HN user

polymatter

1,380 karma

software developer and accountant

interested in functional programming, learning chinese

email: glisher (underscore) rock (at) hotmail (dot) com

Posts1
Comments567
View on HN
OpenAI: Model Spec 2 years ago

Question is asked in traditional characters which are common in Taiwan (and Chinese diaspora). China uses simplified characters (mostly). Might make a difference.

How do you use AI to learn internals of massive repos? Surely the context is too big and/or it requires opening org code up and I'd never get approval for that. What LLM do you use?

Companies House (part of UK civil service) | Senior Developer | Full time | Remote in UK or Hybrid in Cardiff | Can sponsor Visa to UK

We can offer an incredible defined benefit pension and an extremely family friendly flex system (You can work your time anytime between 6am and 8pm with no core hours, just clock in and out). If that sounds interesting you should apply.

We are Remote first with the super majority of people working from home most of the time.

We do a lot of training including free Pluralsight and other courses throughout the year.

I like to be perfectly frank though that we can not offer London salaries.

See the role at https://www.civilservicejobs.service.gov.uk/csr/index.cgi?SI...

Exactly, the chain of custody is absolutely required for this to be legal because no oracle can exist. It must be able to attribute exactly who contributed the suspect code. It must be able to handle the edge case where some humans might publish code without permission.

Either that or we effectively get rid of software copyright as copilot can be used (or even claim to be used) to launder code of license restrictions. Eg No I didn't copy your code, I used copilot and it copied your code so I did nothing wrong.

Dragonbox algebra uses the normal algebra notation for the later stages, so it could be you just need to persist.

For example initially the unknown variable is the namesake cute dragon in a box, but it later becomes "x", the bubbles become brackets, the green whirlpool becomes 0, the cute animals become other variables etc.

Unlucky. It might not even be keyword matching, AI systems in use a far dumber than that. AI systems are mostly trained random noise and it can be impossible to know what set them off.

Even if you get this resolved, its not a bad idea to create a new YouTube account, reupload videos you want to keep. And keep good OpSec to avoid them being "linked".

In general, I'm beginning to feel that habitually updating software is a risk.

I have a paperwhite, but I haven't connected it for a while. Generally when I do connect it to a PC to upload more books, I'll habitually update the software. Its good practice right? Guess I'll now have to remember in perpetuity not to.

Recently I updated my Raspberry Pi setup to find that the latest version of Raspberry Pi OS does not support the official Raspberry Pi cameras (it does add a ton of functionality, including a speed boost for Raspberry Pi 4, but nothing that adds stuff in my setup). So I rolled back everything. And I'll have to remember not to update unless they (or someone else) has added camera support back in.

The point is that I used to laugh at the old guys who refused to update their software. Now I'm turning into them. If everything is working the way I want, why update? Especially for personal devices where the impact of "security vulnerabilities" is so low.

If you are a member of a professional organisation, there will normally be some professional ethics rules which, if broken, will lose you your licence to work as a professional. Software developers don't have that yet, but civil engineers certainly do. As do lawyers, accountants and other professionals.

In a fundamental sort of way you're quite correct. Science can never _prove_ things correct in the same way mathematicians do. In fact, science constantly revises over time. It is not "done". And as a reminder, the way something is proved is when recognised scientific minds perform peer review and conclude it is proved. And they often get it "wrong" as most published science is later proven incorrect. This is science working.

However, this doesn't open the floodgates to any old theory because a simpler theory that explains the data is preferrable to a more complex one. Your green goblin theory might fit the data, but then leaves open questions like what do green goblins eat, why were they not detected previously, what else might they affect etc etc.

fwiw, it probably depends on the bank and the customer. I used to work for a large insurance company in London. There I could (and did) authorise 6 digit transfers to new accounts with just a fax with the (perpetually busy) directors signature. One time the bank made a mistake, and one simple phonecall from me was all it took to resolve (with no verification of who I was - unless they did some callerid check?). The level of customer service was astonishing to me, but that account was probably worth it to the bank to eat any fraud until it got too serious.