My main AGENTS.md has ten lines about the most important coding principles:

The instruction on "meaning" ended up carrying a lot of weight to the AI. It adopted the habit of putting a comment on every single field and function that starts with the word "// Meaning: " and they're honestly, genuinely good ones! Single-line sentences on fields that carry a lot of good weight.

Separately, I have a LEARNINGS.md file which I have the AI auto-update every time it gets course corrected by me. Over the first two weeks there were a lot of course corrections, but now there are only a few a day. The file ended up carrying my senior engineer wisdom, more or less, the kind of things I normally mentor to junior developers on the team over several years. Here's an extract: https://gist.github.com/ljw100...

My CS degree had a lot of theorem proofs in it, invariants, that kind of thing. I've always had the habit of aiming to prove my code correct under all possible circumstances. Usually not a formal proof, but using the same skeleton as a formal proof would.

It got me a job on the C# language design team (when I tried to prove an algorithm correct, couldn't, discovered a counter-proof that the runtime had a flaw).

As I mentor junior devs and review their code, I'm always telling them to reason about their invariants better and document them.

Now in the age of AI, I find that invariant-heavy and proof-heavy guidance to the AI ends up getting its work done quicker and higher quality. OpenAI mentioned the same thing in a blog post in February.

Sure, there are many paths to professional success and engineering excellence that don't involve this kind of CS heavy approach. But, there are many that do...

I think a tool that lets one person do the work of 20 will result in 17 people being laid off, as the company triples its rate of work. (I can see triple being feasible and achievable in the software industry, but maybe not more)

Oh, and a "train" is a bit like 100 cars back to back.

"this belief is utterly false. To quote the U.S. Supreme Court opinion in the recent Hobby Lobby case: 'Modern corporate law does not require for-profit corporations to pursue profit at the expense of everything else, and many do not.'"

https://www.nytimes.com/roomfo...

"We ... show that [the Shareholder Primacy Norm] is not a legal requirement, at least under the guise of shareholder value maximization. This is in contrast to the common assertion that managers are legally constrained from addressing corporate social responsibility issues if doing so would be inconsistent with the economic interests of shareholders."

https://papers.ssrn.com/sol3/p...

> Am I the only one that can't imagine any possible value an AI assistant would bring to a game?

I use AI assistants lots when playing games!

At the moment it's Minecraft. I want to figure how to build something, e.g. a golem farm. I look for tutorials online but (1) they're all videos which I hate watching, (2) they're all hyper-specific and concrete, "place this block here then that block there", but what I want to understand are the foundational principles so I can know how to adapt the golem farm to my own purposes -- what are the mechanics, how do they spawn, how does water flow, what is the SOLUTION SPACE of possibilities.

Gemini AI has been really good at this kind of thing.

The other time is when I get stuck, or want advice on how to make a character build to achieve a certain end. Once again the online advice is typically in the form of "walkthroughs", do step 1 then step 2 then step 3, in other words just one possible way to play the game, and it's too easy to accidentally read too far and spoil the rest of it. I don't want that. I like the feeling of openness and possibilities. I again ask Gemini, and it gives me advice on just the particular bit I'm stuck on, and is better at showing for me the available options.

Lua remains the commonest choice today for games to offer scripting/modding. It's pretty much the industry standard. (Outside C# for Unity).

Fox News is just about always truthful. You just have to watch out for the tricks they use (on 95%+ of their stories)...

(1) non-representative selection. Headline "illegal immigrant murders local mother", which is true in this case, but they don't report the other 99 murders that went by immigrants, and don't report a general trend of immigrants causing less crime overall per capita. (I made up this specific example to illustrate their trick)

(2) report quotes: headline "Biden's senility was covered up, says person". They are 100% factually reporting that the person did indeed say this.

In both cases the reader is left with an untrue impression despite the stories containing only truth. It's because it's not the whole truth.

The irony is, this article itself was AI-generated slop with ridiculous duplication. Maybe a low-effort AI-assisted piece by an author who couldn't be bothered.

I think it can. I've been working on getting an LLM (Claude Sonnet 3.7) to add missing type annotations to python code. When I naively ask it "please add types" then like you said it has about a 60% success rate and 40% hallucination rate as measured by "would an expert human have come up with the same type annotations and did they pass the typechecker".

But when I have a much more careful use of the LLM, micromanaging what sub-tasks it does, then it has a 70% success rate, and 30% rate of declining because it didn't have confidence to come up with an answer. Effectively there were no more hallucinations. (I got these numbers by spot-checking 200 cases).

So I think hallucination can be solved for some tasks, by the right kind of task-specific micromanagement and feedback loops.

OpenAI has $12bn annual revenue, about 3% that of Apple, about $3million per employees per year (compared to $2 million per employee per year at Apple).

I think OpenAI has a huge amount of growth potential even just from predictable growth over the next several years, even if steep changes towards AGI don't come.

Like I said, I figured 8k would be enough resolution for soccer. As for 16k, I imagine that something with bandwidth for 16k would translate that bandwidth into twice the frequency for 8k, which would be ideal for soccer.

Do you watch soccer? 4k resolution means a player's head is about 14 pixels high, not enough to make out much beyond a blob of color; their jersey is 60 pixels high, enough to make out the number but not much more. Doubling the vertical resolution (i.e. going to 8k) would likely be enough to let you make out similar detail to what you'd see in real life. (Frame rate is another issue: HDMI 2.0 allows 4k at 60hz which is too slow when panning in a soccer game; HDMI 2.1 allows 4k at 120hz which is probably enough). I think that 16k is probably the right bandwidth to get soccer looking good.

Do you do VR? 4k per eye isn't good enough for VR yet. It's possible that 16k will be, but we might still need more.

Do you watch the gorgeous film classics like Lawrence of Arabia? One of the (many) things that make it look great is that it was shot on 65mm, equivalent to about 12k resolution.

The military team scrub *alerts about objects *.

Three days later, the full unscrubbed image is shared with astronomers.

I guess they don't and can't care about stable orbits, but they want to retain surprise for dynamic situations.

It's easy to regulate AI art the state level.

"Any job offer for a job based in California must adhere to the following AI disclosure".

"Any mortgage offered in a Californian property must satisfy the following AI disclosure"

etc.

AI regulation need not be about regulating AI innovation; it's enough merely to make sure it's applied fairly. And almost all real-world applications are indeed local.