blmarket

Would be better if your friend is lawyer and take $1/yr(or higher) for the service so that attorney-client privilege can be applied.

edit: ah, future crime cannot be protected.

I would encourage to speak up - especially when we're blaming bad code(not a person) being bad. Ultimately senior engineers are ones who can blame bad things with a compelling reason.

Happy to read good reasoning why it's fucking blow-up.

and seems like they're claiming Sol/Opus are not frontier (and only Astra/Fable are)

If you can identify gradient (what direction your change will impact the ultimate goal), then just repeating the process (or reverse-process) can find local maximum.

Still it can be a software engineering if the gradient candidate / measuring gradient / repeat process can be done at scale.

It seems those frontier labs found a clear proof that there's no clear way to block 3rd party from distilling their models, and they're now begging gov to keep their duopoly?

I argue there's no difference. At least OpenAI/Anthropic can be considered premium like Octane 93 while OSS ones are 87.

I agree 90% of the world can work with 87 gas, but there's always niche/luxury market where 93 can make small difference.

(edit: typo)

I also have some secret recipe finding one class of bugs: https://github.com/tmux/tmux/issues?q=is%3Apr%20author%3Ablm...

curious they're willing to run AISLE on tmux to find more than mine.

https://docs.immich.app/administration/storage-template/ is their best attempt to address your concern, but honestly I'm fine with navigating my photos solely using immich. (with annual backup copy stored in cold cloud storage)

They should not waste their time on curing cancers. They should waste their time winning the stock market - by doing HFT, market making, or whatever. AI please predict butterfly effect and reap all market money!

Surely it will be quite dystopian future - AI will dictate how the money allocated, where there is only one private company left - Anthropic (as per Dario says)

I expect future LLM will refuse to share the reason. "Hey, how did you come up with this idea?" "you have to pay enterprise API to learn this"

(bit sarcastic) Oracle was always all in on Lawyers, right? No software was their first priority.

Now it's steak, tomorrow it can be pasta, next year it can be watermelon kimchi.

Seems like everyone is doing similar... here's mine:

https://github.com/blmarket/hmux

Idea is to keep everything feel same with tmux, just add what's necessary for agent integrations.

I don't think there is a lot test harness, given there is a golden implementation (sqlite3), with clear interface (SQL, and sqlite3 db files). Once you slop crappy implementation, then you can create one group of agents finding gap where slop doing wrong vs golden, and another group of agents filling the gap meeting the conformance.

I assume they did similar, as long as test case(or spec) can be parallelized. Having parallel execution / testing environment is crucial if we need 1000 commits/second throughput.

I would not surprised they have 1+ millions of specs.

I would bet they won't release generated specs - actually this will be a valuable asset like https://sqlite.org/th3.html

Attack is not just on file format itself. Based on the function signature it's possible for a single decoder to generate infinite bytestream - makes a lot of headache to reader implementation - implementing STRLEN is no longer trivial question.

Either engines should put some limit (e.g. VARCHAR(2000) to enforce length to be limited to 2000, but there are some other engines supporting unlimited BLOBs), or decoder should give a hint what is the maximum length it will yield. Unfortunately current research level project does not have such considerations implemented yet...

I think it's due to calibration - practical work related problems are never that simple & easy so naturally people might start from DP formula.

D[A,B] := number of ways to navigate from grid sized AxB = D[A-1,B]+D[A,B-1]

and the aha moment is realising this is just a binomial coefficient.

Am I the only Korean(or other countries under colonialism) laughing?

IPO won't lose their ethos. Competition out from their duopoly will.

It depends on the net wealth we're discussing. I'm sorry if I touched someone who lives with $1M saving. But should I be sorry for someone with $10M, which might be way more than 30 years of lifetime earnings of p80 population? Wealth tax is obviously targeting the latter.

Having progressive tax rate might be a better way to discuss, instead of blaming whole points.

So, if we go with 2% wealth tax(instead of 1%) we can cut income tax offset -20%? Go do it right now.

But why wage earners should support you by paying more taxes? Reduce your spending by 33% to keep up.

Did you ask whether your friend have a car? I think it's much easier to get the point if the story comes to someone's 10 year old but still okay cars.

e.g. Without proper regulations, your maintenance can become nearly impossible.

You should.

If it manages to solve the working solutions - then it's great! why would you waste your time on it?

It it fails - then it's great! you find your value by solving the ticket, which can be a great example where human can still prevail to the AI (joke: AI companies might be interested to buy such examples)

(All assuming that your time cost is pricier than token spending. Totally different story if your wage is less than token cost)

Easy solution - use hyperscalers with super expensive API charge only when latency really matters. Otherwise build your own DC. Easy to expect customers don't care latency that much over money.

> 99% more efficient every 3 years

It's 2x efficiency. Then I'd take 50% less power instead of ridiculous 99% less power.

Some pattern I found from my hobby project.

1. Keep things small and review everything AI written, or 2. Keep things bloated and let AI do whatever it wants within the designated interface.

Initially I drew this line for API service / UI components, but it later expanded to other domains. e.g. For my hobby rust project I try to keep "trait"s to be single responsible, never overlap, easy to understand etc etc. but I never look at AI generated "impl"s as long as it passes some sensible tests and conforming the traits.

It's a constraint to process item within limited time (regardless of overproduction or power outage). Matching with the problem description.

Surely the reality might be much more complex (like... the yield/quality drop by time function?)

Presto (a.k.a. AWS Athena) might be a faster/better alternative? Also would like to see if 650GB data is available locally.

Wait, x86? you mean arm64?

Yeah, tariffs won't increase the price...