Hacker Newsnew | past | comments | ask | show | jobs | submit | dimava's commentslogin

1. Owning your own company stock is not having money, it's merely ability to control the company. Elon tries to be BDFL of his companies, so selling the stock is not really an option.

2. Stock valuation and money are not the same thing. As you could've noticed from the recent Bitcoin price drop, the moment you start selling it it starts to worth less. Tesla is bubble-priced, so it's sell price is at least 5 times less then valuation

3. Even if you get the money, distribution is much harder problem then you think after agreeing with this sensense. IIRC it was Musk who said "So you say you can fix the world hunger for $nnn? Bring me the plan how you are going to do it. Noone brought me a plan so far".


Sure. If you completely ignore the ability to borrow against your assets at extremely low rates. And those assets include stock.

Since I know russian well, here's a proper translation for y'all

    FUNC FACT (N);
       NAMES: P;           (* variable names *)
       1 -> P;
       FOR I FROM 1 TO N ::
          P * I -> P
       DONE                (* endif *)
    RET: P                 (* return value *)
    END;                   (* end of function *)
    
    FOR N FROM 0 TO 6 ::
       ? "FACT(", N, ") = ", FACT(Н)   (* print *)
    DONE;


If it's purely bot traffic, then Anubis could help

You could have seen it on some websites already

https://anubis.techaro.lol/


anubis only works against lazy scrapers, and at a cost to your users. I'd prefer people not use it.

Bot traffic comes from machines that usually have a lot of idle cpu (since they're largely blocked on network IO as they scrape a bunch of sites in parallel), so they can trivially solve the anubis "proof of work" challenge, save the cookie, and then not solve it again for that site.

The only reason scrapers don't solve it is if the developers were too lazy to implement it... and modern scrapers also do, codeberg stopped using anubis because modern scrapers were updated to solve it.

The "proof of work" has to be easy or else people on old cell phones couldn't access your site (since an old android phone would start to overheat and throttle trying to solve a challenge that would take a modern server even several seconds), and it also consumes your cell-phone user's batteries, which is a really precious resource for them compared to the idle cpu on a server.


Just to add to the two negative replies, I find Anubis to be the only system that doesn't ever get in the way. My browsers have Javascript enabled and, so far, it never took more than a fraction of a second to complete the checks

Every other system I've run into has constant false positives, e.g. Google captchas will sometimes say I've failed and make me do the hardest level (if it wasn't giving me that already), Cloudflare regularly thinks I'm a bot, Codeberg blocked me before, Github signup captchas used to take ~15 minutes to complete and then still said "well you failed, try again", Github's general rate limiting has false positives (some days I browse a lot, other days little, and on the little days it'll sometimes go "slow down" with no recourse whatsoever, you're just blocked for an indeterminate amount of time), OpenStreetMap blocks my browser at work because I'm using Firefox ESR instead of latest stable and it finds that user agent string to be implausible, whatever the german railway operator uses since a few days is triggering on me constantly, etc.,

etc.,

etc. Constant blocks everywhere.

With Anubis, my understanding is that you do the proof of work (with whatever implementation you like, it doesn't have to be the Javascript one that they provide) and you can move on without ever doing any task yourself. The power consumption is a shame, but so long as attackers aren't even doing this much, the couple Joules it takes doesn't seem to be an issue

Of course, the attackers will evolve, but for now...


Please no. I'm a non-bot who gets stopped and turned away all the time by that menace. Anubis doesn't work without JS.

One of the things I give duckduckgo a lot of credit for is that while they're quick to interrupt me for a bot check (sometimes multiple times in a span of minutes) they'll let me identify ducks even on the most locked down browsers I use.


The amount of compute you get is not documented anywhere.

Cursor docs do say "You get $70 of api priced tokens on $60 tier"

Anthropic limits were 8.33x with free cached tokens, but those numbers themselves were a leak rather then an official statement


Cursor may be losing money only on $200 sub people who do over $200 of usage (it grants $400)

Everyone else pays them at API prices


And then you have a

    const CatDog = { bark(){}, meow(){} }
and

    const TreeCat = { bark: "oak", meow(){} }
and your code stops working

To make types discriminatable you need either

    type A = { bark: fn, meow?: never } | { bark?: never, meow: fn }
    type B = { species: "dog", bark: fn } | { species: "cat", meow: fn }
    or use instanceof with a class


Just to reiterate

ANTHROPIC IS GIVING EVERY DECENTLY LARGE MAINTAINER $1000 WORTH OF INFERENCE (~x8 that in API prices)

They likely made a marketing budget for this of $1M or so

Other OSS stuff like Copilot or JetBrains costs to providers much less, $100/yr most (licenses are not expenses, only inference is)

Anthropic may get $500(average total for all 6mo) per user of just inference costs

6 months is because this is experimental and they have no idea what to expect

(their devrel department is meh as you could've noticed already), when they see it working they'll make it autorenew or something

ESR (Eric S.Raymond) asked OpenAI to match and got one, so the same offer from OpenAI will likely follow soon[tm]


Also check the https://botblock.ai/ , AI extension to detect AI replies on twitter


A better option is to stop using it entirely.


That's a curious one, Twitter is worthless anyway. Before AI bots proliferated, the change to rank paid accounts high in replies turned it into a de facto entry level $8/month advertising tier.


Does it even work?

I ask becaue it considers @lilycoy__ (an obvious AI generated account, as quoted by Robin Hanson <https://x.com/robinhanson/status/2025332066552819782>) to be "100% human"

https://i.imgur.com/NQHVcdM.png


Open source models costs are determined only by electricity usage, as anyone can rent a GPU qnd host them Closed source models cost x10 more just because they can A simple example is Claude Opus, which costs ~1/10 if not less in Claude Code that doesn't have that price multiplier


But Kimi seems so big that renting the necessary number of GPUs is a non trivial exercise.


Exactly! Electricity, hosting, and amortized cost of the GPUs would be the baseline costs.


Tokens will cost same on Mac and on API because electricity is not free

And you can only generate like $20 of tokens a month

Cloud tokens made on TPU will always be cheaper and waaay faster then anything you can make at home


This generally isn't true. Cloud vendors have to make back the cost of electricity and the cost of the GPUs. If you already bought the Mac for other purposes, also using it for LLM generation means your marginal cost is just the electricity.

Also, vendors need to make a profit! So tack a little extra on as well.

However, you're right that it will be much slower. Even just an 8xH100 can do 100+ tps for GLM-4.7 at FP8; no Mac can get anywhere close to that decode speed. And for long prompts (which are compute constrained) the difference will be even more stark.


A question on the 100+ tps - is this for short prompts? For large contexts that generate a chunk of tokens at context sizes at 120k+, I was seeing 30-50 - and that's with 95% KV cache hit rate. Am wondering if I'm simply doing something wrong here...


Depends on how well the speculator predicts your prompts, assuming you're using speculative decoding — weird prompts are slower, but e.g. TypeScript code diffs should be very fast. For SGLang, you also want to use a larger chunked prefill size and larger max batch sizes for CUDA graphs than the defaults IME.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: