It is an interesting article and some of the results are unexpected, but the layout is way too long and could be greatly condensed. At first glance and speaking from how I like to lay things out in a paper:
- You do not really need to show the prompt interface. You can show it once and thereafter use a bulleted list format, or simply show the input image if it responded correctly. - Your figures should be about half of their size, they don't need to fit the width of the body. - For comparative results against other models you can use a table with colored cells, with the test name on the row and model names on the columns. - For the dog, show a side-by-side figure with the raw image on the left and the box on the right, and include the coordinates it gave you in the body. - In your conclusion show the full matrix table of comparative results and summarize the relative strengths of the model against the others.
In terms of the writing and methods your conclusion says little and your tests do not go into significant depth. For example with the tire image you could show that it succeeds when cropped but as the photo gets wider it begins to fail to correctly identify the text in the image's center. For example see the methodology and presentation this article used: https://dynomight.net/ducks/
Also, the OCR test is too simple, even a 20-year-old OCR algorithm would probably recognize that. Experimenting with progressive degradation of the image could show its strengths, and analysis could show its accuracy at each level of degradation.
I'd say, trying to read this, the biggest problems are:
- tons of visual clutter, all those gradients and lines like the header or hero image
- a floating ToC which insists on jamming in 'recommended links' (?!) the entire time
- no outlines. Every single image or screenshot blends into the actual article.
- a visual summary which is hard to read because it has tiny text and looks like a correlation heatmap instead of a table
- highly inconsistent use of linking. Like, why does 'We have evaluated Gemini across four separate vision tasks:' link only 2 of the 4, and then not to the section in this article?
- highly repetitive screenshots, which add nothing, and in conjunction with the lack of outlines for the images and the many outlines inside the images, means that the benchmark sections are a frustrating visual jigsaw puzzle where you have to decode the screenshot again and again to look at the tiny text inside it. It would be better to provide one (1) screenshot of each model's UI, which is all I need to see to get an idea of what it looks like and the implied workflow and what sort of metadata/options it has, and then for each task simply show the image/prompt and each model's responses as a normal blockquote or text.
- tons of visual clutter, all those gradients and lines like the header or hero image
- a floating ToC which insists on jamming in 'recommended links' (?!) the entire time
- no outlines. Every single image or screenshot blends into the actual article.
- a visual summary which is hard to read because it has tiny text and looks like a correlation heatmap instead of a table
- highly inconsistent use of linking. Like, why does 'We have evaluated Gemini across four separate vision tasks:' link only 2 of the 4, and then not to the section in this article?
- highly repetitive screenshots, which add nothing, and in conjunction with the lack of outlines for the images and the many outlines inside the images, means that the benchmark sections are a frustrating visual jigsaw puzzle where you have to decode the screenshot again and again to look at the tiny text inside it. It would be better to provide one (1) screenshot of each model's UI, which is all I need to see to get an idea of what it looks like and the implied workflow and what sort of metadata/options it has, and then for each task simply show the image/prompt and each model's responses as a normal blockquote or text.
All: I sincerely appreciate the time spent sharing feedback. Your notes and comments are helpful and give me tools to be a better writer .
Regarding the screenshots, I am not a fan of this approach. We adopted it because of the early trend to share ChatGPT screenshots, and to ensure people could see the origin of our prompting (the web interface).
I will start a discussion about screenshots with the team. This can be better.
I will also discuss the layout, too. Machine learning and AI is difficult enough. To the extent to which we can focus attention on the most important part of the page — the content — we should.
While I don't necessarily agree with all of these points,
> link only 2 of the 4, and then not to the section in this article?
This one is particularly prevalent on websites and it's quite annoying. When the site has any topic explainer articles, the terms that refer to those topics are always linked to those other articles, presumably to increase ad impressions and keep users on their site- but when there are legitimate article-specific links (which are almost always what I want), I have no way to locate those links (for instance when finding an original source).
Back in the day websites would use a different link style for this sort of "internal plug" style links, which was helpful. I guess it died out because users didn't want to click them. So the solution is, make it hard to tell which ones are internal plugs!
Does anyone know why apple acquired them just to shut it down?
It doesn’t seem like the weather data was unique - what was unique was its user focussed design.
To acquire it just to shut it down would have only made sense to me had they actually folded the design of dark sky into their weather app and put their founders in charge of the app. Doesn’t seem like that happened
The weather data actually was pretty unique in that a lot of the rain / snow predictions were based on visually analysing the satellite imagery. I'd understand buying it just for that reason, but to then seemingly leave that out of the stock weather app seems bizarre. (I'm guessing they haven't ported that to the stock app as the stock app is dreadful at predicting imminent rain)
Did Apple include the local weather prediction in their app? That was the other impressive part of Dark Sky. The combination of prediction and design was why Dark Sky was so good.
I use Android and never used the nice app, but used the predictions in other apps. Knowing when it was going to rain was really handy. Apps now their own predictions but not as good.
The Dark Sky API was very popular, and at the time many weather apps on iOS did sketchy things with people's data. I sensed that Apple maybe wanted to provide a free weather API for weather apps to make it easier for anyone to make a weather app.
