The project would benefit from explaining the test setup.
From what I can tell (from code), the machine is probably well under capacity, so I am unsure about how to interpret the results.
If you consider that a pixel is a container for RGB value, then the length and area collapse into simple storage requirement.
When an image is some px wide and some px high, it provides a shape (ratio between them), because a camera of 12mega px isn't 1px x 12M px in shape, but it does occupy 12M storage units nevertheless.
I agree though that like many IT related terms, their usage is twisted and simplified to amplify the buzz. Because it is more marketing friendly to put a bigger number than exact dimensions and proper terminology.
This is very interesting. I have not tested it (yet) but I believe this is what will come next when talking about "AI will be everywhere".
The self-contained disconnected mode is very convenient, though it could be useful to have an "online access" mode to allow it to access the Internet and perform lookup.
My own guess would be that what matters most for the frontend is UX (and accessibility).
So if you are able to properly evaluate the outcome, it is probably fine.
It will also depend on the type of frontend and its target complexity. From what I understand from your question, you would handle the backend and infrastructure so these do need higher rigour.
If you apply the principle of separation of concerns, the frontend shall be decoupled from the rest and may evolve on its own end. That means that you have some degree of freedom to test and fail the frontend without harming the rest of your project.
Your end users may not like it much, but from a dev perspective, there is no "flaw".
Also consider that LLM training dataset and available data is huge about frontends, so it is way better at doing these.
Did you consider that star growth stalling could mean product maturity?
We always want more stars, but if your innovators user base (those that try at the early days and give stars) is now turning to average users that will use the product without rating.
While I agree on the fact, this is the typical case where you need a business analyst in between to 'translate' that requirement and confront the leadership if it is unclear.
If you mean that leadership IS the business analyst, then you are in trouble :D
The goal of the AI act is not to determine if an "oh yeah!" comment was AI generated. The target is long papers that falsely claim human review and can have real significant consequences.
E.g. research paper, law makers, lawyers, state policies, notaries,...
These are much longer content and thus statistically they will disclose a better guess at AI generated content.
Asking another AI to paraphrase will not erase the mark (which they are unaware about) but rather cumulatively add their own mark and make it easier to detect.
The problem is not to use AI, but to endorse the responsibility of the content you (as a human) deliver and somehow make sure that fake-news, biased content or unverified output is detected as early as possible.
You'd think that the humans being paid to review these things are actually reviewing them. Journals, laywers etc. are expensive. With the addition of AI, it should be easier than ever to review things on their merits.
Maybe some part of it is that the deluge of slop is uncovering how poorly/sloppily these social institutions were working in the first place.
Peer reviews is largely done for free. In some cases, an author needs to commit to reviewing someone else's work for their work to be reviewed. However, this has led to an increase in LLM usage for review generation, even if conference/journal guidelines forbid it. I've seen nonsensical reviews from people who obviously haven't read the paper beyond an LLM-generated summary. Area chairs are supposed to catch this, but I imagine they're using LLMs too.
I think you are very correct here! Now imagine how much worse it can get with AI in the way.
It is as always (think cybersecurity) the cat and mouse game, what is a weapon is also a defense. You, the simple fact that you are on this very site, means that you are probably more educated to AI than most, so it is not necessarily you that will really benefit from any constraining framework.
There are many people believing in many weird theories and those are more prone to be convinced by a nice narrative. AI did not introduce that, it just made it easier and available to anyone with any intent.
Anecdote: I've had at two friends publish "papers" just by slapping their names on something they had literally nothing to do with, out of pure nepotism.
This wouldn't be helped by an AI watermark. What would help is if the reviewer used AI to look up the authors and see the authorship claims are dubious. The papers are still up.
I do think the academic publishing field is corrupt, which is why I'm not convinced it should be on AI providers to help bail it out of doing its one job (verification and trust).
Verifying authorship is an ambiguous task. How much involvement is necessary to qualify for authorship? If someone reads the paper and gives some small feedback, is that enough? What if they were present in one meeting and raised a question that turned out not to be interesting? What if they have no clue about the work but helped with data validation?
I'd argue all of these could justify authorship, even if they're just in the middle of the author list. At least in NLP, which can be seen in the generally high number of authors in papers.
The standards also vary by field. It's not unusual for supervisors to be the last author on any paper that a group publishes, even if they had basically no input into it directly. I've been listed on papers just because I designed and built the equipment that happened to be used for the experiments, even though I didn't do anything but a quick review of the actual paper. For most papers, unless there's an indication otherwise, it's generally only safe to assume the first author listed that actually did the bulk of the work and write-up, and the others are there mainly for having some potentially quite indirect contribution.
I agree completely. That's why I pushed back on "nothing to do" from the OP. Maybe it was a small contribution, but calling it "corruption" is ridiculous.
From what I understand of the not-very-detailed text watermark, it comes up to masking the generated output with specific bias in the weights of the token generation (or something like this). Anthropic was explaining that it would survive edits (not full rewrite). So asking another AI to paraphrase will surely add its own, but since the bias is not disclosed, it will depend on how the 2nd AI is considering the existing tokens to steer its own.
Fictive example: imagine the bias is to inject notion of colorfullness in the text regardless or its content:
"I eat an apple" becomes "I eat a red apple"
The 2nd AI is about injecting size component, so the text becomes " I eat a big red apple"
There you get both watermarks.
It is not that obvious obviously
I think that part of the problem is that you try to apply this to low level variables.
If you consider a framework on top of any existing language that provides serialization mechanics that include metadata about the item, then you can approach your 2 goals.
I think that one important aspect (whether that is feasible or not is another question) is to have valid coercion. JS has some, but as your example shows, it is not complete.
Is the binary equal value a requirement or is it the semantic value which can take another form but can be interpreted in the same way?
Very interesting topic overall and I've been touching these questions a bit (until I had nightmares).
reply