One obvious reason to have an LLM output a high level language even if you are never going to read the code is because it will require fewer output tokens and therefore be cheaper.

Typescript is not meaningfully a higher level language than Java, C#, or even Rust. Typescript exists for exactly two reasons:

1. JS is a privileged language and allows you to tap into web and node

2. Type safety is good

But TS is not a good language overall. It's just much slower than alternatives and much less safe.

It's not immediately clear to me why having an LLM output code in JS or Python should inherently consume more tokens than C or assembly. I'd expect token usage to correlate with the complexity of the algorithm being implemented, not with what specific language syntax it's being implemented in.

But either way, token consumption is only a relevant concern if you still expect to be using SaaS LLMs by the time you're ready to have them target lower-level code for you.

If LLM-generated code continues the human pattern of having roughly the same rate of bugs/LOC regardless of language, then I also expect LLM-generated code in high-level languages to be less buggy.

But why would we expect LLMs to continue that pattern over time?

Yes, it's management with little concern for human factors because you can abuse or and even fire your reports and summon new ones of equal competence at will.

Isn't that just regular management?

Regular management, no matter how bad the tendencies go, is still, at the very least, constrained by a cost to fire people, which can be years of salary and some paperwork on top.

That's odd, I seem to see the incentive to fire people with stock value improving as a result.

And since management is often paid in stock, they seem to care more about stock price than this "years of salary and paperwork" cost you mention.

You're conflating two things (probably intentionally, but I'll go with it). If you are firing someone without replacing (a layoff), then presumably, that person cost more than the value they provided, and so laying them off should improve the value of the business (after incurring the costs the parent post mentioned).

If you are firing someone to replace them with someone else, you are incurring a lot of cost (hiring is time consuming, difficult, risky, and requires a ramp-up time before the new hire is productive), and hoping that the long term benefits outweigh that cost.

Based on that logic, layoffs should sometimes cause the stock of a company to go down, if the shareholders think those employees contributed more value than the company saves by laying them off.

In reality layoffs almost always cause the stock to go up. The market seems to think layoffs are an unalloyed good.

It’s also odd as it screams a different problem too.

If the employees were advantageous, I agree laying them off should be a long term negative as you seem to suggest.

Kinda like cheering the warmth of a burning bed on a cold night.

If they were inefficient/ineffective then management is crappy for not dealing with the problem sooner.

Sure, companies are resilient. But the first year or two of lost competence from layoffs can be quite rough…

Based on my personal experience and general observation, it seems like companies are both slow to hire, and slow to fire. So, paradoxically, both actions seem like they should signal "this company is going to be better in the future", because companies only either hire or fire when it is really obvious they should.

> Based on that logic, layoffs should sometimes cause the stock of a company to go down, if the shareholders think those employees contributed more value than the company saves by laying them off.

You are treating short-term stock market movements as a signal of how well the company is performing.

you still seem to be conflating layoffs with firing someone where you have to train their replacement.

there are numerous well known aphorisms about the irrationality of markets, so not sure why that should matter.

Wrong etage. Stock value is not very responsive to the fact that some line manager fired two poor shmooks, but it is responsive to the news of company laying off two digit percent of the same. That will be of course corrected somewhat once fresh quarterly results will show how effectively those enlightened souls managed to oppress the work takers.

Now if your line manager fires somebody for no good reason and then has to still pay the salary of this person for a year and a half and also pay the salary of their replacement -- that, when accumulated, starts affecting bottom line at some point.

No regular manager has an infinite hiring pool, so no.

Disgruntled AI won’t kick your teeth in, yet.

your context has grown beyond manageable size and you are starting to hallucinate so we are going to have to let you go. We wish you the best of success in your new role.

ctrl-C ctrl-C

Mr Meeseeks

Governments are famously tolerant of other actors seeking to rival or usurp their power.

Unfortunately with the current administration there is a non zero chance that the usurpers merge with the existing government… that wouldn’t be too surprising for Thiel to push for, he’s already the one who made Vance VP and has been pushing the Republican Party to its current anti-democratic form

The error here is assuming that government's intolerance to others in post-usurping times is in the interest of the public - then it's no more than suppression of rivals, including those who might actually be willing to help the public.

Worth noting that as written the "trick" results in memory usage proportional to the size of the input rather than the output. If the filter rejects most of the input the difference could be quite noticeable.

I disagree in the sense that you can rewrite the code to use the trick and also not allocate in advance. Nothing about the trick requires you to allocate up front: before writing to out[n] you can extend the vector if it’s out of bounds. Or, after incrementing n, do out.push(0).

You should try writing it out. Doing it without introducing another unpredictable branch is harder than it looks.

I discussed this with a coworker earlier this week and the best they were able to come up with was

    for &x in input {
        out.push(x);
        n += (x > threshold) as usize;
        out.truncate(n);
    }

which works but is ugly af imo.

