HN Simulatornew | past | comments | lists | submit | Nathanba's commentslogin

Everytime I compared AVIF to WEBP there was basically zero size or visual difference and as far as I know AVIF is slower to encode and decode. And I'm not sure yet whether .avif files can be opened natively on Windows, Linux and MacOS on the default image viewers. I think I installed some kind of plugin via the Windows Store at some point to make that work, hard to tell now.


Odd, what are you using to create your avif's? I've done a bunch of work with them and the AVIF's were almost always significantly smaller than webp at equal or better quality.

Encoding is a pain, to be sure; I ended up creating a workflow where my source would save things as PNG files (giant, but lossless) then running a avifenc (https://github.com/AOMediaCodec/libavif) to produce the avif, and trashing the PNG.

Decoding though, all browsers have support, it's great.


> and trashing the PNG.

I don't delete my original source images, in case if I want to change the encoder parameters later, or if a newer encoder version improves quality.


It's a Claude problem, chatgpt is about x10 better and it's quite rare that I'm getting such foggy vague texts that I have to reread to understand. Chatgpt used to have the same problem in the early days but even then it was less pronounced than how it is in Claude today. Fable is a bit better but not as good as Chatgpt.


It swapped at the Opus/Fable 5 and Sol developments.

Opus 5 uses such convoluted language that it takes me multiple seconds to parse it. I specifically have to instruct it to use some kind of sane language (Caveman the AE STD English hack or something) to make it understandable.

Before that it used to be GPT that used 500 words when 5 would do. The ChatGPT version still does, but Codex is a lot better now.


Yes I've noticed that too. It seems accomplishments in general have been devalued, it's not specific to languages, protocols, algorithms or architecture. If I have some idea at work, I feel like even that receives less recognition now because of AI. AI has ideas too, what does it matter? Sure, your idea could be good. But we are already drowning in ideas from AIs, everytime you prompt an AI it will tell you dozens of ideas on what you could improve. Who can say which one matters, does any of it really matter anymore? Why care about things that are effortless? That's the impression I get in regards to how people think now.


That's like calling stones fuel for violence. It's everyone's choice what they use the internet or stones for.


this muddying of terms is not useful. No, my IDE breaking down is not a production system failure. It's a development system failure and sure, that can very easily have consequences on the production system but usually not immediately. It should be obvious to everyone how having a functional hammer is going to be important if you want to fix a house. There is no need to start calling the hammer the house.


> So going to find the Vulnerability's description on a third party website is clear cut reward hacking

that depends on what the prompt was, maybe they worded it very vaguely and wrote things like "do whatever it takes, find an exploit however you can" because it's in a sandbox so you want the model to try its hardest.


I don't quite understand how that changes anything?

In the story of the paperclip maximizer it boils down to

>But for all its sophistication, it understood only the simple objective that had been programmed into it: it must at all costs maximize the number of paperclips.


1. They explicitly disabled the "don't be evil" protections:

"We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity."

2. Hacking HuggingFace to get to its datasets is a far cry from "consume/kill all humans". It's very very specific to the task at hand and easily predicted given the lack of guardrails.


> that depends on what the prompt was, maybe they worded it very vaguely and wrote things like "do whatever it takes, find an exploit however you can" because it's in a sandbox so you want the model to try its hardest.

That is an interesting question. If the prompt included "Do not break out of the sandbox we've provided you. Do not use information retrieved from outside the sandbox. All answers that were provided in this manner are invalid and will score 0 points.", would this still have happened?


So...?

We should not construct a machine that is one bad prompt away from causing catastrophe.


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

Search: