world of atoms, they flatten into step changes, because you lose the compounding effect.

I do think we are on the verge of something tho. Once the compounding effect happens in the world of atoms (recursive robotics), it's over.

> you will simply be able to put the camera in front of a high resolution screen

Are you sure it's that simple? How high does the resolution need to be for the camera to not be able to tell? And I'm sure there are sublet clues. Remember, you can't modify the photo or change the camera.

I'm working on a search engine for clinical trials around the world.

I was frustrated by the process of searching for clinical trial info using the clunky and slow registry websites. So I aggregated all trials around the world and made the search faster. Additional features:

- You can watch a trial and get email updates

- Sometimes a trial is done and a paper published, but no updates to the official page. I try to find these papers and link them

- I try to link trials with the same drug together, showing the drug going through different phases

https://findtrial.co/

I have a theory: I think the recent advance in coding agent has shocked everyone. It's something of such unheard-of novelty that everyone thinks they've discovered something profound. Naturally, they all think they are the only ones in on it and feel the need to share. But in reality, it's already public knowledge, so they add no value. I've been in this trap many times in the last couple years.

Interesting article. What prevents me from dumping the memory of the shell process and reading the local variable?

My program requires a password on startup. To run it in a loop, I used a script that takes the password as input, stores it in a local variable, and echoes it to the program. At the time I thought the only weakness was the memory of the shell process. But it was the best I could come up with.

My outdated/incorrect mental model of how well databases work.

OP here. Roughly 50GB in db size. Fairly standard queries (full-text search + filters). Most queries are on the order of 10-100ms. Some more complex ones involving business logic exceeds 100ms.

This is well within my budget, but it sounds like there might be room for improvements?

I recently started digging into databases for the first time since college, and from a novice's perspective, postgres is absolutely magical. You can throw in 10M+ rows across twenty columns, spread over five tables, add some indices, and get sub-100ms queries for virtually anything you want. If something doesn't work, you just ask it for an analysis and immediately know what index to add or how to fix your query. It blows my mind. Modern databases are miracles.

You are right. But in the future we'll be refueling the satellites anyway. Might as well maintain the servers using robots all in one go.

I feel like there's this weird information gap that shouldn't be there in this day and age. FSD drives me around every day. 99% of the time, I don't touch the steering wheel for the entire trip, both city and highway. It's obvious to me that it is working well and has tremendous value. As far as I know, no other car on the market can do this. But apparently this info is not reaching a large number of people. I'm genuinely confused.

Damn, I just wrote a note "search is free" in my aggressively-automate-everything-using-llms personal project plan.md. I guess not anymore.

So you prefer

does tue 11am work for you?

no, does tue 2pm work for you?

no, does wed 10am work for you?

yes, wait which timezone are you in? central? then no, does...

Or do you mean you don't like the link in the initial email?

I believe it's a play on "Your margin is my opportunity".

There's a weird conflict going on here and I've experienced it myself. Essentially we hear 2 claims:

- You all should build your own software. AI is so good!

- You all should use the software I built with AI. It's so good!

Is insider betting illegal? I know it's illegal in the sports context, but not necessarily in other areas.

Even for insider trading, it's illegal not because it's unfair, but because the insider is considered to be stealing information from shareholders. For example, it's not illegal for a company to buy back stocks while holding insider information.

Not legal advice of course.

Exactly! This is how the future should be: solar panels + portable huge battery arrays doubled as vehicles + grid as backup.

Off topic: I think there should be an extension to DHCP that distributes AI session keys, so that your vacuum, thermostats and robot chef can all call the LLM as soon as they connect to wifi.

> Am I missing something obvious?

To list a few: the risk of Polymarket going under, the risk of Polymarket mishandling your money, the risk of Polygon going under, the risk of Ethereum going under, the risk of USDC depegging, the risk of interests going up, and, most obvious of all, the risk of a civil war.

I'm curious: what's the use case? Usually when I need a server I need it to stay on and have a public IP.

And when you layer on top networking, it's another level of sw/hw complexity.

Does meditation really work for everyone? I tried it pretty seriously for 2 months, and my mind just wouldn't stop. I'd think of nothing for 10 seconds and then get extremely bored, so essentially every 10 seconds I'd force myself back. It didn't really have any calming effect.

My prediction is that software will be so cheap that very soon, economy of scale gives way to maximum customization which means everyone writes their own software. There will be no software market in the future.

We often hear claims like "do this and you'll make so much money" or "this is so good it's unbelievable." Occasionally, they turn out to be true. I've experienced a couple myself and feel fairly positive about the current one. The initial resistance comes from not wanting them to be true because then I know I either have to take action or risk missing out. It's a weird psychological thing that I now try hard to avoid.

Every time I want to watch a movie, I go on Netflix and search. Then I want to scream, "What movie DOES Netflix have??"

If an outsider is ever allowed to have a simplified mental model of why nvidia was unbeatable, here's mine:

- they were way ahead, and they didn't make any big mistakes

- they weren't waiting for others to catch up. They were aggressively improving

- memory bandwidth is almost always the bottleneck. Hence systolic array is "overrated". Furthermore, interconnect is the new bottleneck now

- cuda offers the most flexibility in the world of ever changing model requirements

Apparently it has an integrated guide on the right side of the page, which is "Based on \"The Kernel in The Mind\" by Moon Hee Lee".