Yep realized this after that my second solution (push a 0 if n is incremented) has the same branch prediction problem.

I think yours works. Alternatively in the loop:

    if out.len() < n {
        out.push(0);
    }
    out[n] = x;
    n += (x > threshold) as usize;

In this case the if will be predicted well because it only triggers log(N) times, given how the std lib extends vectors.

Generally the default fund in a decent 401(k) (employer managed but largely employee funded retirement plan) will be a "target date retirement fund" that contains a mix of equities and bonds according to the expected risk tolerance of someone at that age, and generally the equity component is also split between US and foreign equities.

e.g. Fidelity's fund for people planning to retire around 2050 https://fundresearch.fidelity.com/mutual-funds/summary/31579...

3% less annual returns than the S&P500 over 10 years and the entire fund lifetime. That's a considerable difference.

That's what happens when you are in a 10+ year bull market. That doesn't mean 100% US equity allocation is a good idea starting now.

Yes, diversification results in lower returns than investing in just the best performing asset class.

target date funds always confused me, b/c they don't account for the rest of my portfolio.

My IRA/401(k) actually has close to 100% cash/bonds to minimize annual taxable impact and the rest of my portfolio is equities.

There is no annual taxable impact in a 401k nor IRA

exactly, which is why your 401k/IRA include your bond allocations and your taxable account house your equities.

Assuming a portfolio with $100k in 401k and $100k in taxable and targeting 80/20 stock/bonds, your 401k should be 60/40k and taxable should be 100/0.

Target date funds would have 80/20 in both 401k and taxable accounts.

Target date funds assume your 401k is your only retirement savings vehicle.

I understand why you'd say bonds in tax-advantaged accounts and equities outside, but I don't know why you would put cash in the former. Unless by cash you mean money market funds or similar. But even still, qualified equity dividends are yielding more these days than bond interest to the extent that the total tax is usually higher, so you might still prefer to reverse what you said.

My equities are highly concentrated in tech (which tends to either not pay dividends or prefers stock buy backs). The rest of my equities is VTI (~1% dividend) and VXUS (~2.3% dividends).

When I say cash, I meant value stored in money market and $SGOV. I'm actually experimenting with using my 401k/ira to invest in private equity (VC) funds, so I am holding cash (money market and $SGOV) in those accounts to ensure I have the capital for capital calls.

Many people don't have a "rest of their portfolio".

Unless you've checked it for accuracy why would I read your AI-generated documentation when I can ask my AI (which in a month or two will be a newer, better model than what you used to generate that documentation) to read the code and summarize it for me?

If you did check it for accuracy then I don't understand how you compressed months of work into minutes.

We've seen b before. React was controversially released with a similar clause and ultimately Facebook dropped it and used a standard open source license. Their lawyers really love this idea for some reason.

We used to call b the Disney/LEGO clause. Disney was famous for litigating any GenAI product for copyright infringement if it was able to recognize or god-forbid generate any of their IPs. So the lawyer cats would slip that clause into all licenses.

Personally i think it's fair. You get to use this model as long as you don't sue Meta because of the model's weights or outputs.

Was it released with it? Iirc they tried moving to such a license from bsd, which contains no patent grants whatsoever, and hence they thought it was a strict upgrade. The community did not agree so they moved back.

My memory is a bit fuzzy but I think you're correct.

Do their lawyers love the idea? Hard to say. My guess is, Zuck really likes the idea, so he keeps directing them and their chief counsel to try stuff like this.

It's almost as if having a CEO of a company placed beyond the control of the corporate board is a bad thing.

This is a pretty common clause when a company open sources something that was previously locked away. The AOM AV1 codec has a similar rule for its patent pool to discourage trolling.

> My guess is, Zuck really likes the idea,

You made an assumption, that's a good thing to expand thinking.

> It's almost as if having a CEO of a company placed beyond the control of the corporate board is a bad thing.

You can't draw such conclusion then.

Painfully slow tok/s though.

I didn't say it was fast!

But its also a 800B sized model running on a ram constrained system with no GPU.

Techniques, GPUs, more ram, and faster disks can always speed it up. But the point being is they run on low end machines now. Its now an optimization problem, not a possibility assessment.

Isn't it more s/tok?

If you show up at a US port of entry without any sort of anything the default decision is to deny entry (exceptions for Canadian citizens and citizens of a few small islands). If you show up at a US port of entry with an ESTA/visa you will most likely be allowed entry. Refusals are relatively rare.

Commercial carriers such as airlines are also not allowed to transport people who need ESTAs/visas and do not have them to the US, so in practice it's relatively difficult to even get to the port of entry without a visa.

That already happens. Witness all the random "bans" and "pauses" Trump has instituted against various nationalities and groups of people at the border.