Privacy.com is a great service. I use them all the time to generate 1 time use card numbers for sites & then cancel the card so they cannot mysteriously charge me. I've been with them for years & their CEO is a wonderful & smart person.
When you're allowing strangers to perform financial transactions - you're taking on risk that the money that is sent needs to actually be funded. They need to conform to KYC laws like all fintech providers - so yes, they will require knowing a little bit about you to operate in the United States like all financial institutions.
But they apparently require government photo ID that gets sent off into the cloud? You can most certainly get a regular credit card without that, so it doesn't seem analogous. They also intentionally hide this requirement so that you don't find out about it until after you've signed up, which is a pretty devious dark pattern.
That part is fine. The part that isn't is where they require you to give your personal information to a third party who has very weak controls on how they share it. Your endorsement makes me interested in trying the service, but I have the same "nope" reaction to the non-privacy as the author of this post.
What makes you think OnFido is sketchy? It’s a pretty popular platform for ID verification.
3rd party verification has become a standard in the fintech/insuretech industries since its very hard and risky to do KYC on your own. Also personally I don’t trust having all the random companies I transact with maintain my KYC info. At least in theory, the experts at ID verification have strong enough incentives, motivation and expertise to keep my data safe, reducing the attack surface area.
FWIW, they are at least willing to put this in their privacy policy:
> Whenever legally possible, we seek to protect the information we share by imposing contractual privacy and security safeguards on the recipient of the information. This is particularly important in cases where the recipient is located in a country that has different or lesser privacy laws than those of the country where the information was originally collected. In some cases, however, it’s not possible for us to do so — for example, when we have a legal obligation to disclose information to a government authority and that government authority isn’t willing to enter into such contractual safeguards.
Check the article. It has quotes from their ToS that can be roughly summarized as "we'll sell all your data to whoever pays and you have no control over this".
No, that's not what they are saying at all. The quote on the blog is misleading and leaves out important pieces. Here's the full thing:
"As part of a business transfer. Onfido may disclose your personal information to an actual or potential buyer, investor or partner (and its agents and advisers) in relation to any actual or proposed divestiture, merger, acquisition, joint venture, bankruptcy, dissolution, reorganization, or any other similar transaction or proceeding"
Nothing sketchy about Onfido. You're hating on them all over this thread based on nothing but one sentence from their privacy policy, quoted on a random blog. A sentence that's not even saying what you think it's saying.
Here's the full paragraph:
"As part of a business transfer. Onfido may disclose your personal information to an actual or potential buyer, investor or partner (and its agents and advisers) in relation to any actual or proposed divestiture, merger, acquisition, joint venture, bankruptcy, dissolution, reorganization, or any other similar transaction or proceeding"
I've been using it since the Raspberry Pi 3 days (before rpi 4 came about) because ads affected them so severely. And now that we have the pi4 that can handle the load, I still have not changed from reading CNN from the lite page. And that has carried over to all my other devices. I just Google the title if I want to see pictures or video in the article.
I used to do the same with Twitter, but these days there is nearly no difference between the mobile and the regular versions.
It would do well with a tiny amount of additional CSS to make it work well on large screens, but I guess they don't want to make it too nice to use for that segment.
IIRC, this site was put together, in particular, for low-bandwidth users (like mobile phone users back in the feature phone day) where every byte cost. CSS would be an unnecessary and potentially expensive addition for those users.
It's not optimized to that level. There's 1.1k of inline tracking javascript. And a reference to an external stylesheet for the styling, along with a bunch of wastefully long hex identifiers.
Took a look at the dev console, there are 2 tracking scripts, segment and CNN specific one. Those take up the majority of bandwidth. There's also a separate CSS style sheet. The main DOM and the stylesheet are <10% of the total bytes transferred. The favicon is the same byte size as the entire DOM.
Don't think there is necessarily a consideration of byte cost, given the size of the tracking scripts. A few lines to make this manageable on wide monitors aren't prohibitive for those users.
The website already serves a 496 byte CSS file (359 minified, even less gzipped), addition of a max-width to make it readable on large screens wouldn't make a difference.
I read HN on Kindle[1], So low bloat websites like these are always a pleasure to read on Kindle's forever experimental browser.
The websites needn't be text only, Reducing useless elements like Techcrunch's content wrap can go a long way to improve readability and accessibility.
There are a lot of efforts to increase diversity in media.
One effort that has been happening is providing incentives to have diverse top of the line staff on productions so that they share their stories and angle on things.
We at wrapbook (https://www.wrapbook.com) at making it easy for productions to measure their diverse staff so that we can incentivize diversity down the line.
I happen to have co-founded a company (Wrapbook) that makes it easy to work with gig-based workers as employees. The entertainment space long ago settled this fight and classifies everyone as employees, not as contractors, even when they work a single day.
This fight has nothing to do with independence and only has to do with employers trying to save some 15% of an employee's wage in taxes & keep themselves off the hook if people get injured on the job.
The savings are real - somebody making $1000 a day - classified as an employee, costs an additional estimated $150.
You can pay short term workers as employees pretty easily (we facilitate it).
